docs(fleet): define declarative config requirements

Refs #758

.gitignore

@@ -15,3 +15,10 @@ infra/step-ca/dev-password
 
 # Scratch dirs created by the framework git-wrapper shell test harnesses
 .mosaic-test-work/
+
+# Transient config files vite/vitest/esbuild write next to a *.config.ts while
+# loading it, then unlink. They are untracked but were not ignored, so turbo's
+# package traversal hashed them and intermittently failed CI with "Package
+# traversal error: ... .timestamp-*.mjs: No such file or directory" when the
+# file vanished mid-scan. Ignoring them removes the race.
+*.timestamp-*.mjs

.npmrc

@@ -1 +1,5 @@
 @mosaicstack:registry=https://git.mosaicstack.dev/api/packages/mosaicstack/npm/
+# Pin the pnpm store to the same path the ci-base image warms (Dockerfile.ci),
+# so the pipeline `pnpm install --prefer-offline` consumes the baked store
+# instead of repopulating a fresh one.
+store-dir=/root/.local/share/pnpm/store

.prettierignore

@@ -5,3 +5,5 @@ pnpm-lock.yaml
 **/drizzle
 **/.next
 .claude/
+docs/tess/TASKS.md
+docs/scratchpads/

.woodpecker/ci-image.yml

@@ -0,0 +1,40 @@
+# Build & push the pre-baked CI base image (Dockerfile.ci) to the Gitea
+# registry CI already publishes to. Reuses the exact kaniko + auth pattern
+# from publish.yml (REGISTRY_USER/REGISTRY_PASS from_secret, /kaniko/.docker
+# config.json). Other pipelines (ci.yml, publish.yml) pull `ci-base:latest`
+# for their install step.
+#
+# Rebuild ONLY when the dependency set or the image recipe changes — a normal
+# code push must not trigger a 25-min image build. `path` applies to push/PR
+# events; `event: tag` (releases) rebuilds unconditionally so a tagged release
+# always ships a fresh base.
+when:
+  - event: tag
+  - event: [push, manual]
+    branch: main
+    path:
+      include:
+        - 'pnpm-lock.yaml'
+        - 'Dockerfile.ci'
+
+steps:
+  build-ci-base:
+    image: gcr.io/kaniko-project/executor:debug
+    environment:
+      REGISTRY_USER:
+        from_secret: gitea_username
+      REGISTRY_PASS:
+        from_secret: gitea_password
+      CI_COMMIT_BRANCH: ${CI_COMMIT_BRANCH}
+      CI_COMMIT_TAG: ${CI_COMMIT_TAG}
+      CI_COMMIT_SHA: ${CI_COMMIT_SHA}
+    commands:
+      - mkdir -p /kaniko/.docker
+      - echo "{\"auths\":{\"git.mosaicstack.dev\":{\"username\":\"$REGISTRY_USER\",\"password\":\"$REGISTRY_PASS\"}}}" > /kaniko/.docker/config.json
+      - |
+        # Lockfile-hash tag: an immutable identity for the exact dep set baked
+        # into this image. `:latest` is the mutable pointer pipelines consume.
+        LOCK_HASH=$(sha256sum pnpm-lock.yaml | cut -c1-12)
+        DESTINATIONS="--destination git.mosaicstack.dev/mosaicstack/stack/ci-base:latest"
+        DESTINATIONS="$DESTINATIONS --destination git.mosaicstack.dev/mosaicstack/stack/ci-base:lock-$LOCK_HASH"
+        /kaniko/executor --context . --dockerfile Dockerfile.ci $DESTINATIONS

.woodpecker/ci.yml

@@ -1,9 +1,20 @@
+# &node_image is the pre-baked CI base built by .woodpecker/ci-image.yml:
+# node:24-alpine + python3/make/g++/postgresql-client + pnpm + a warm pnpm
+# store. The install step resolves from the baked store (--prefer-offline)
+# instead of paying a ~731s cold fetch + native compile every run.
 variables:
-  - &node_image 'node:22-alpine'
+  - &node_image 'git.mosaicstack.dev/mosaicstack/stack/ci-base:latest'
   - &enable_pnpm 'corepack enable'
 
 when:
-  - event: [push, pull_request, manual]
+  # PR + manual CI run on any branch — the pull_request pipeline is the merge gate.
+  # push CI is restricted to protected branches (main) so a feature-branch push no
+  # longer fires a redundant SECOND pipeline alongside its PR pipeline. This ~halves
+  # CI load on the storage-constrained runner with zero loss of gating (branch
+  # protection requires no push/ci status context; main still gets full push CI).
+  - event: [pull_request, manual]
+  - event: push
+    branch: main
 
 # Turbo remote cache (turbo.mosaicstack.dev) is configured via Woodpecker
 # repository-level environment variables (TURBO_API, TURBO_TEAM, TURBO_TOKEN).
@@ -15,8 +26,9 @@ steps:
     image: *node_image
     commands:
       - corepack enable
-      - apk add --no-cache python3 make g++
-      - pnpm install --frozen-lockfile
+      # python3/make/g++ are baked into ci-base; --prefer-offline resolves from
+      # the baked pnpm store.
+      - pnpm install --frozen-lockfile --prefer-offline
 
   # Blocking gate: public framework package must contain no operator-specific
   # personal data or private $HOME defaults. Runs early (no node_modules needed).
@@ -64,8 +76,7 @@ steps:
       DATABASE_URL: postgresql://mosaic:mosaic@ci-postgres:5432/mosaic
     commands:
       - *enable_pnpm
-      # Install postgresql-client for pg_isready
-      - apk add --no-cache postgresql-client
+      # postgresql-client (pg_isready) is baked into ci-base.
       # Wait up to 60s for CI postgres to be ready; fail fast if it never comes up.
       - |
         ready=0

.woodpecker/publish.yml

@@ -2,8 +2,27 @@
 # Runs only on main branch push/tag
 
 variables:
-  - &node_image 'node:22-alpine'
+  # Pre-baked CI base (see .woodpecker/ci-image.yml): node:24-alpine +
+  # toolchain + warm pnpm store. Kills the second cold install publish pays.
+  - &node_image 'git.mosaicstack.dev/mosaicstack/stack/ci-base:latest'
   - &enable_pnpm 'corepack enable'
+  # Heavy kaniko image builds (~25 min) — gate them so a merge that only touches
+  # the npm-only CLI (@mosaicstack/mosaic) or docs does NOT rebuild the platform
+  # images (gateway/appservice/web do not depend on @mosaicstack/mosaic). Releases
+  # (tags) always build everything. Exclude-list keeps the default SAFE: any
+  # non-excluded change still builds, so no transitive dep can silently go stale.
+  # (Woodpecker: `when` entries are OR'd; `path` applies to push/PR only — hence
+  # the separate `event: tag` entry.)
+  - &image_build_when
+    - event: tag
+    - event: [push, manual]
+      branch: main
+      path:
+        exclude:
+          - 'packages/mosaic/**'
+          - 'docs/**'
+          - '**/*.md'
+          - '.woodpecker/**'
 
 when:
   - branch: [main]
@@ -14,7 +33,8 @@ steps:
     image: *node_image
     commands:
       - corepack enable
-      - pnpm install --frozen-lockfile
+      # Resolve from the baked pnpm store instead of a cold network fetch.
+      - pnpm install --frozen-lockfile --prefer-offline
 
   build:
     image: *node_image
@@ -26,6 +46,15 @@ steps:
 
   publish-npm:
     image: *node_image
+    # Publish only when a publishable package changed (or on a release tag); a
+    # pure-docs merge runs no publish. Cheap step, but gated for cleanliness.
+    when:
+      - event: tag
+      - event: [push, manual]
+        branch: main
+        path:
+          include:
+            - 'packages/**'
     environment:
       NPM_TOKEN:
         from_secret: gitea_token
@@ -91,6 +120,7 @@ steps:
 
   build-gateway:
     image: gcr.io/kaniko-project/executor:debug
+    when: *image_build_when
     environment:
       REGISTRY_USER:
         from_secret: gitea_username
@@ -116,6 +146,7 @@ steps:
 
   build-appservice:
     image: gcr.io/kaniko-project/executor:debug
+    when: *image_build_when
     environment:
       REGISTRY_USER:
         from_secret: gitea_username
@@ -141,6 +172,7 @@ steps:
 
   build-web:
     image: gcr.io/kaniko-project/executor:debug
+    when: *image_build_when
     environment:
       REGISTRY_USER:
         from_secret: gitea_username

Dockerfile.ci

@@ -0,0 +1,45 @@
+# Pre-baked CI base image for Woodpecker pipelines.
+#
+# Purpose: eliminate the cold `pnpm install` that dominates every pipeline
+# (~731s median). This image ships the native toolchain (no per-run `apk add`)
+# AND a warm, content-addressable pnpm store with the dependency-tree tarballs
+# already fetched at build time. `pnpm fetch` only populates the store from the
+# lockfile — it does NOT run the native node-gyp builds (better-sqlite3,
+# node-pty, sqlite3, canvas, sharp); those still compile at `pnpm install`,
+# which is exactly why the musl toolchain stays baked into this image. A
+# pipeline `pnpm install --frozen-lockfile --prefer-offline` then resolves
+# tarballs from local hard-links (no network) and compiles natives against the
+# already-present toolchain, in tens of seconds instead of ~731s.
+#
+# Rebuilt only when `pnpm-lock.yaml` or this Dockerfile change
+# (see .woodpecker/ci-image.yml).
+#
+# Node version is pinned to 24 (Active LTS). This is the follow-up bump from
+# node:22 — sequenced AFTER the CI cache work landed so the runtime change
+# carries zero cache variables. node:26 stays held until it reaches LTS
+# (Oct 2026); the Current line risks native-module (node-gyp) breakage on a
+# runner that compiles better-sqlite3 / canvas / sharp / node-pty from source.
+FROM node:24-alpine
+
+# Native toolchain required to compile node-gyp deps on musl, plus the
+# postgresql-client used by the test step's pg_isready readiness probe. `bash`
+# is baked here too — the sanitization step in ci.yml otherwise does a per-run
+# `apk add bash`.
+RUN apk add --no-cache python3 make g++ postgresql-client bash
+
+# Pin pnpm to the repo's packageManager version via corepack.
+RUN corepack enable && corepack prepare pnpm@10.6.2 --activate
+
+WORKDIR /app
+
+# Pin the store location so the pipeline can point `store-dir` at the same path.
+ENV PNPM_HOME=/root/.local/share/pnpm
+RUN pnpm config set store-dir /root/.local/share/pnpm/store
+
+# Warm the store. `pnpm fetch` populates the content-addressable store with the
+# dependency tarballs directly from the lockfile (no package.json / workspace
+# needed), so a baked store stays valid until the lockfile changes. Note:
+# `fetch` does NOT compile native modules — that happens later at `pnpm install`
+# in the pipeline, against the toolchain baked above.
+COPY pnpm-lock.yaml ./
+RUN pnpm fetch --frozen-lockfile

apps/gateway/package.json

@@ -31,6 +31,7 @@
     "@mariozechner/pi-ai": "^0.65.0",
     "@mariozechner/pi-coding-agent": "^0.65.0",
     "@modelcontextprotocol/sdk": "^1.27.1",
+    "@mosaicstack/agent": "workspace:^",
     "@mosaicstack/auth": "workspace:^",
     "@mosaicstack/brain": "workspace:^",
     "@mosaicstack/config": "workspace:^",

apps/gateway/src/__tests__/integration/tess-cross-surface.integration.test.ts

@@ -0,0 +1,192 @@
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import { InMemoryDurableSessionStore } from '@mosaicstack/agent';
+import {
+  createDiscordIngressEnvelope,
+  DiscordPlugin,
+  type DiscordIngressPayload,
+} from '@mosaicstack/discord-plugin';
+import { InteractionController } from '../../agent/interaction.controller.js';
+import { RuntimeProviderService } from '../../agent/runtime-provider-registry.service.js';
+import { DurableSessionService } from '../../agent/durable-session.service.js';
+import { ChatGateway } from '../../chat/chat.gateway.js';
+import { CommandAuthorizationService } from '../../commands/command-authorization.service.js';
+
+const SERVICE_TOKEN = 'test-discord-service-token';
+const envKeys = [
+  'DISCORD_SERVICE_TOKEN',
+  'DISCORD_SERVICE_TENANT_ID',
+  'DISCORD_INTERACTION_BINDINGS',
+  'DISCORD_ALLOWED_GUILD_IDS',
+  'DISCORD_ALLOWED_CHANNEL_IDS',
+  'DISCORD_ALLOWED_USER_IDS',
+  'MOSAIC_AGENT_NAME',
+] as const;
+const priorEnv = new Map<string, string | undefined>();
+
+function payload(content: string, messageId: string, correlationId: string): DiscordIngressPayload {
+  return {
+    content,
+    messageId,
+    correlationId,
+    guildId: 'guild-1',
+    channelId: 'channel-1',
+    userId: 'discord-admin-1',
+    conversationId: 'Nova:discord:channel-1',
+  };
+}
+
+function authorization(): CommandAuthorizationService {
+  const entries = new Map<string, string>();
+  return new CommandAuthorizationService(
+    {
+      select: () => ({
+        from: () => ({ where: () => ({ limit: async () => [{ role: 'admin' }] }) }),
+      }),
+    } as never,
+    {
+      get: async (key: string) => entries.get(key) ?? null,
+      set: async (key: string, value: string) => entries.set(key, value),
+      del: async (key: string) => Number(entries.delete(key)),
+    },
+  );
+}
+
+describe('interaction Discord/CLI durable-session integration', () => {
+  afterEach(() => {
+    for (const key of envKeys) {
+      const value = priorEnv.get(key);
+      if (value === undefined) delete process.env[key];
+      else process.env[key] = value;
+    }
+    priorEnv.clear();
+  });
+
+  it('enrolls through the CLI surface then resolves the same durable session from Discord', async () => {
+    for (const key of envKeys) priorEnv.set(key, process.env[key]);
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    process.env['DISCORD_SERVICE_TOKEN'] = SERVICE_TOKEN;
+    process.env['DISCORD_SERVICE_TENANT_ID'] = 'tenant-1';
+    process.env['DISCORD_ALLOWED_GUILD_IDS'] = 'guild-1';
+    process.env['DISCORD_ALLOWED_CHANNEL_IDS'] = 'channel-1';
+    process.env['DISCORD_ALLOWED_USER_IDS'] = 'discord-admin-1';
+    process.env['DISCORD_INTERACTION_BINDINGS'] = JSON.stringify([
+      {
+        instanceId: 'Nova',
+        guildId: 'guild-1',
+        channelId: 'channel-1',
+        pairedUsers: {
+          'discord-admin-1': { role: 'admin', mosaicUserId: 'mosaic-admin-1' },
+        },
+      },
+    ]);
+
+    const durable = new DurableSessionService(
+      new InMemoryDurableSessionStore() as never,
+      {} as never,
+    );
+    const enrollmentRuntime = {
+      listSessions: vi.fn().mockResolvedValue([{ id: 'runtime-1' }]),
+    };
+    const controller = new InteractionController(enrollmentRuntime as never, durable);
+    await controller.enroll(
+      'Nova',
+      'Nova:discord:channel-1',
+      { providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+      { id: 'mosaic-admin-1', tenantId: 'tenant-1' },
+      'cli-enrollment-correlation',
+    );
+
+    const authz = authorization();
+    const terminated = vi.fn().mockResolvedValue(undefined);
+    const runtime = new RuntimeProviderService(
+      {
+        require: () => ({
+          capabilities: async () => ({ supported: ['session.terminate'] }),
+          terminate: terminated,
+        }),
+      } as never,
+      { record: async () => undefined } as never,
+      {
+        consume: (approvalId, action) =>
+          authz.consumeRuntimeTerminationApproval(approvalId, action),
+      },
+    );
+    const gateway = new ChatGateway(
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      authz,
+      runtime,
+      durable,
+    );
+    const client = { data: { discordService: true }, emit: vi.fn() };
+    const plugin = new DiscordPlugin({
+      token: 'unused',
+      gatewayUrl: 'http://unused',
+      serviceToken: SERVICE_TOKEN,
+      allowedGuildIds: ['guild-1'],
+      allowedChannelIds: ['channel-1'],
+      allowedUserIds: ['discord-admin-1'],
+      interactionBindings: [
+        {
+          instanceId: 'Nova',
+          guildId: 'guild-1',
+          channelId: 'channel-1',
+          pairedUsers: {
+            'discord-admin-1': { role: 'admin', mosaicUserId: 'mosaic-admin-1' },
+          },
+        },
+      ],
+    });
+    const pluginInternals = plugin as unknown as {
+      client: { user: { id: string } };
+      socket: { connected: boolean; emit: ReturnType<typeof vi.fn> };
+      handleDiscordMessage(message: unknown): void;
+    };
+    const pluginSocket = { connected: true, emit: vi.fn() };
+    pluginInternals.client = { user: { id: 'bot-1' } };
+    pluginInternals.socket = pluginSocket;
+    pluginInternals.handleDiscordMessage({
+      id: 'approve-1',
+      guildId: 'guild-1',
+      channelId: 'channel-1',
+      author: { id: 'discord-admin-1', bot: false },
+      mentions: { has: () => true },
+      content: '<@bot-1> /approve',
+      channel: { parentId: null },
+      attachments: new Map(),
+    });
+
+    expect(pluginSocket.emit).toHaveBeenCalledWith('discord:approve', expect.any(Object));
+    const approvalEnvelope = pluginSocket.emit.mock.calls[0]?.[1];
+    await gateway.handleDiscordApproval(client as never, approvalEnvelope);
+    const approval = client.emit.mock.calls.find(
+      ([event]) => event === 'discord:approval',
+    )?.[1] as {
+      approvalId: string;
+      success: boolean;
+    };
+    expect(approval.success).toBe(true);
+
+    await gateway.handleDiscordStop(
+      client as never,
+      createDiscordIngressEnvelope(
+        payload(`/stop ${approval.approvalId}`, 'stop-1', 'discord-stop-correlation'),
+        SERVICE_TOKEN,
+      ),
+    );
+
+    expect(terminated).toHaveBeenCalledWith(
+      'runtime-1',
+      approval.approvalId,
+      expect.objectContaining({ actorId: 'mosaic-admin-1' }),
+    );
+    expect(client.emit).toHaveBeenCalledWith('discord:stop', {
+      correlationId: 'discord-stop-correlation',
+      success: true,
+    });
+  });
+});

apps/gateway/src/admin/admin-health.controller.ts

@@ -20,7 +20,7 @@ export class AdminHealthController {
   async check(): Promise<HealthStatusDto> {
     const [database, cache] = await Promise.all([this.checkDatabase(), this.checkCache()]);
 
-    const sessions = this.agentService.listSessions();
+    const sessions = this.agentService.listAllSessionsForSystem();
     const providers = this.providerService.listProviders();
 
     const allOk = database.status === 'ok' && cache.status === 'ok';

apps/gateway/src/agent/__tests__/agent-service-ownership.test.ts

@@ -0,0 +1,179 @@
+import { ForbiddenException } from '@nestjs/common';
+import { describe, expect, it, vi } from 'vitest';
+import { AgentService, type AgentSession } from '../agent.service.js';
+import type { ActorTenantScope } from '../../auth/session-scope.js';
+
+const CONVERSATION_ID = '22222222-2222-4222-8222-222222222222';
+const OWNER_SCOPE: ActorTenantScope = { userId: 'owner-user', tenantId: 'owner-tenant' };
+const FOREIGN_SCOPE: ActorTenantScope = { userId: 'foreign-user', tenantId: 'foreign-tenant' };
+
+type AgentServiceInternals = {
+  sessions: Map<string, AgentSession>;
+  creating: Map<string, Promise<AgentSession>>;
+};
+
+function makeService(operatorMemory: unknown = null): AgentService {
+  return new AgentService(
+    {
+      getDefaultModel: vi.fn(() => null),
+      getRegistry: vi.fn(() => ({})),
+      findModel: vi.fn(),
+      listAvailableModels: vi.fn(() => []),
+    } as never,
+    {} as never,
+    {} as never,
+    { available: false } as never,
+    {} as never,
+    { getToolDefinitions: vi.fn(() => []) } as never,
+    { loadForSession: vi.fn(async () => ({ metaTools: [], promptAdditions: [] })) } as never,
+    null,
+    null,
+    { collect: vi.fn().mockResolvedValue(undefined) } as never,
+    operatorMemory as never,
+  );
+}
+
+function internals(service: AgentService): AgentServiceInternals {
+  return service as unknown as AgentServiceInternals;
+}
+
+function makeSession(scope: ActorTenantScope = OWNER_SCOPE): AgentSession {
+  return {
+    id: CONVERSATION_ID,
+    provider: 'test-provider',
+    modelId: 'test-model',
+    piSession: {
+      thinkingLevel: 'off',
+      getAvailableThinkingLevels: vi.fn().mockReturnValue(['off', 'low', 'high']),
+      setThinkingLevel: vi.fn(),
+      abort: vi.fn().mockResolvedValue(undefined),
+      prompt: vi.fn().mockResolvedValue(undefined),
+      dispose: vi.fn(),
+      getSessionStats: vi.fn(),
+      getContextUsage: vi.fn(),
+    } as unknown as AgentSession['piSession'],
+    listeners: new Set(),
+    unsubscribe: vi.fn(),
+    createdAt: Date.now(),
+    promptCount: 0,
+    channels: new Set(),
+    skillPromptAdditions: [],
+    sandboxDir: '/tmp/tess-session-ownership-test',
+    allowedTools: null,
+    userId: scope.userId,
+    tenantId: scope.tenantId,
+    metrics: {
+      tokens: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+      modelSwitches: 0,
+      messageCount: 0,
+      lastActivityAt: new Date('2026-07-12T00:00:00Z').toISOString(),
+    },
+  };
+}
+
+describe('AgentService owner/tenant scope enforcement', () => {
+  it('allows owner-scoped operations and rejects foreign scopes for seeded sessions', async () => {
+    const service = makeService();
+    const session = makeSession();
+    internals(service).sessions.set(CONVERSATION_ID, session);
+
+    expect(service.getSession(CONVERSATION_ID, OWNER_SCOPE)).toBe(session);
+    expect(service.getSession(CONVERSATION_ID, FOREIGN_SCOPE)).toBeUndefined();
+    expect(service.getSessionInfo(CONVERSATION_ID, FOREIGN_SCOPE)).toBeUndefined();
+    expect(service.listSessions(OWNER_SCOPE)).toHaveLength(1);
+    expect(service.listSessions(FOREIGN_SCOPE)).toEqual([]);
+
+    service.addChannel(CONVERSATION_ID, 'websocket:owner', OWNER_SCOPE);
+    expect(session.channels.has('websocket:owner')).toBe(true);
+    expect(() => service.addChannel(CONVERSATION_ID, 'websocket:foreign', FOREIGN_SCOPE)).toThrow(
+      ForbiddenException,
+    );
+    expect(() => service.removeChannel(CONVERSATION_ID, 'websocket:owner', FOREIGN_SCOPE)).toThrow(
+      ForbiddenException,
+    );
+
+    expect(() =>
+      service.updateSessionModel(CONVERSATION_ID, 'foreign-model', FOREIGN_SCOPE),
+    ).toThrow(ForbiddenException);
+    service.updateSessionModel(CONVERSATION_ID, 'owner-model', OWNER_SCOPE);
+    expect(session.modelId).toBe('owner-model');
+
+    expect(() =>
+      service.applyAgentConfig(CONVERSATION_ID, 'agent-foreign', 'Foreign Agent', FOREIGN_SCOPE),
+    ).toThrow(ForbiddenException);
+    service.applyAgentConfig(CONVERSATION_ID, 'agent-owner', 'Owner Agent', OWNER_SCOPE);
+    expect(session.agentConfigId).toBe('agent-owner');
+
+    expect(() => service.onEvent(CONVERSATION_ID, vi.fn(), FOREIGN_SCOPE)).toThrow(
+      ForbiddenException,
+    );
+    const cleanup = service.onEvent(CONVERSATION_ID, vi.fn(), OWNER_SCOPE);
+    cleanup();
+
+    await expect(
+      service.prompt(CONVERSATION_ID, 'foreign prompt', FOREIGN_SCOPE),
+    ).rejects.toBeInstanceOf(ForbiddenException);
+    await service.prompt(CONVERSATION_ID, 'owner prompt', OWNER_SCOPE);
+    expect(session.piSession.prompt).toHaveBeenCalledWith('owner prompt');
+
+    await expect(service.destroySession(CONVERSATION_ID, FOREIGN_SCOPE)).rejects.toBeInstanceOf(
+      ForbiddenException,
+    );
+    expect(internals(service).sessions.has(CONVERSATION_ID)).toBe(true);
+
+    await service.destroySession(CONVERSATION_ID, OWNER_SCOPE);
+    expect(session.piSession.dispose).toHaveBeenCalled();
+    expect(internals(service).sessions.has(CONVERSATION_ID)).toBe(false);
+  });
+
+  it('derives the operator-memory scope on the createSession production path', async () => {
+    const plugin = { capture: vi.fn(), search: vi.fn() };
+    const service = makeService(plugin);
+    const buildTools = vi.spyOn(service as never, 'buildToolsForSandbox').mockReturnValue([]);
+
+    // Session construction reaches the real scope derivation before the intentionally incomplete
+    // Pi test double rejects later in createAgentSession.
+    await service.createSession(CONVERSATION_ID, OWNER_SCOPE).catch(() => undefined);
+
+    expect(buildTools).toHaveBeenCalledWith(expect.any(String), OWNER_SCOPE.userId, {
+      tenantId: OWNER_SCOPE.tenantId,
+      ownerId: OWNER_SCOPE.userId,
+      sessionId: CONVERSATION_ID,
+    });
+  });
+
+  it('denies a foreign actor before it can obtain another session operator-memory scope', async () => {
+    const plugin = { capture: vi.fn(), search: vi.fn() };
+    const service = makeService(plugin);
+    internals(service).sessions.set(CONVERSATION_ID, makeSession());
+    const buildTools = vi.spyOn(service as never, 'buildToolsForSandbox');
+
+    await expect(service.createSession(CONVERSATION_ID, FOREIGN_SCOPE)).rejects.toBeInstanceOf(
+      ForbiddenException,
+    );
+
+    expect(buildTools).not.toHaveBeenCalled();
+    expect(plugin.capture).not.toHaveBeenCalled();
+    expect(plugin.search).not.toHaveBeenCalled();
+  });
+
+  it('checks owner/tenant scope before returning an in-flight session creation', async () => {
+    const service = makeService();
+    const session = makeSession();
+    internals(service).creating.set(CONVERSATION_ID, Promise.resolve(session));
+
+    await expect(
+      service.createSession(CONVERSATION_ID, {
+        userId: FOREIGN_SCOPE.userId,
+        tenantId: FOREIGN_SCOPE.tenantId,
+      }),
+    ).rejects.toBeInstanceOf(ForbiddenException);
+
+    await expect(
+      service.createSession(CONVERSATION_ID, {
+        userId: OWNER_SCOPE.userId,
+        tenantId: OWNER_SCOPE.tenantId,
+      }),
+    ).resolves.toBe(session);
+  });
+});

apps/gateway/src/agent/__tests__/runtime-provider-registry.service.test.ts

@@ -0,0 +1,370 @@
+import { describe, expect, it } from 'vitest';
+import type {
+  AgentRuntimeProvider,
+  RuntimeAttachHandle,
+  RuntimeAttachMode,
+  RuntimeCapability,
+  RuntimeCapabilitySet,
+  RuntimeHealth,
+  RuntimeMessage,
+  RuntimeScope,
+  RuntimeSession,
+  RuntimeSessionTree,
+  RuntimeStreamEvent,
+} from '@mosaicstack/types';
+import { AgentRuntimeProviderRegistry } from '@mosaicstack/agent';
+import type { ActorTenantScope } from '../../auth/session-scope.js';
+import {
+  RuntimeProviderAuditService,
+  RuntimeProviderService,
+  type RuntimeAuditEvent,
+  type RuntimeAuditSink,
+  type RuntimeApprovalVerifier,
+} from '../runtime-provider-registry.service.js';
+
+process.env['MOSAIC_AGENT_NAME'] ??= 'test-runtime-agent';
+
+const OWNER_SCOPE: ActorTenantScope = { userId: 'owner-1', tenantId: 'tenant-1' };
+const CONTEXT = {
+  actorScope: OWNER_SCOPE,
+  channelId: 'cli',
+  correlationId: 'correlation-1',
+};
+
+class RecordingRuntimeProvider implements AgentRuntimeProvider {
+  readonly id = 'fleet';
+  readonly receivedScopes: RuntimeScope[] = [];
+  readonly sentMessages: RuntimeMessage[] = [];
+  terminateCalls = 0;
+  throwAfterSend = false;
+  throwAuthorization = false;
+
+  constructor(private readonly supported: RuntimeCapability[]) {}
+
+  async capabilities(scope: RuntimeScope): Promise<RuntimeCapabilitySet> {
+    this.receivedScopes.push(scope);
+    return { supported: this.supported };
+  }
+
+  async health(scope: RuntimeScope): Promise<RuntimeHealth> {
+    this.receivedScopes.push(scope);
+    return { status: 'healthy', checkedAt: '2026-07-12T00:00:00.000Z' };
+  }
+
+  async listSessions(scope: RuntimeScope): Promise<RuntimeSession[]> {
+    this.receivedScopes.push(scope);
+    return [];
+  }
+
+  async getSessionTree(scope: RuntimeScope): Promise<RuntimeSessionTree[]> {
+    this.receivedScopes.push(scope);
+    return [];
+  }
+
+  async *streamSession(
+    _sessionId: string,
+    _cursor: string | undefined,
+    scope: RuntimeScope,
+  ): AsyncIterable<RuntimeStreamEvent> {
+    this.receivedScopes.push(scope);
+    return;
+  }
+
+  async sendMessage(
+    _sessionId: string,
+    message: RuntimeMessage,
+    scope: RuntimeScope,
+  ): Promise<void> {
+    this.receivedScopes.push(scope);
+    this.sentMessages.push(message);
+    if (this.throwAuthorization) {
+      throw Object.assign(new Error('provider authorization denied'), { code: 'forbidden' });
+    }
+    if (this.throwAfterSend) {
+      throw new Error('provider acknowledgement failed');
+    }
+  }
+
+  async attach(
+    sessionId: string,
+    mode: RuntimeAttachMode,
+    scope: RuntimeScope,
+  ): Promise<RuntimeAttachHandle> {
+    this.receivedScopes.push(scope);
+    return {
+      attachmentId: 'attachment-1',
+      sessionId,
+      mode,
+      expiresAt: '2026-07-12T00:00:00.000Z',
+    };
+  }
+
+  async detach(_attachmentId: string, scope: RuntimeScope): Promise<void> {
+    this.receivedScopes.push(scope);
+  }
+
+  async terminate(_sessionId: string, _approvalRef: string, scope: RuntimeScope): Promise<void> {
+    this.receivedScopes.push(scope);
+    this.terminateCalls += 1;
+  }
+}
+
+class RecordingAuditSink implements RuntimeAuditSink {
+  readonly events: RuntimeAuditEvent[] = [];
+
+  async record(event: RuntimeAuditEvent): Promise<void> {
+    this.events.push(event);
+  }
+}
+
+class DenyingApprovalVerifier implements RuntimeApprovalVerifier {
+  async consume(): Promise<boolean> {
+    return false;
+  }
+}
+
+class AcceptingApprovalVerifier implements RuntimeApprovalVerifier {
+  consumedAction: Parameters<RuntimeApprovalVerifier['consume']>[1] | undefined;
+
+  async consume(
+    _approvalRef: string,
+    action: Parameters<RuntimeApprovalVerifier['consume']>[1],
+  ): Promise<boolean> {
+    this.consumedAction = action;
+    return true;
+  }
+}
+
+function makeService(
+  provider: RecordingRuntimeProvider,
+  audit: RuntimeAuditSink = new RecordingAuditSink(),
+  approval: RuntimeApprovalVerifier = new DenyingApprovalVerifier(),
+): RuntimeProviderService {
+  const registry = new AgentRuntimeProviderRegistry();
+  registry.register(provider);
+  return new RuntimeProviderService(registry, audit, approval);
+}
+
+describe('RuntimeProviderService security boundary', (): void => {
+  it('derives and freezes only the authenticated actor scope while preserving correlation metadata', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.send']);
+    const audit = new RecordingAuditSink();
+    const service = makeService(provider, audit);
+
+    await service.sendMessage(
+      'fleet',
+      'session-1',
+      { content: 'hello', idempotencyKey: 'key-1' },
+      CONTEXT,
+    );
+
+    const providerScope = provider.receivedScopes[0];
+    expect(providerScope).toEqual({
+      actorId: OWNER_SCOPE.userId,
+      tenantId: OWNER_SCOPE.tenantId,
+      channelId: CONTEXT.channelId,
+      correlationId: CONTEXT.correlationId,
+    });
+    expect(Object.isFrozen(providerScope)).toBe(true);
+    expect(audit.events).toContainEqual(
+      expect.objectContaining({
+        providerId: 'fleet',
+        operation: 'session.send',
+        outcome: 'succeeded',
+        actorId: OWNER_SCOPE.userId,
+        tenantId: OWNER_SCOPE.tenantId,
+        channelId: CONTEXT.channelId,
+        correlationId: CONTEXT.correlationId,
+        resourceId: 'session-1',
+        durationMs: expect.any(Number),
+      }),
+    );
+    expect(JSON.stringify(audit.events)).not.toContain('hello');
+    expect(JSON.stringify(audit.events)).not.toContain('key-1');
+  });
+
+  it('does not block a provider operation when an unsafe resource ID is redacted in durable audit', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.send']);
+    let persisted: unknown;
+    const durableAudit = new RuntimeProviderAuditService({
+      logs: {
+        ingest: async (entry: unknown): Promise<unknown> => {
+          persisted = entry;
+          return entry;
+        },
+      },
+    } as never);
+    const service = makeService(provider, durableAudit);
+
+    await service.sendMessage(
+      'fleet',
+      'session/credential-canary=secret-value',
+      { content: 'safe message', idempotencyKey: 'key-1' },
+      CONTEXT,
+    );
+
+    expect(provider.sentMessages).toHaveLength(1);
+    expect(JSON.stringify(persisted)).not.toContain('secret-value');
+  });
+
+  it('fails closed before a provider side effect when a capability is missing', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider([]);
+    const service = makeService(provider);
+
+    await expect(
+      service.sendMessage(
+        'fleet',
+        'session-1',
+        { content: 'hello', idempotencyKey: 'key-1' },
+        CONTEXT,
+      ),
+    ).rejects.toThrow(/capability denied/);
+    expect(provider.sentMessages).toEqual([]);
+  });
+
+  it('requires a consumed exact-action approval before termination', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.terminate']);
+    const approval = new DenyingApprovalVerifier();
+    const audit = new RecordingAuditSink();
+    const service = makeService(provider, audit, approval);
+
+    await expect(
+      service.terminate('fleet', 'session-1', 'forged-approval', CONTEXT),
+    ).rejects.toThrow(/approval denied/);
+    expect(provider.terminateCalls).toBe(0);
+    expect(audit.events.at(-1)).toMatchObject({ outcome: 'denied', errorCode: 'policy_denied' });
+  });
+
+  it('binds an accepted termination approval to provider, session, immutable scope, and correlation', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.terminate']);
+    const approval = new AcceptingApprovalVerifier();
+    const service = makeService(provider, new RecordingAuditSink(), approval);
+
+    await service.terminate('fleet', 'session-1', 'approval-1', CONTEXT);
+
+    expect(approval.consumedAction).toEqual({
+      providerId: 'fleet',
+      sessionId: 'session-1',
+      actorId: OWNER_SCOPE.userId,
+      tenantId: OWNER_SCOPE.tenantId,
+      channelId: CONTEXT.channelId,
+      correlationId: CONTEXT.correlationId,
+      agentName: process.env['MOSAIC_AGENT_NAME'],
+    });
+    expect(provider.terminateCalls).toBe(1);
+  });
+
+  it('fails closed before invoking a provider when audit persistence rejects the request', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.send']);
+    const unavailableAudit: RuntimeAuditSink = {
+      async record(): Promise<void> {
+        throw new Error('audit unavailable');
+      },
+    };
+    const service = makeService(provider, unavailableAudit);
+
+    await expect(
+      service.sendMessage(
+        'fleet',
+        'session-1',
+        { content: 'hello', idempotencyKey: 'key-1' },
+        CONTEXT,
+      ),
+    ).rejects.toThrow(/audit unavailable/);
+    expect(provider.sentMessages).toEqual([]);
+  });
+
+  it('records a provider error after invocation as failed rather than denied', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.send']);
+    provider.throwAfterSend = true;
+    const audit = new RecordingAuditSink();
+    const service = makeService(provider, audit);
+
+    await expect(
+      service.sendMessage(
+        'fleet',
+        'session-1',
+        { content: 'hello', idempotencyKey: 'key-1' },
+        CONTEXT,
+      ),
+    ).rejects.toThrow(/provider acknowledgement failed/);
+    expect(provider.sentMessages).toHaveLength(1);
+    expect(audit.events.map((event: RuntimeAuditEvent): string => event.outcome)).toEqual([
+      'requested',
+      'failed',
+    ]);
+    expect(audit.events.at(-1)).toMatchObject({
+      errorCode: 'provider_error',
+      durationMs: expect.any(Number),
+    });
+  });
+
+  it('records a provider authorization rejection as denied rather than provider failure', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.send']);
+    provider.throwAuthorization = true;
+    const audit = new RecordingAuditSink();
+    const service = makeService(provider, audit);
+
+    await expect(
+      service.sendMessage(
+        'fleet',
+        'session-1',
+        { content: 'hello', idempotencyKey: 'key-1' },
+        CONTEXT,
+      ),
+    ).rejects.toThrow(/provider authorization denied/);
+    expect(audit.events.at(-1)).toMatchObject({ outcome: 'denied', errorCode: 'policy_denied' });
+  });
+
+  it('persists only metadata-only runtime audit fields', async (): Promise<void> => {
+    let persisted: unknown;
+    const ingest = async (entry: unknown): Promise<unknown> => {
+      persisted = entry;
+      return entry;
+    };
+    const service = new RuntimeProviderAuditService({ logs: { ingest } } as never);
+
+    await service.record({
+      providerId: 'fleet',
+      operation: 'session.send',
+      outcome: 'succeeded',
+      actorId: 'owner-1',
+      tenantId: 'tenant-1',
+      channelId: 'cli',
+      correlationId: 'correlation-1',
+      resourceId: 'session-1',
+      durationMs: 12,
+    });
+
+    expect(persisted).toMatchObject({
+      content: 'runtime.provider.audit',
+      metadata: expect.objectContaining({ correlationId: 'correlation-1', durationMs: 12 }),
+    });
+    expect(JSON.stringify(persisted)).not.toContain('approval');
+  });
+
+  it('does not misreport a completed provider side effect when completion auditing fails', async (): Promise<void> => {
+    const provider = new RecordingRuntimeProvider(['session.send']);
+    let auditCalls = 0;
+    const audit: RuntimeAuditSink = {
+      async record(): Promise<void> {
+        auditCalls += 1;
+        if (auditCalls === 2) {
+          throw new Error('completion audit unavailable');
+        }
+      },
+    };
+    const service = makeService(provider, audit);
+
+    await expect(
+      service.sendMessage(
+        'fleet',
+        'session-1',
+        { content: 'hello', idempotencyKey: 'key-1' },
+        CONTEXT,
+      ),
+    ).resolves.toBeUndefined();
+    expect(provider.sentMessages).toHaveLength(1);
+    expect(auditCalls).toBe(2);
+  });
+});

apps/gateway/src/agent/__tests__/session-ownership.test.ts

@@ -0,0 +1,262 @@
+import { readFileSync } from 'node:fs';
+import { resolve } from 'node:path';
+import { ForbiddenException, NotFoundException } from '@nestjs/common';
+import { describe, expect, it, vi } from 'vitest';
+
+vi.mock('../agent.service.js', () => ({ AgentService: class AgentService {} }));
+vi.mock('../../commands/command-executor.service.js', () => ({
+  CommandExecutorService: class CommandExecutorService {},
+}));
+vi.mock('../routing/routing-engine.service.js', () => ({
+  RoutingEngineService: class RoutingEngineService {},
+}));
+
+import { SessionsController } from '../sessions.controller.js';
+import { ChatController } from '../../chat/chat.controller.js';
+import { ChatGateway } from '../../chat/chat.gateway.js';
+import type { AgentSession } from '../agent.service.js';
+import type { SessionInfoDto } from '../session.dto.js';
+
+const USER_A = { id: 'user-a', tenantId: 'tenant-a' };
+const USER_B = { id: 'user-b', tenantId: 'tenant-b' };
+const CONVERSATION_ID = '11111111-1111-4111-8111-111111111111';
+
+function makeSessionInfo(overrides?: Partial<SessionInfoDto>): SessionInfoDto {
+  return {
+    id: CONVERSATION_ID,
+    provider: 'test-provider',
+    modelId: 'test-model',
+    createdAt: new Date('2026-07-12T00:00:00Z').toISOString(),
+    promptCount: 0,
+    channels: [],
+    durationMs: 0,
+    metrics: {
+      tokens: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+      modelSwitches: 0,
+      messageCount: 0,
+      lastActivityAt: new Date('2026-07-12T00:00:00Z').toISOString(),
+    },
+    ...overrides,
+  };
+}
+
+function makeAgentSession(owner = USER_A): AgentSession {
+  return {
+    id: CONVERSATION_ID,
+    provider: 'test-provider',
+    modelId: 'test-model',
+    piSession: {
+      thinkingLevel: 'off',
+      getAvailableThinkingLevels: vi.fn().mockReturnValue(['off', 'low', 'high']),
+      setThinkingLevel: vi.fn(),
+      abort: vi.fn().mockResolvedValue(undefined),
+      prompt: vi.fn().mockResolvedValue(undefined),
+      dispose: vi.fn(),
+      getSessionStats: vi.fn(),
+      getContextUsage: vi.fn(),
+    } as unknown as AgentSession['piSession'],
+    listeners: new Set(),
+    unsubscribe: vi.fn(),
+    createdAt: Date.now(),
+    promptCount: 0,
+    channels: new Set(),
+    skillPromptAdditions: [],
+    sandboxDir: '/tmp',
+    allowedTools: null,
+    userId: owner.id,
+    tenantId: owner.tenantId,
+    metrics: {
+      tokens: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+      modelSwitches: 0,
+      messageCount: 0,
+      lastActivityAt: new Date('2026-07-12T00:00:00Z').toISOString(),
+    },
+  };
+}
+
+function makeScopedAgentService() {
+  const foreign = makeAgentSession(USER_A);
+  return {
+    listSessions: vi.fn((scope?: { userId: string; tenantId?: string }) =>
+      scope?.userId === USER_B.id ? [] : [makeSessionInfo({ id: foreign.id })],
+    ),
+    getSessionInfo: vi.fn((_id: string, scope?: { userId: string; tenantId?: string }) =>
+      scope?.userId === USER_B.id ? undefined : makeSessionInfo({ id: foreign.id }),
+    ),
+    destroySession: vi.fn(),
+    getSession: vi.fn((_id: string, scope?: { userId: string; tenantId?: string }) =>
+      scope?.userId === USER_B.id ? undefined : foreign,
+    ),
+    createSession: vi.fn().mockRejectedValue(new ForbiddenException('Session scope mismatch')),
+    onEvent: vi.fn(() => vi.fn()),
+    addChannel: vi.fn(),
+    removeChannel: vi.fn(),
+    recordMessage: vi.fn(),
+    prompt: vi.fn().mockResolvedValue(undefined),
+  };
+}
+
+describe('TESS-M1-SEC-002 AgentService ownership boundary', () => {
+  it('requires explicit owner+tenant scope on protected session operations', () => {
+    const source = readFileSync(resolve('src/agent/agent.service.ts'), 'utf8');
+
+    expect(source).toContain('getSession(sessionId: string, scope: ActorTenantScope)');
+    expect(source).toContain('listSessions(scope: ActorTenantScope)');
+    expect(source).toContain('getSessionInfo(sessionId: string, scope: ActorTenantScope)');
+    expect(source).toContain(
+      'addChannel(sessionId: string, channel: string, scope: ActorTenantScope)',
+    );
+    expect(source).toContain(
+      'removeChannel(sessionId: string, channel: string, scope: ActorTenantScope)',
+    );
+    expect(source).toContain(
+      'async prompt(sessionId: string, message: string, scope: ActorTenantScope)',
+    );
+    expect(source).toContain('scope: ActorTenantScope,');
+    expect(source).toContain('async destroySession(sessionId: string, scope: ActorTenantScope)');
+    expect(source).not.toContain('scope?: ActorTenantScope');
+  });
+});
+
+describe('TESS-M1-SEC-002 REST session ownership and tenant binding', () => {
+  it('lists only sessions owned by the authenticated owner+tenant scope', () => {
+    const agentService = makeScopedAgentService();
+    const controller = new SessionsController(agentService as never);
+
+    expect(controller.list(USER_B)).toEqual({ sessions: [], total: 0 });
+    expect(agentService.listSessions).toHaveBeenCalledWith({
+      userId: USER_B.id,
+      tenantId: USER_B.tenantId,
+    });
+  });
+
+  it('does not reveal another owner/tenant session by guessed id', () => {
+    const agentService = makeScopedAgentService();
+    const controller = new SessionsController(agentService as never);
+
+    expect(() => controller.findOne(CONVERSATION_ID, USER_B)).toThrow(NotFoundException);
+    expect(agentService.getSessionInfo).toHaveBeenCalledWith(CONVERSATION_ID, {
+      userId: USER_B.id,
+      tenantId: USER_B.tenantId,
+    });
+  });
+
+  it('does not terminate another owner/tenant session by guessed id', async () => {
+    const agentService = makeScopedAgentService();
+    const controller = new SessionsController(agentService as never);
+
+    await expect(controller.destroy(CONVERSATION_ID, USER_B)).rejects.toBeInstanceOf(
+      NotFoundException,
+    );
+    expect(agentService.destroySession).not.toHaveBeenCalled();
+  });
+});
+
+describe('TESS-M1-SEC-002 REST chat send ownership and tenant binding', () => {
+  it('does not send a prompt into another owner/tenant session by guessed conversationId', async () => {
+    const agentService = makeScopedAgentService();
+    const controller = new ChatController(agentService as never);
+
+    await expect(
+      controller.chat({ conversationId: CONVERSATION_ID, content: 'take over' }, USER_B),
+    ).rejects.toMatchObject({ status: 404 });
+
+    expect(agentService.getSession).toHaveBeenCalledWith(CONVERSATION_ID, {
+      userId: USER_B.id,
+      tenantId: USER_B.tenantId,
+    });
+    expect(agentService.prompt).not.toHaveBeenCalled();
+  });
+});
+
+describe('TESS-M1-SEC-002 WebSocket session ownership and tenant binding', () => {
+  function makeGateway(agentService = makeScopedAgentService()) {
+    const brain = {
+      conversations: {
+        findById: vi.fn().mockResolvedValue(undefined),
+        create: vi.fn().mockResolvedValue(undefined),
+        update: vi.fn().mockResolvedValue(undefined),
+        findMessages: vi.fn().mockResolvedValue([]),
+        addMessage: vi.fn().mockResolvedValue(undefined),
+      },
+    };
+    const commandRegistry = { getManifest: vi.fn().mockReturnValue([]) };
+    const commandExecutor = { execute: vi.fn() };
+    const routingEngine = {
+      resolve: vi.fn().mockResolvedValue({ provider: 'test', model: 'test-model' }),
+    };
+    const gateway = new ChatGateway(
+      agentService as never,
+      {} as never,
+      brain as never,
+      commandRegistry as never,
+      commandExecutor as never,
+      routingEngine as never,
+    );
+    return { gateway, agentService };
+  }
+
+  function makeSocket() {
+    return {
+      id: 'socket-b',
+      connected: true,
+      data: { user: USER_B, session: { id: 'auth-session-b', userId: USER_B.id } },
+      emit: vi.fn(),
+      disconnect: vi.fn(),
+    };
+  }
+
+  it('does not attach or send to another owner/tenant session by guessed conversationId', async () => {
+    const { gateway, agentService } = makeGateway();
+    const socket = makeSocket();
+
+    await gateway.handleMessage(socket as never, {
+      conversationId: CONVERSATION_ID,
+      content: 'attach to foreign session',
+    });
+
+    expect(agentService.getSession).toHaveBeenCalledWith(CONVERSATION_ID, {
+      userId: USER_B.id,
+      tenantId: USER_B.tenantId,
+    });
+    expect(agentService.onEvent).not.toHaveBeenCalled();
+    expect(agentService.addChannel).not.toHaveBeenCalled();
+    expect(agentService.prompt).not.toHaveBeenCalled();
+    expect(socket.emit).toHaveBeenCalledWith(
+      'error',
+      expect.objectContaining({ conversationId: CONVERSATION_ID }),
+    );
+  });
+
+  it('does not mutate thinking level on another owner/tenant session', () => {
+    const { gateway, agentService } = makeGateway();
+    const socket = makeSocket();
+
+    gateway.handleSetThinking(socket as never, { conversationId: CONVERSATION_ID, level: 'high' });
+
+    expect(agentService.getSession).toHaveBeenCalledWith(CONVERSATION_ID, {
+      userId: USER_B.id,
+      tenantId: USER_B.tenantId,
+    });
+    expect(socket.emit).toHaveBeenCalledWith(
+      'error',
+      expect.objectContaining({ conversationId: CONVERSATION_ID }),
+    );
+  });
+
+  it('does not terminate another owner/tenant session over WebSocket abort', async () => {
+    const { gateway, agentService } = makeGateway();
+    const socket = makeSocket();
+
+    await gateway.handleAbort(socket as never, { conversationId: CONVERSATION_ID });
+
+    expect(agentService.getSession).toHaveBeenCalledWith(CONVERSATION_ID, {
+      userId: USER_B.id,
+      tenantId: USER_B.tenantId,
+    });
+    expect(socket.emit).toHaveBeenCalledWith(
+      'error',
+      expect.objectContaining({ conversationId: CONVERSATION_ID }),
+    );
+  });
+});

apps/gateway/src/agent/agent.module.ts

@@ -1,4 +1,5 @@
 import { Global, Module } from '@nestjs/common';
+import { AgentRuntimeProviderRegistry, HermesRuntimeProvider } from '@mosaicstack/agent';
 import { AgentService } from './agent.service.js';
 import { ProviderService } from './provider.service.js';
 import { ProviderCredentialsService } from './provider-credentials.service.js';
@@ -8,24 +9,66 @@ import { SkillLoaderService } from './skill-loader.service.js';
 import { ProvidersController } from './providers.controller.js';
 import { SessionsController } from './sessions.controller.js';
 import { AgentConfigsController } from './agent-configs.controller.js';
+import { InteractionController } from './interaction.controller.js';
 import { RoutingController } from './routing/routing.controller.js';
+import { DurableSessionRepository } from './durable-session.repository.js';
+import { DurableSessionService } from './durable-session.service.js';
 import { CoordModule } from '../coord/coord.module.js';
 import { McpClientModule } from '../mcp-client/mcp-client.module.js';
 import { SkillsModule } from '../skills/skills.module.js';
 import { GCModule } from '../gc/gc.module.js';
+import { LogModule } from '../log/log.module.js';
+import { CommandsModule } from '../commands/commands.module.js';
+import { CommandRuntimeApprovalVerifier } from '../commands/runtime-approval-verifier.js';
+import { GatewayHermesRuntimeTransport } from './hermes-runtime.transport.js';
+import {
+  AGENT_RUNTIME_PROVIDER_REGISTRY,
+  RUNTIME_APPROVAL_VERIFIER,
+  RUNTIME_PROVIDER_AUDIT_SINK,
+  RuntimeProviderAuditService,
+  RuntimeProviderService,
+} from './runtime-provider-registry.service.js';
+
+export function createGatewayRuntimeProviderRegistry(): AgentRuntimeProviderRegistry {
+  const registry = new AgentRuntimeProviderRegistry();
+  registry.register(new HermesRuntimeProvider(new GatewayHermesRuntimeTransport()));
+  return registry;
+}
 
 @Global()
 @Module({
-  imports: [CoordModule, McpClientModule, SkillsModule, GCModule],
+  imports: [CoordModule, McpClientModule, SkillsModule, GCModule, LogModule, CommandsModule],
   providers: [
     ProviderService,
     ProviderCredentialsService,
     RoutingService,
     RoutingEngineService,
     SkillLoaderService,
+    DurableSessionRepository,
+    DurableSessionService,
+    {
+      provide: AGENT_RUNTIME_PROVIDER_REGISTRY,
+      useFactory: createGatewayRuntimeProviderRegistry,
+    },
+    RuntimeProviderAuditService,
+    {
+      provide: RUNTIME_PROVIDER_AUDIT_SINK,
+      useExisting: RuntimeProviderAuditService,
+    },
+    {
+      provide: RUNTIME_APPROVAL_VERIFIER,
+      useExisting: CommandRuntimeApprovalVerifier,
+    },
+    RuntimeProviderService,
     AgentService,
   ],
-  controllers: [ProvidersController, SessionsController, AgentConfigsController, RoutingController],
+  controllers: [
+    ProvidersController,
+    SessionsController,
+    AgentConfigsController,
+    InteractionController,
+    RoutingController,
+  ],
   exports: [
     AgentService,
     ProviderService,
@@ -33,6 +76,9 @@ import { GCModule } from '../gc/gc.module.js';
     RoutingService,
     RoutingEngineService,
     SkillLoaderService,
+    DurableSessionService,
+    RuntimeProviderService,
+    AGENT_RUNTIME_PROVIDER_REGISTRY,
   ],
 })
 export class AgentModule {}

apps/gateway/src/agent/agent.service.ts

@@ -1,4 +1,11 @@
-import { Inject, Injectable, Logger, Optional, type OnModuleDestroy } from '@nestjs/common';
+import {
+  ForbiddenException,
+  Inject,
+  Injectable,
+  Logger,
+  Optional,
+  type OnModuleDestroy,
+} from '@nestjs/common';
 import {
   createAgentSession,
   DefaultResourceLoader,
@@ -8,9 +15,10 @@ import {
   type ToolDefinition,
 } from '@mariozechner/pi-coding-agent';
 import type { Brain } from '@mosaicstack/brain';
-import type { Memory } from '@mosaicstack/memory';
+import type { Memory, OperatorMemoryPlugin } from '@mosaicstack/memory';
 import { BRAIN } from '../brain/brain.tokens.js';
 import { MEMORY } from '../memory/memory.tokens.js';
+import { OPERATOR_MEMORY_PLUGIN } from '../memory/memory.module.js';
 import { EmbeddingService } from '../memory/embedding.service.js';
 import { CoordService } from '../coord/coord.service.js';
 import { ProviderService } from './provider.service.js';
@@ -28,6 +36,7 @@ import type { SessionInfoDto, SessionMetrics } from './session.dto.js';
 import { SystemOverrideService } from '../preferences/system-override.service.js';
 import { PreferencesService } from '../preferences/preferences.service.js';
 import { SessionGCService } from '../gc/session-gc.service.js';
+import type { ActorTenantScope } from '../auth/session-scope.js';
 
 /** A single message from DB conversation history, used for context injection. */
 export interface ConversationHistoryMessage {
@@ -68,6 +77,8 @@ export interface AgentSessionOptions {
   agentConfigId?: string;
   /** ID of the user who owns this session. Used for preferences and system override lookups. */
   userId?: string;
+  /** Server-derived tenant scope that owns this session. Falls back to userId for solo users. */
+  tenantId?: string;
   /**
    * Prior conversation messages to inject as context when resuming a session.
    * These messages are formatted and prepended to the system prompt so the
@@ -94,6 +105,8 @@ export interface AgentSession {
   allowedTools: string[] | null;
   /** User ID that owns this session, used for preference lookups. */
   userId?: string;
+  /** Server-derived tenant scope that owns this session. Falls back to userId for solo users. */
+  tenantId?: string;
   /** Agent config ID applied to this session, if any (M5-001). */
   agentConfigId?: string;
   /** Human-readable agent name applied to this session, if any (M5-001). */
@@ -123,6 +136,9 @@ export class AgentService implements OnModuleDestroy {
     @Inject(PreferencesService)
     private readonly preferencesService: PreferencesService | null,
     @Inject(SessionGCService) private readonly gc: SessionGCService,
+    @Optional()
+    @Inject(OPERATOR_MEMORY_PLUGIN)
+    private readonly operatorMemory: OperatorMemoryPlugin | null = null,
   ) {}
 
   /**
@@ -134,6 +150,7 @@ export class AgentService implements OnModuleDestroy {
   private buildToolsForSandbox(
     sandboxDir: string,
     sessionUserId: string | undefined,
+    sessionScope?: { tenantId: string; ownerId: string; sessionId: string },
   ): ToolDefinition[] {
     return [
       ...createBrainTools(this.brain),
@@ -142,6 +159,9 @@ export class AgentService implements OnModuleDestroy {
         this.memory,
         this.embeddingService.available ? this.embeddingService : null,
         sessionUserId,
+        this.operatorMemory && sessionScope
+          ? { plugin: this.operatorMemory, scope: sessionScope }
+          : undefined,
       ),
       ...createFileTools(sandboxDir),
       ...createGitTools(sandboxDir),
@@ -174,12 +194,20 @@ export class AgentService implements OnModuleDestroy {
       .filter((t) => t.length > 0);
   }
 
-  async createSession(sessionId: string, options?: AgentSessionOptions): Promise<AgentSession> {
+  async createSession(sessionId: string, options: AgentSessionOptions): Promise<AgentSession> {
+    const scope = this.scopeFromOptions(options);
     const existing = this.sessions.get(sessionId);
-    if (existing) return existing;
+    if (existing) {
+      this.assertSessionScope(existing, scope);
+      return existing;
+    }
 
     const inflight = this.creating.get(sessionId);
-    if (inflight) return inflight;
+    if (inflight) {
+      const session = await inflight;
+      this.assertSessionScope(session, scope);
+      return session;
+    }
 
     const promise = this.doCreateSession(sessionId, options).finally(() => {
       this.creating.delete(sessionId);
@@ -208,6 +236,7 @@ export class AgentService implements OnModuleDestroy {
           isAdmin: options.isAdmin,
           agentConfigId: options.agentConfigId,
           userId: options.userId,
+          tenantId: options.tenantId,
           conversationHistory: options.conversationHistory,
         };
         this.logger.log(
@@ -247,7 +276,15 @@ export class AgentService implements OnModuleDestroy {
     }
 
     // Build per-session tools scoped to the sandbox directory and authenticated user
-    const sandboxTools = this.buildToolsForSandbox(sandboxDir, mergedOptions?.userId);
+    const sessionUserId = mergedOptions?.userId;
+    const sessionTenantId = this.tenantIdFor(sessionUserId, mergedOptions?.tenantId);
+    const sandboxTools = this.buildToolsForSandbox(
+      sandboxDir,
+      sessionUserId,
+      sessionUserId && sessionTenantId
+        ? { tenantId: sessionTenantId, ownerId: sessionUserId, sessionId }
+        : undefined,
+    );
 
     // Combine static tools with dynamically discovered MCP client tools and skill tools
     const mcpTools = this.mcpClientService.getToolDefinitions();
@@ -342,6 +379,7 @@ export class AgentService implements OnModuleDestroy {
       sandboxDir,
       allowedTools,
       userId: mergedOptions?.userId,
+      tenantId: sessionTenantId,
       agentConfigId: mergedOptions?.agentConfigId,
       agentName: resolvedAgentName,
       metrics: {
@@ -473,38 +511,70 @@ export class AgentService implements OnModuleDestroy {
     return this.providerService.getDefaultModel() ?? null;
   }
 
-  getSession(sessionId: string): AgentSession | undefined {
-    return this.sessions.get(sessionId);
+  getSession(sessionId: string, scope: ActorTenantScope): AgentSession | undefined {
+    const session = this.sessions.get(sessionId);
+    if (!session || !this.sessionMatchesScope(session, scope)) return undefined;
+    return session;
   }
 
-  listSessions(): SessionInfoDto[] {
+  listSessions(scope: ActorTenantScope): SessionInfoDto[] {
     const now = Date.now();
-    return Array.from(this.sessions.values()).map((s) => ({
-      id: s.id,
-      provider: s.provider,
-      modelId: s.modelId,
-      ...(s.agentName ? { agentName: s.agentName } : {}),
-      createdAt: new Date(s.createdAt).toISOString(),
-      promptCount: s.promptCount,
-      channels: Array.from(s.channels),
-      durationMs: now - s.createdAt,
-      metrics: { ...s.metrics },
-    }));
+    return Array.from(this.sessions.values())
+      .filter((s) => this.sessionMatchesScope(s, scope))
+      .map((s) => this.toSessionInfo(s, now));
   }
 
-  getSessionInfo(sessionId: string): SessionInfoDto | undefined {
+  listAllSessionsForSystem(): SessionInfoDto[] {
+    const now = Date.now();
+    return Array.from(this.sessions.values()).map((s) => this.toSessionInfo(s, now));
+  }
+
+  getSessionInfo(sessionId: string, scope: ActorTenantScope): SessionInfoDto | undefined {
     const s = this.sessions.get(sessionId);
-    if (!s) return undefined;
+    if (!s || !this.sessionMatchesScope(s, scope)) return undefined;
+    return this.toSessionInfo(s);
+  }
+
+  private scopeFromOptions(options: AgentSessionOptions): ActorTenantScope {
+    if (!options.userId) {
+      throw new ForbiddenException('Session owner scope is required');
+    }
     return {
-      id: s.id,
-      provider: s.provider,
-      modelId: s.modelId,
-      ...(s.agentName ? { agentName: s.agentName } : {}),
-      createdAt: new Date(s.createdAt).toISOString(),
-      promptCount: s.promptCount,
-      channels: Array.from(s.channels),
-      durationMs: Date.now() - s.createdAt,
-      metrics: { ...s.metrics },
+      userId: options.userId,
+      tenantId: this.tenantIdFor(options.userId, options.tenantId) ?? options.userId,
+    };
+  }
+
+  private tenantIdFor(
+    userId: string | undefined,
+    tenantId: string | undefined,
+  ): string | undefined {
+    return tenantId ?? userId;
+  }
+
+  private sessionMatchesScope(session: AgentSession, scope: ActorTenantScope): boolean {
+    return (
+      session.userId === scope.userId && (session.tenantId ?? session.userId) === scope.tenantId
+    );
+  }
+
+  private assertSessionScope(session: AgentSession, scope: ActorTenantScope): void {
+    if (!this.sessionMatchesScope(session, scope)) {
+      throw new ForbiddenException('Session does not belong to the current owner/tenant scope');
+    }
+  }
+
+  private toSessionInfo(session: AgentSession, now = Date.now()): SessionInfoDto {
+    return {
+      id: session.id,
+      provider: session.provider,
+      modelId: session.modelId,
+      ...(session.agentName ? { agentName: session.agentName } : {}),
+      createdAt: new Date(session.createdAt).toISOString(),
+      promptCount: session.promptCount,
+      channels: Array.from(session.channels),
+      durationMs: now - session.createdAt,
+      metrics: { ...session.metrics },
     };
   }
 
@@ -553,9 +623,10 @@ export class AgentService implements OnModuleDestroy {
    * not reconstructed — the model is used on the next createSession call for
    * the same conversationId when the session is torn down or a new one is created.
    */
-  updateSessionModel(sessionId: string, modelId: string): void {
+  updateSessionModel(sessionId: string, modelId: string, scope: ActorTenantScope): void {
     const session = this.sessions.get(sessionId);
     if (!session) return;
+    this.assertSessionScope(session, scope);
     const prev = session.modelId;
     session.modelId = modelId;
     this.recordModelSwitch(sessionId);
@@ -572,48 +643,51 @@ export class AgentService implements OnModuleDestroy {
     sessionId: string,
     agentConfigId: string,
     agentName: string,
+    scope: ActorTenantScope,
     modelId?: string,
   ): void {
     const session = this.sessions.get(sessionId);
     if (!session) return;
+    this.assertSessionScope(session, scope);
     session.agentConfigId = agentConfigId;
     session.agentName = agentName;
     if (modelId) {
-      this.updateSessionModel(sessionId, modelId);
+      this.updateSessionModel(sessionId, modelId, scope);
     }
     this.logger.log(
       `Session ${sessionId}: agent switched to "${agentName}" (${agentConfigId}) (M5-003)`,
     );
   }
 
-  addChannel(sessionId: string, channel: string): void {
+  addChannel(sessionId: string, channel: string, scope: ActorTenantScope): void {
     const session = this.sessions.get(sessionId);
-    if (session) {
-      session.channels.add(channel);
-    }
+    if (!session) return;
+    this.assertSessionScope(session, scope);
+    session.channels.add(channel);
   }
 
-  removeChannel(sessionId: string, channel: string): void {
+  removeChannel(sessionId: string, channel: string, scope: ActorTenantScope): void {
     const session = this.sessions.get(sessionId);
-    if (session) {
-      session.channels.delete(channel);
-    }
+    if (!session) return;
+    this.assertSessionScope(session, scope);
+    session.channels.delete(channel);
   }
 
-  async prompt(sessionId: string, message: string): Promise<void> {
+  async prompt(sessionId: string, message: string, scope: ActorTenantScope): Promise<void> {
     const session = this.sessions.get(sessionId);
     if (!session) {
       throw new Error(`No agent session found: ${sessionId}`);
     }
+    this.assertSessionScope(session, scope);
     session.promptCount += 1;
 
     // Prepend session-scoped system override if present (renew TTL on each turn)
     let effectiveMessage = message;
     if (this.systemOverride) {
-      const override = await this.systemOverride.get(sessionId);
+      const override = await this.systemOverride.get(sessionId, scope);
       if (override) {
         effectiveMessage = `[System Override]\n${override}\n\n${message}`;
-        await this.systemOverride.renew(sessionId);
+        await this.systemOverride.renew(sessionId, scope);
         this.logger.debug(`Applied system override for session ${sessionId}`);
       }
     }
@@ -629,16 +703,28 @@ export class AgentService implements OnModuleDestroy {
     }
   }
 
-  onEvent(sessionId: string, listener: (event: AgentSessionEvent) => void): () => void {
+  onEvent(
+    sessionId: string,
+    listener: (event: AgentSessionEvent) => void,
+    scope: ActorTenantScope,
+  ): () => void {
     const session = this.sessions.get(sessionId);
     if (!session) {
       throw new Error(`No agent session found: ${sessionId}`);
     }
+    this.assertSessionScope(session, scope);
     session.listeners.add(listener);
     return () => session.listeners.delete(listener);
   }
 
-  async destroySession(sessionId: string): Promise<void> {
+  async destroySession(sessionId: string, scope: ActorTenantScope): Promise<void> {
+    const session = this.sessions.get(sessionId);
+    if (!session) return;
+    this.assertSessionScope(session, scope);
+    await this.destroySessionForSystem(sessionId);
+  }
+
+  private async destroySessionForSystem(sessionId: string): Promise<void> {
     const session = this.sessions.get(sessionId);
     if (!session) return;
     this.logger.log(`Destroying agent session ${sessionId}`);
@@ -667,7 +753,7 @@ export class AgentService implements OnModuleDestroy {
 
   async onModuleDestroy(): Promise<void> {
     this.logger.log('Shutting down all agent sessions');
-    const stops = Array.from(this.sessions.keys()).map((id) => this.destroySession(id));
+    const stops = Array.from(this.sessions.keys()).map((id) => this.destroySessionForSystem(id));
     const results = await Promise.allSettled(stops);
     for (const result of results) {
       if (result.status === 'rejected') {

apps/gateway/src/agent/durable-session.dto.ts

@@ -0,0 +1,10 @@
+import type { RuntimeProviderRequestContext } from './runtime-provider-registry.service.js';
+
+/** Server-side request for a replay-safe provider message. */
+export interface ProviderOutboxDto {
+  sessionId: string;
+  idempotencyKey: string;
+  correlationId: string;
+  content: string;
+  context: RuntimeProviderRequestContext;
+}

apps/gateway/src/agent/durable-session.repository.test.ts

@@ -0,0 +1,416 @@
+import { mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { createHash } from 'node:crypto';
+import { eq, sql, interactionCheckpoints, interactionInbox } from '@mosaicstack/db';
+import { DurableSessionCoordinator, type DurableSessionIdentity } from '@mosaicstack/agent';
+import { afterAll, beforeAll, beforeEach, describe, expect, it, vi } from 'vitest';
+import { createPgliteDb, runPgliteMigrations, type DbHandle } from '@mosaicstack/db';
+import { DurableSessionRepository } from './durable-session.repository.js';
+import { DurableSessionService } from './durable-session.service.js';
+
+const IDENTITY: DurableSessionIdentity = {
+  agentName: 'Nova',
+  sessionId: 'tess-pglite-session',
+  tenantId: 'tenant-pglite',
+  ownerId: 'tess-owner',
+  providerId: 'fleet',
+  runtimeSessionId: 'nova',
+};
+
+describe('DurableSessionRepository', () => {
+  let dataDir: string | undefined;
+  let handle: DbHandle;
+  let previousAuthSecret: string | undefined;
+
+  beforeAll(async (): Promise<void> => {
+    previousAuthSecret = process.env['BETTER_AUTH_SECRET'];
+    process.env['BETTER_AUTH_SECRET'] = 'tess-durable-state-test-sealing-key';
+    dataDir = mkdtempSync(join(tmpdir(), 'tess-durable-state-'));
+    handle = createPgliteDb(dataDir);
+    await runPgliteMigrations(handle);
+    await seedOwner(handle);
+  }, 30_000);
+
+  beforeEach(async (): Promise<void> => {
+    await handle.db.execute(sql`DELETE FROM interaction_handoffs`);
+    await handle.db.execute(sql`DELETE FROM interaction_checkpoints`);
+    await handle.db.execute(sql`DELETE FROM interaction_inbox`);
+    await handle.db.execute(sql`DELETE FROM interaction_outbox`);
+    await handle.db.execute(sql`DELETE FROM interaction_sessions`);
+  });
+
+  afterAll(async (): Promise<void> => {
+    await handle.close();
+    if (dataDir) rmSync(dataDir, { recursive: true, force: true });
+    if (previousAuthSecret === undefined) delete process.env['BETTER_AUTH_SECRET'];
+    else process.env['BETTER_AUTH_SECRET'] = previousAuthSecret;
+  });
+
+  it('survives a full PGlite close/reopen mid-session without duplicate inbox or outbox side effects', async () => {
+    const beforeRestart = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    await beforeRestart.create(IDENTITY);
+    await beforeRestart.receive({
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'inbox-before-kill',
+      correlationId: 'correlation-before-kill',
+      content: 'resume after a kill',
+    });
+    await beforeRestart.enqueueOutbox({
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'outbox-before-kill',
+      correlationId: 'correlation-before-kill',
+      channelId: 'cli',
+      kind: 'provider.send',
+      content: 'one response only',
+    });
+    await beforeRestart.checkpoint({
+      sessionId: IDENTITY.sessionId,
+      checkpointId: 'checkpoint-before-kill',
+      cursor: 'cursor-before-kill',
+      summary: 'restart-safe state',
+      compactionEpoch: 1,
+    });
+    await beforeRestart.handoff({
+      sessionId: IDENTITY.sessionId,
+      handoffId: 'handoff-before-kill',
+      destination: 'mos',
+      correlationId: 'correlation-before-kill',
+      checkpointId: 'checkpoint-before-kill',
+      status: 'pending',
+    });
+    await beforeRestart.checkpoint({
+      sessionId: IDENTITY.sessionId,
+      checkpointId: 'checkpoint-after-handoff',
+      cursor: 'cursor-after-handoff',
+      summary: 'newer state cannot strand the portable handoff',
+      compactionEpoch: 2,
+    });
+
+    await handle.close();
+    handle = createPgliteDb(dataDir!);
+
+    const afterRestart = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    const recovered = await afterRestart.recover(IDENTITY.sessionId);
+    const resumedHandoff = await afterRestart.resumeHandoff('handoff-before-kill');
+    const handled: string[] = [];
+    const effects: string[] = [];
+
+    await afterRestart.drainInbox(IDENTITY.sessionId, async (entry): Promise<void> => {
+      handled.push(entry.idempotencyKey);
+    });
+    await afterRestart.dispatchOutbox(IDENTITY.sessionId, async (entry): Promise<void> => {
+      effects.push(entry.idempotencyKey);
+    });
+    await afterRestart.drainInbox(IDENTITY.sessionId, async (entry): Promise<void> => {
+      handled.push(entry.idempotencyKey);
+    });
+    await afterRestart.dispatchOutbox(IDENTITY.sessionId, async (entry): Promise<void> => {
+      effects.push(entry.idempotencyKey);
+    });
+
+    expect(recovered.identity).toEqual(IDENTITY);
+    expect(recovered.checkpoint).toMatchObject({ checkpointId: 'checkpoint-after-handoff' });
+    expect(recovered.handoffs).toMatchObject([{ handoffId: 'handoff-before-kill' }]);
+    expect(resumedHandoff.checkpoint).toMatchObject({ checkpointId: 'checkpoint-before-kill' });
+    expect(handled).toEqual(['inbox-before-kill']);
+    expect(effects).toEqual(['outbox-before-kill']);
+  }, 30_000);
+
+  it('redacts sensitive durable payloads before persistence', async () => {
+    const coordinator = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    await coordinator.create(IDENTITY);
+    await coordinator.receive({
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'redacted-inbox',
+      correlationId: 'correlation-redaction',
+      content: 'api_key=super-secret-canary',
+    });
+    await coordinator.enqueueOutbox({
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'redacted-outbox',
+      correlationId: 'correlation-redaction',
+      channelId: 'cli',
+      kind: 'provider.send',
+      content: 'email operator@example.test api_key=super-secret-canary',
+    });
+    await coordinator.checkpoint({
+      sessionId: IDENTITY.sessionId,
+      checkpointId: 'redacted-checkpoint',
+      cursor: 'bearer super-secret-canary',
+      summary: 'email operator@example.test',
+      compactionEpoch: 0,
+    });
+
+    const snapshot = await coordinator.snapshot(IDENTITY.sessionId);
+    const [persisted] = await handle.db
+      .select({ content: interactionInbox.content })
+      .from(interactionInbox)
+      .where(eq(interactionInbox.idempotencyKey, 'redacted-inbox'));
+
+    expect(JSON.stringify(snapshot)).not.toContain('super-secret-canary');
+    expect(JSON.stringify(snapshot)).not.toContain('operator@example.test');
+    expect(persisted?.content).not.toContain('super-secret-canary');
+    expect(persisted?.content).not.toContain('[REDACTED]');
+  }, 30_000);
+
+  it('fails closed when the configured idempotency secret is unavailable', async () => {
+    const coordinator = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    await coordinator.create(IDENTITY);
+    const secret = process.env['BETTER_AUTH_SECRET'];
+    delete process.env['BETTER_AUTH_SECRET'];
+    try {
+      await expect(
+        coordinator.receive({
+          sessionId: IDENTITY.sessionId,
+          idempotencyKey: 'requires-idempotency-secret',
+          correlationId: 'correlation-secret',
+          content: 'sensitive payload',
+        }),
+      ).rejects.toThrow(/required for durable idempotency digests/);
+    } finally {
+      if (secret === undefined) delete process.env['BETTER_AUTH_SECRET'];
+      else process.env['BETTER_AUTH_SECRET'] = secret;
+    }
+  }, 30_000);
+
+  it('uses keyed pre-redaction digests to reject distinct sensitive checkpoint payloads', async () => {
+    const coordinator = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    await coordinator.create(IDENTITY);
+    const input = {
+      sessionId: IDENTITY.sessionId,
+      checkpointId: 'checkpoint-secret-conflict',
+      cursor: 'api_key=secret-one',
+      summary: 'bearer secret-one',
+      compactionEpoch: 1,
+    };
+    await coordinator.checkpoint(input);
+
+    await expect(
+      coordinator.checkpoint({
+        ...input,
+        cursor: 'api_key=secret-two',
+        summary: 'bearer secret-two',
+      }),
+    ).rejects.toThrow(/checkpoint identity conflict/);
+
+    const [persisted] = await handle.db
+      .select({
+        digest: interactionCheckpoints.contentDigest,
+        cursor: interactionCheckpoints.cursor,
+      })
+      .from(interactionCheckpoints)
+      .where(eq(interactionCheckpoints.checkpointId, input.checkpointId));
+    expect(persisted?.cursor).not.toContain('secret-one');
+    expect(persisted?.digest).not.toBe(
+      createHash('sha256')
+        .update(JSON.stringify([input.cursor, input.summary]))
+        .digest('hex'),
+    );
+
+    await coordinator.checkpoint({
+      ...input,
+      checkpointId: 'checkpoint-delimiter-conflict',
+      cursor: 'a\u0000b',
+      summary: 'c',
+    });
+    await expect(
+      coordinator.checkpoint({
+        ...input,
+        checkpointId: 'checkpoint-delimiter-conflict',
+        cursor: 'a',
+        summary: 'b\u0000c',
+      }),
+    ).rejects.toThrow(/checkpoint identity conflict/);
+  }, 30_000);
+
+  it('rejects distinct sensitive inbox and outbox payloads under reused idempotency keys', async () => {
+    const coordinator = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    await coordinator.create(IDENTITY);
+    const inbox = {
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'inbox-secret-conflict',
+      correlationId: 'correlation-inbox-secret',
+      content: 'api_key=secret-one',
+    };
+    const outbox = {
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'outbox-secret-conflict',
+      correlationId: 'correlation-outbox-secret',
+      channelId: 'cli',
+      kind: 'provider.send',
+      content: 'api_key=secret-one',
+    };
+    await coordinator.receive(inbox);
+    await coordinator.enqueueOutbox(outbox);
+
+    await expect(coordinator.receive({ ...inbox, content: 'api_key=secret-two' })).rejects.toThrow(
+      /idempotency conflict/,
+    );
+    await expect(
+      coordinator.enqueueOutbox({ ...outbox, content: 'api_key=secret-two' }),
+    ).rejects.toThrow(/idempotency conflict/);
+  }, 30_000);
+
+  it('rejects database inbox and outbox idempotency-key conflicts', async () => {
+    const coordinator = new DurableSessionCoordinator(new DurableSessionRepository(handle.db));
+    await coordinator.create(IDENTITY);
+    await coordinator.receive({
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'inbox-conflict',
+      correlationId: 'correlation-inbox',
+      content: 'original inbox',
+    });
+    await coordinator.enqueueOutbox({
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'outbox-conflict',
+      correlationId: 'correlation-outbox',
+      channelId: 'cli',
+      kind: 'provider.send',
+      content: 'original outbox',
+    });
+
+    await expect(
+      coordinator.receive({
+        sessionId: IDENTITY.sessionId,
+        idempotencyKey: 'inbox-conflict',
+        correlationId: 'forged-correlation',
+        content: 'original inbox',
+      }),
+    ).rejects.toThrow(/idempotency conflict/);
+    await expect(
+      coordinator.enqueueOutbox({
+        sessionId: IDENTITY.sessionId,
+        idempotencyKey: 'outbox-conflict',
+        correlationId: 'correlation-outbox',
+        channelId: 'forged-channel',
+        kind: 'provider.send',
+        content: 'original outbox',
+      }),
+    ).rejects.toThrow(/idempotency conflict/);
+  }, 30_000);
+
+  it('does not requeue a live outbox claim during a normal scoped dispatch', async () => {
+    const repository = new DurableSessionRepository(handle.db);
+    const coordinator = new DurableSessionCoordinator(repository);
+    const runtimeProviders = { sendMessage: vi.fn().mockResolvedValue(undefined) };
+    const service = new DurableSessionService(repository, runtimeProviders as never);
+    const input = {
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'live-effect',
+      correlationId: 'correlation-live',
+      content: 'must not duplicate',
+      context: {
+        actorScope: { userId: IDENTITY.ownerId, tenantId: IDENTITY.tenantId },
+        channelId: 'cli',
+        correlationId: 'correlation-live',
+      },
+    };
+
+    await coordinator.create(IDENTITY);
+    await service.queueProviderSend(input);
+    expect(await repository.claimOutbox(IDENTITY.sessionId)).toMatchObject({
+      status: 'processing',
+    });
+
+    await service.dispatchProviderOutbox(IDENTITY.sessionId, input);
+    await expect(
+      service.recoverProviderSession(IDENTITY.sessionId, {
+        ...input,
+        context: {
+          ...input.context,
+          actorScope: { userId: 'intruder', tenantId: 'tenant-pglite' },
+        },
+      }),
+    ).rejects.toThrow(/scope or correlation mismatch/);
+
+    expect(runtimeProviders.sendMessage).not.toHaveBeenCalled();
+    expect(await coordinator.snapshot(IDENTITY.sessionId)).toMatchObject({
+      outbox: [{ idempotencyKey: 'live-effect', status: 'processing' }],
+    });
+  }, 30_000);
+
+  it('rejects an outbox correlation mismatch before claiming the pending effect', async () => {
+    const repository = new DurableSessionRepository(handle.db);
+    const coordinator = new DurableSessionCoordinator(repository);
+    const runtimeProviders = { sendMessage: vi.fn().mockResolvedValue(undefined) };
+    const service = new DurableSessionService(repository, runtimeProviders as never);
+    const input = {
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'mismatch-effect',
+      correlationId: 'correlation-expected',
+      content: 'must remain pending',
+      context: {
+        actorScope: { userId: IDENTITY.ownerId, tenantId: IDENTITY.tenantId },
+        channelId: 'cli',
+        correlationId: 'correlation-expected',
+      },
+    };
+
+    await coordinator.create(IDENTITY);
+    await service.queueProviderSend(input);
+    await expect(
+      service.dispatchProviderOutbox(IDENTITY.sessionId, {
+        ...input,
+        correlationId: 'correlation-forged',
+        context: { ...input.context, correlationId: 'correlation-forged' },
+      }),
+    ).rejects.toThrow(/scope or correlation mismatch/);
+
+    expect(runtimeProviders.sendMessage).not.toHaveBeenCalled();
+    expect(await coordinator.snapshot(IDENTITY.sessionId)).toMatchObject({
+      outbox: [{ idempotencyKey: 'mismatch-effect', status: 'pending' }],
+    });
+  }, 30_000);
+
+  it('dispatches only the outbox record bound to the supplied correlation and channel', async () => {
+    const repository = new DurableSessionRepository(handle.db);
+    const coordinator = new DurableSessionCoordinator(repository);
+    const runtimeProviders = { sendMessage: vi.fn().mockResolvedValue(undefined) };
+    const service = new DurableSessionService(repository, runtimeProviders as never);
+    const first = {
+      sessionId: IDENTITY.sessionId,
+      idempotencyKey: 'scoped-effect-one',
+      correlationId: 'correlation-one',
+      content: 'first result',
+      context: {
+        actorScope: { userId: IDENTITY.ownerId, tenantId: IDENTITY.tenantId },
+        channelId: 'cli',
+        correlationId: 'correlation-one',
+      },
+    };
+    const second = {
+      ...first,
+      idempotencyKey: 'scoped-effect-two',
+      correlationId: 'correlation-two',
+      content: 'second result',
+      context: { ...first.context, correlationId: 'correlation-two' },
+    };
+
+    await coordinator.create(IDENTITY);
+    await service.queueProviderSend(first);
+    await service.queueProviderSend(second);
+    await service.dispatchProviderOutbox(IDENTITY.sessionId, first);
+
+    expect(runtimeProviders.sendMessage).toHaveBeenCalledTimes(1);
+    expect(runtimeProviders.sendMessage).toHaveBeenCalledWith(
+      IDENTITY.providerId,
+      IDENTITY.runtimeSessionId,
+      { content: 'first result', idempotencyKey: 'scoped-effect-one' },
+      first.context,
+    );
+    expect(await coordinator.snapshot(IDENTITY.sessionId)).toMatchObject({
+      outbox: [
+        { idempotencyKey: 'scoped-effect-one', status: 'delivered' },
+        { idempotencyKey: 'scoped-effect-two', status: 'pending' },
+      ],
+    });
+  }, 30_000);
+});
+
+async function seedOwner(handle: DbHandle): Promise<void> {
+  await handle.db.execute(sql`
+    INSERT INTO users (id, name, email, email_verified, created_at, updated_at)
+    VALUES ('tess-owner', 'Tess Owner', 'tess-owner@example.test', false, now(), now())
+  `);
+}

apps/gateway/src/agent/durable-session.repository.ts

@@ -0,0 +1,529 @@
+import { createHash, createHmac } from 'node:crypto';
+import { Inject, Injectable } from '@nestjs/common';
+import {
+  and,
+  asc,
+  desc,
+  eq,
+  interactionCheckpoints,
+  interactionHandoffs,
+  interactionInbox,
+  interactionOutbox,
+  interactionSessions,
+  type Db,
+} from '@mosaicstack/db';
+import { seal, unseal } from '@mosaicstack/auth';
+import { redactSensitiveContent } from '@mosaicstack/log';
+import type {
+  DurableCheckpoint,
+  DurableCheckpointInput,
+  DurableEnqueueResult,
+  DurableHandoff,
+  DurableHandoffInput,
+  DurableInboxEntry,
+  DurableInboxInput,
+  DurableInboxStatus,
+  DurableOutboxEntry,
+  DurableOutboxInput,
+  DurableOutboxStatus,
+  DurableSessionIdentity,
+  DurableSessionSnapshot,
+  DurableSessionStore,
+} from '@mosaicstack/agent';
+import { DB } from '../database/database.module.js';
+
+@Injectable()
+export class DurableSessionRepository implements DurableSessionStore {
+  constructor(@Inject(DB) private readonly db: Db) {}
+
+  async create(identity: DurableSessionIdentity): Promise<void> {
+    await this.db
+      .insert(interactionSessions)
+      .values({
+        id: identity.sessionId,
+        agentName: identity.agentName,
+        tenantId: identity.tenantId,
+        ownerId: identity.ownerId,
+        providerId: identity.providerId,
+        runtimeSessionId: identity.runtimeSessionId,
+      })
+      .onConflictDoNothing();
+
+    const existing = await this.session(identity.sessionId);
+    if (!existing || !sameEnrollmentScope(existing, identity)) {
+      throw new Error(`Durable session identity conflict: ${identity.sessionId}`);
+    }
+    // A recovered/re-enrolled runtime can receive a new provider session ID;
+    // the conversation handle and owner scope remain immutable.
+    if (
+      existing.providerId !== identity.providerId ||
+      existing.runtimeSessionId !== identity.runtimeSessionId
+    ) {
+      await this.db
+        .update(interactionSessions)
+        .set({ providerId: identity.providerId, runtimeSessionId: identity.runtimeSessionId })
+        .where(eq(interactionSessions.id, identity.sessionId));
+    }
+  }
+
+  async snapshot(sessionId: string): Promise<DurableSessionSnapshot | null> {
+    const identity = await this.session(sessionId);
+    if (!identity) return null;
+
+    const [inbox, outbox, checkpoints, handoffs] = await Promise.all([
+      this.db
+        .select()
+        .from(interactionInbox)
+        .where(eq(interactionInbox.sessionId, sessionId))
+        .orderBy(asc(interactionInbox.createdAt)),
+      this.db
+        .select()
+        .from(interactionOutbox)
+        .where(eq(interactionOutbox.sessionId, sessionId))
+        .orderBy(asc(interactionOutbox.createdAt)),
+      this.db
+        .select()
+        .from(interactionCheckpoints)
+        .where(eq(interactionCheckpoints.sessionId, sessionId))
+        .orderBy(
+          desc(interactionCheckpoints.compactionEpoch),
+          desc(interactionCheckpoints.createdAt),
+        )
+        .limit(1),
+      this.db
+        .select()
+        .from(interactionHandoffs)
+        .where(eq(interactionHandoffs.sessionId, sessionId))
+        .orderBy(asc(interactionHandoffs.createdAt)),
+    ]);
+
+    const checkpoint = checkpoints[0];
+    return {
+      identity,
+      inbox: inbox.map(toInbox),
+      outbox: outbox.map(toOutbox),
+      ...(checkpoint ? { checkpoint: toCheckpoint(checkpoint) } : {}),
+      handoffs: handoffs.map(toHandoff),
+    };
+  }
+
+  async enqueueInbox(input: DurableInboxInput): Promise<DurableEnqueueResult<DurableInboxStatus>> {
+    const digest = contentDigest(input.content);
+    const record: DurableInboxInput = {
+      ...input,
+      content: redactSensitiveContent(input.content).content,
+    };
+    const inserted = await this.db
+      .insert(interactionInbox)
+      .values({
+        ...record,
+        content: seal(record.content),
+        contentDigest: digest,
+        status: 'pending',
+      })
+      .onConflictDoNothing()
+      .returning({ status: interactionInbox.status });
+    if (inserted[0]) return { accepted: true, status: inserted[0].status };
+
+    const existing = await this.db
+      .select()
+      .from(interactionInbox)
+      .where(
+        and(
+          eq(interactionInbox.sessionId, input.sessionId),
+          eq(interactionInbox.idempotencyKey, input.idempotencyKey),
+        ),
+      )
+      .limit(1);
+    if (!existing[0]) throw new Error(`Durable inbox enqueue failed: ${input.idempotencyKey}`);
+    const entry = toInbox(existing[0]);
+    if (
+      !sameInbox(entry, record) ||
+      !matchesContentDigest(existing[0].contentDigest, input.content)
+    ) {
+      throw new Error(`Durable inbox idempotency conflict: ${input.idempotencyKey}`);
+    }
+    return { accepted: false, status: entry.status };
+  }
+
+  async claimInbox(sessionId: string): Promise<DurableInboxEntry | null> {
+    for (let attempt = 0; attempt < 3; attempt += 1) {
+      const candidate = await this.db
+        .select()
+        .from(interactionInbox)
+        .where(
+          and(eq(interactionInbox.sessionId, sessionId), eq(interactionInbox.status, 'pending')),
+        )
+        .orderBy(asc(interactionInbox.createdAt))
+        .limit(1);
+      const entry = candidate[0];
+      if (!entry) return null;
+      const claimed = await this.db
+        .update(interactionInbox)
+        .set({ status: 'processing', updatedAt: new Date() })
+        .where(and(eq(interactionInbox.id, entry.id), eq(interactionInbox.status, 'pending')))
+        .returning();
+      if (claimed[0]) return toInbox(claimed[0]);
+    }
+    return null;
+  }
+
+  async completeInbox(sessionId: string, idempotencyKey: string): Promise<void> {
+    await this.db
+      .update(interactionInbox)
+      .set({ status: 'processed', updatedAt: new Date() })
+      .where(
+        and(
+          eq(interactionInbox.sessionId, sessionId),
+          eq(interactionInbox.idempotencyKey, idempotencyKey),
+          eq(interactionInbox.status, 'processing'),
+        ),
+      );
+  }
+
+  async releaseInbox(sessionId: string, idempotencyKey: string): Promise<void> {
+    await this.db
+      .update(interactionInbox)
+      .set({ status: 'pending', updatedAt: new Date() })
+      .where(
+        and(
+          eq(interactionInbox.sessionId, sessionId),
+          eq(interactionInbox.idempotencyKey, idempotencyKey),
+          eq(interactionInbox.status, 'processing'),
+        ),
+      );
+  }
+
+  async enqueueOutbox(
+    input: DurableOutboxInput,
+  ): Promise<DurableEnqueueResult<DurableOutboxStatus>> {
+    const digest = contentDigest(input.content);
+    const record: DurableOutboxInput = {
+      ...input,
+      content: redactSensitiveContent(input.content).content,
+    };
+    const inserted = await this.db
+      .insert(interactionOutbox)
+      .values({
+        ...record,
+        content: seal(record.content),
+        contentDigest: digest,
+        status: 'pending',
+      })
+      .onConflictDoNothing()
+      .returning({ status: interactionOutbox.status });
+    if (inserted[0]) return { accepted: true, status: inserted[0].status };
+
+    const existing = await this.db
+      .select()
+      .from(interactionOutbox)
+      .where(
+        and(
+          eq(interactionOutbox.sessionId, input.sessionId),
+          eq(interactionOutbox.idempotencyKey, input.idempotencyKey),
+        ),
+      )
+      .limit(1);
+    if (!existing[0]) throw new Error(`Durable outbox enqueue failed: ${input.idempotencyKey}`);
+    const entry = toOutbox(existing[0]);
+    if (
+      !sameOutbox(entry, record) ||
+      !matchesContentDigest(existing[0].contentDigest, input.content)
+    ) {
+      throw new Error(`Durable outbox idempotency conflict: ${input.idempotencyKey}`);
+    }
+    return { accepted: false, status: entry.status };
+  }
+
+  async claimOutbox(sessionId: string): Promise<DurableOutboxEntry | null> {
+    for (let attempt = 0; attempt < 3; attempt += 1) {
+      const candidate = await this.db
+        .select()
+        .from(interactionOutbox)
+        .where(
+          and(eq(interactionOutbox.sessionId, sessionId), eq(interactionOutbox.status, 'pending')),
+        )
+        .orderBy(asc(interactionOutbox.createdAt))
+        .limit(1);
+      const entry = candidate[0];
+      if (!entry) return null;
+      const claimed = await this.db
+        .update(interactionOutbox)
+        .set({ status: 'processing', updatedAt: new Date() })
+        .where(and(eq(interactionOutbox.id, entry.id), eq(interactionOutbox.status, 'pending')))
+        .returning();
+      if (claimed[0]) return toOutbox(claimed[0]);
+    }
+    return null;
+  }
+
+  async claimOutboxByKey(
+    sessionId: string,
+    idempotencyKey: string,
+  ): Promise<DurableOutboxEntry | null> {
+    const claimed = await this.db
+      .update(interactionOutbox)
+      .set({ status: 'processing', updatedAt: new Date() })
+      .where(
+        and(
+          eq(interactionOutbox.sessionId, sessionId),
+          eq(interactionOutbox.idempotencyKey, idempotencyKey),
+          eq(interactionOutbox.status, 'pending'),
+        ),
+      )
+      .returning();
+    return claimed[0] ? toOutbox(claimed[0]) : null;
+  }
+
+  async completeOutbox(sessionId: string, idempotencyKey: string): Promise<void> {
+    await this.db
+      .update(interactionOutbox)
+      .set({ status: 'delivered', updatedAt: new Date() })
+      .where(
+        and(
+          eq(interactionOutbox.sessionId, sessionId),
+          eq(interactionOutbox.idempotencyKey, idempotencyKey),
+          eq(interactionOutbox.status, 'processing'),
+        ),
+      );
+  }
+
+  async releaseOutbox(sessionId: string, idempotencyKey: string): Promise<void> {
+    await this.db
+      .update(interactionOutbox)
+      .set({ status: 'pending', updatedAt: new Date() })
+      .where(
+        and(
+          eq(interactionOutbox.sessionId, sessionId),
+          eq(interactionOutbox.idempotencyKey, idempotencyKey),
+          eq(interactionOutbox.status, 'processing'),
+        ),
+      );
+  }
+
+  async checkpoint(input: DurableCheckpointInput): Promise<void> {
+    // Compute identity before redaction.  The persisted digest is keyed so a database
+    // reader cannot use it as an offline oracle for sensitive cursor/summary values.
+    const digest = contentDigest(JSON.stringify([input.cursor, input.summary]));
+    const checkpoint: DurableCheckpointInput = {
+      ...input,
+      cursor: redactSensitiveContent(input.cursor).content,
+      summary: redactSensitiveContent(input.summary).content,
+    };
+    const inserted = await this.db
+      .insert(interactionCheckpoints)
+      .values({
+        ...checkpoint,
+        contentDigest: digest,
+        cursor: seal(checkpoint.cursor),
+        summary: seal(checkpoint.summary),
+      })
+      .onConflictDoNothing()
+      .returning({ checkpointId: interactionCheckpoints.checkpointId });
+    if (inserted[0]) return;
+
+    const existing = await this.db
+      .select()
+      .from(interactionCheckpoints)
+      .where(
+        and(
+          eq(interactionCheckpoints.sessionId, input.sessionId),
+          eq(interactionCheckpoints.checkpointId, input.checkpointId),
+        ),
+      )
+      .limit(1);
+    if (
+      !existing[0] ||
+      !sameCheckpoint(toCheckpoint(existing[0]), checkpoint) ||
+      !matchesCheckpointDigest(existing[0].contentDigest, digest)
+    ) {
+      throw new Error(`Durable checkpoint identity conflict: ${input.checkpointId}`);
+    }
+  }
+
+  async findCheckpoint(sessionId: string, checkpointId: string): Promise<DurableCheckpoint | null> {
+    const checkpoints = await this.db
+      .select()
+      .from(interactionCheckpoints)
+      .where(
+        and(
+          eq(interactionCheckpoints.sessionId, sessionId),
+          eq(interactionCheckpoints.checkpointId, checkpointId),
+        ),
+      )
+      .limit(1);
+    const checkpoint = checkpoints[0];
+    return checkpoint ? toCheckpoint(checkpoint) : null;
+  }
+
+  async handoff(input: DurableHandoffInput): Promise<void> {
+    const checkpoint = await this.findCheckpoint(input.sessionId, input.checkpointId);
+    if (!checkpoint) {
+      throw new Error(`Durable handoff checkpoint is unavailable: ${input.checkpointId}`);
+    }
+    const inserted = await this.db
+      .insert(interactionHandoffs)
+      .values({ ...input })
+      .onConflictDoNothing()
+      .returning({ handoffId: interactionHandoffs.handoffId });
+    if (inserted[0]) return;
+
+    const existing = await this.findHandoff(input.handoffId);
+    if (!existing || !sameHandoff(existing, input)) {
+      throw new Error(`Durable handoff identity conflict: ${input.handoffId}`);
+    }
+  }
+
+  async findHandoff(handoffId: string): Promise<DurableHandoff | null> {
+    const handoffs = await this.db
+      .select()
+      .from(interactionHandoffs)
+      .where(eq(interactionHandoffs.handoffId, handoffId))
+      .limit(1);
+    const handoff = handoffs[0];
+    return handoff ? toHandoff(handoff) : null;
+  }
+
+  async requeueInFlight(sessionId: string): Promise<void> {
+    // Inbox handlers are process-local work. A provider outbox claim may have
+    // reached an external target before a crash, so it is deliberately not
+    // replayed by generic recovery.
+    await this.db
+      .update(interactionInbox)
+      .set({ status: 'pending', updatedAt: new Date() })
+      .where(
+        and(eq(interactionInbox.sessionId, sessionId), eq(interactionInbox.status, 'processing')),
+      );
+  }
+
+  private async session(sessionId: string): Promise<DurableSessionIdentity | null> {
+    const sessions = await this.db
+      .select()
+      .from(interactionSessions)
+      .where(eq(interactionSessions.id, sessionId))
+      .limit(1);
+    const session = sessions[0];
+    return session
+      ? {
+          agentName: session.agentName,
+          sessionId: session.id,
+          tenantId: session.tenantId,
+          ownerId: session.ownerId,
+          providerId: session.providerId,
+          runtimeSessionId: session.runtimeSessionId,
+        }
+      : null;
+  }
+}
+
+function contentDigest(content: string): string {
+  const secret = process.env['BETTER_AUTH_SECRET'];
+  if (!secret) {
+    throw new Error('BETTER_AUTH_SECRET is required for durable idempotency digests');
+  }
+  return `hmac:v1:${createHmac('sha256', secret).update(content).digest('hex')}`;
+}
+
+function matchesContentDigest(stored: string, content: string): boolean {
+  return (
+    stored === contentDigest(content) ||
+    stored === createHash('sha256').update(content).digest('hex')
+  );
+}
+
+function matchesCheckpointDigest(stored: string, digest: string): boolean {
+  // Legacy rows predate any pre-redaction identity and cannot safely prove equality.
+  // Reject rather than let redaction collapse distinct sensitive checkpoint payloads.
+  return stored === digest;
+}
+
+function sameEnrollmentScope(left: DurableSessionIdentity, right: DurableSessionIdentity): boolean {
+  return (
+    left.agentName === right.agentName &&
+    left.sessionId === right.sessionId &&
+    left.tenantId === right.tenantId &&
+    left.ownerId === right.ownerId
+  );
+}
+
+function sameInbox(left: DurableInboxEntry, right: DurableInboxInput): boolean {
+  return (
+    left.sessionId === right.sessionId &&
+    left.idempotencyKey === right.idempotencyKey &&
+    left.correlationId === right.correlationId &&
+    left.content === right.content
+  );
+}
+
+function sameOutbox(left: DurableOutboxEntry, right: DurableOutboxInput): boolean {
+  return (
+    left.sessionId === right.sessionId &&
+    left.idempotencyKey === right.idempotencyKey &&
+    left.correlationId === right.correlationId &&
+    left.channelId === right.channelId &&
+    left.kind === right.kind &&
+    left.content === right.content
+  );
+}
+
+function sameCheckpoint(left: DurableCheckpoint, right: DurableCheckpointInput): boolean {
+  return (
+    left.sessionId === right.sessionId &&
+    left.checkpointId === right.checkpointId &&
+    left.compactionEpoch === right.compactionEpoch
+  );
+}
+
+function sameHandoff(left: DurableHandoff, right: DurableHandoffInput): boolean {
+  return (
+    left.sessionId === right.sessionId &&
+    left.handoffId === right.handoffId &&
+    left.destination === right.destination &&
+    left.correlationId === right.correlationId &&
+    left.checkpointId === right.checkpointId &&
+    left.status === right.status
+  );
+}
+
+function toInbox(row: typeof interactionInbox.$inferSelect): DurableInboxEntry {
+  return {
+    sessionId: row.sessionId,
+    idempotencyKey: row.idempotencyKey,
+    correlationId: row.correlationId,
+    content: unseal(row.content),
+    status: row.status,
+  };
+}
+
+function toOutbox(row: typeof interactionOutbox.$inferSelect): DurableOutboxEntry {
+  return {
+    sessionId: row.sessionId,
+    idempotencyKey: row.idempotencyKey,
+    correlationId: row.correlationId,
+    channelId: row.channelId,
+    kind: row.kind,
+    content: unseal(row.content),
+    status: row.status,
+  };
+}
+
+function toCheckpoint(row: typeof interactionCheckpoints.$inferSelect): DurableCheckpoint {
+  return {
+    sessionId: row.sessionId,
+    checkpointId: row.checkpointId,
+    cursor: unseal(row.cursor),
+    summary: unseal(row.summary),
+    compactionEpoch: row.compactionEpoch,
+  };
+}
+
+function toHandoff(row: typeof interactionHandoffs.$inferSelect): DurableHandoff {
+  return {
+    sessionId: row.sessionId,
+    handoffId: row.handoffId,
+    destination: row.destination,
+    correlationId: row.correlationId,
+    checkpointId: row.checkpointId,
+    status: row.status,
+  };
+}

apps/gateway/src/agent/durable-session.service.ts

@@ -0,0 +1,123 @@
+import { ForbiddenException, Inject, Injectable } from '@nestjs/common';
+import { DurableSessionCoordinator, type DurableSessionIdentity } from '@mosaicstack/agent';
+import type { ProviderOutboxDto } from './durable-session.dto.js';
+import { DurableSessionRepository } from './durable-session.repository.js';
+import {
+  RuntimeProviderService,
+  type RuntimeProviderRequestContext,
+} from './runtime-provider-registry.service.js';
+
+/**
+ * Scoped gateway boundary for the canonical durable session state machine. It deliberately
+ * uses composition: raw state methods cannot be injected into channel, CLI, or
+ * MCP adapters without a server-derived actor/tenant/correlation context.
+ */
+@Injectable()
+export class DurableSessionService {
+  private readonly coordinator: DurableSessionCoordinator;
+
+  constructor(
+    @Inject(DurableSessionRepository) repository: DurableSessionRepository,
+    @Inject(RuntimeProviderService) private readonly runtimeProviders: RuntimeProviderService,
+  ) {
+    this.coordinator = new DurableSessionCoordinator(repository);
+  }
+
+  /** Enroll a verified runtime session under the stable cross-surface conversation handle. */
+  async enroll(
+    identity: DurableSessionIdentity,
+    context: RuntimeProviderRequestContext,
+  ): Promise<void> {
+    if (
+      identity.ownerId !== context.actorScope.userId ||
+      identity.tenantId !== context.actorScope.tenantId
+    ) {
+      throw new ForbiddenException('Durable session enrollment scope mismatch');
+    }
+    await this.coordinator.create(identity);
+  }
+
+  async queueProviderSend(input: ProviderOutboxDto): Promise<void> {
+    const snapshot = await this.coordinator.snapshot(input.sessionId);
+    this.assertScope(snapshot.identity.ownerId, snapshot.identity.tenantId, input);
+    await this.coordinator.enqueueOutbox({
+      sessionId: input.sessionId,
+      idempotencyKey: input.idempotencyKey,
+      correlationId: input.correlationId,
+      channelId: input.context.channelId,
+      kind: 'provider.send',
+      content: input.content,
+    });
+  }
+
+  async dispatchProviderOutbox(sessionId: string, input: ProviderOutboxDto): Promise<void> {
+    if (sessionId !== input.sessionId) {
+      throw new ForbiddenException('Durable outbox session mismatch');
+    }
+    const snapshot = await this.coordinator.snapshot(sessionId);
+    this.assertScope(snapshot.identity.ownerId, snapshot.identity.tenantId, input);
+    const pendingEntry = snapshot.outbox.find(
+      (entry): boolean => entry.idempotencyKey === input.idempotencyKey,
+    );
+    if (!pendingEntry) return;
+    // Validate immutable routing before claiming. A caller with a mismatched
+    // correlation/channel must not strand a pending external side effect.
+    this.assertOutboxScope(pendingEntry, input);
+    await this.coordinator.dispatchOutboxEntry(
+      sessionId,
+      input.idempotencyKey,
+      async (entry): Promise<void> => {
+        this.assertOutboxScope(entry, input);
+        await this.runtimeProviders.sendMessage(
+          snapshot.identity.providerId,
+          snapshot.identity.runtimeSessionId,
+          { content: entry.content, idempotencyKey: entry.idempotencyKey },
+          input.context,
+        );
+      },
+    );
+  }
+
+  /** Read durable identity/state only after deriving and checking the server-side actor scope. */
+  async getSnapshot(sessionId: string, context: RuntimeProviderRequestContext) {
+    const snapshot = await this.coordinator.snapshot(sessionId);
+    this.assertScope(snapshot.identity.ownerId, snapshot.identity.tenantId, {
+      sessionId,
+      content: '',
+      idempotencyKey: 'read-only',
+      correlationId: context.correlationId,
+      context,
+    });
+    return snapshot;
+  }
+
+  /** Startup/recovery-only path; normal queue/dispatch methods never requeue live work. */
+  async recoverProviderSession(sessionId: string, input: ProviderOutboxDto): Promise<void> {
+    const snapshot = await this.coordinator.snapshot(sessionId);
+    this.assertScope(snapshot.identity.ownerId, snapshot.identity.tenantId, input);
+    await this.coordinator.recover(sessionId);
+  }
+
+  private assertOutboxScope(
+    entry: { kind: string; correlationId: string; channelId: string },
+    input: ProviderOutboxDto,
+  ): void {
+    if (
+      entry.kind !== 'provider.send' ||
+      entry.correlationId !== input.correlationId ||
+      entry.channelId !== input.context.channelId
+    ) {
+      throw new ForbiddenException('Durable outbox scope or correlation mismatch');
+    }
+  }
+
+  private assertScope(ownerId: string, tenantId: string, input: ProviderOutboxDto): void {
+    if (
+      input.context.actorScope.userId !== ownerId ||
+      input.context.actorScope.tenantId !== tenantId ||
+      input.context.correlationId !== input.correlationId
+    ) {
+      throw new ForbiddenException('Durable session scope or correlation mismatch');
+    }
+  }
+}

apps/gateway/src/agent/hermes-runtime-reachability.e2e.test.ts

@@ -0,0 +1,176 @@
+import 'reflect-metadata';
+import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest';
+import { Global, Module } from '@nestjs/common';
+import { Test } from '@nestjs/testing';
+import { FastifyAdapter, type NestFastifyApplication } from '@nestjs/platform-fastify';
+import { HermesRuntimeProvider } from '@mosaicstack/agent';
+import { AgentModule } from './agent.module.js';
+import { AUTH } from '../auth/auth.tokens.js';
+import { AuthGuard } from '../auth/auth.guard.js';
+import { BRAIN } from '../brain/brain.tokens.js';
+import { DB } from '../database/database.module.js';
+import { CoordModule } from '../coord/coord.module.js';
+import { McpClientModule } from '../mcp-client/mcp-client.module.js';
+import { SkillsModule } from '../skills/skills.module.js';
+import { GCModule } from '../gc/gc.module.js';
+import { LogModule } from '../log/log.module.js';
+import { CommandsModule } from '../commands/commands.module.js';
+import {
+  AGENT_RUNTIME_PROVIDER_REGISTRY,
+  RUNTIME_APPROVAL_VERIFIER,
+  RUNTIME_PROVIDER_AUDIT_SINK,
+  RuntimeProviderAuditService,
+} from './runtime-provider-registry.service.js';
+import { DurableSessionService } from './durable-session.service.js';
+import { DurableSessionRepository } from './durable-session.repository.js';
+import { AgentService } from './agent.service.js';
+import { ProviderService } from './provider.service.js';
+import { ProviderCredentialsService } from './provider-credentials.service.js';
+import { RoutingService } from './routing.service.js';
+import { RoutingEngineService } from './routing/routing-engine.service.js';
+import { SkillLoaderService } from './skill-loader.service.js';
+
+const authenticatedUser = { id: 'operator-1', tenantId: 'tenant-1' };
+
+@Module({})
+class EmptyAgentDependencyModule {}
+
+@Global()
+@Module({
+  providers: [
+    {
+      provide: AUTH,
+      useValue: {
+        api: {
+          getSession: vi.fn(async ({ headers }: { headers: Headers }) =>
+            headers.get('cookie') === 'session=trusted'
+              ? { user: authenticatedUser, session: { id: 'session-1' } }
+              : null,
+          ),
+        },
+      },
+    },
+    AuthGuard,
+    { provide: BRAIN, useValue: {} },
+    { provide: DB, useValue: {} },
+  ],
+  exports: [AUTH, AuthGuard, BRAIN, DB],
+})
+class AuthenticatedRequestModule {}
+
+/**
+ * This is deliberately an HTTP test rather than a controller unit test: it
+ * exercises AgentModule's actual provider factory, Nest DI, and AuthGuard.
+ */
+describe('Hermes runtime provider reachability', (): void => {
+  let app: NestFastifyApplication | undefined;
+
+  beforeAll(async (): Promise<void> => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const moduleRef = await Test.createTestingModule({
+      imports: [AuthenticatedRequestModule, AgentModule],
+    })
+      .overrideModule(CoordModule)
+      .useModule(EmptyAgentDependencyModule)
+      .overrideModule(McpClientModule)
+      .useModule(EmptyAgentDependencyModule)
+      .overrideModule(SkillsModule)
+      .useModule(EmptyAgentDependencyModule)
+      .overrideModule(GCModule)
+      .useModule(EmptyAgentDependencyModule)
+      .overrideModule(LogModule)
+      .useModule(EmptyAgentDependencyModule)
+      .overrideModule(CommandsModule)
+      .useModule(EmptyAgentDependencyModule)
+      .overrideProvider(RuntimeProviderAuditService)
+      .useValue({ record: vi.fn().mockResolvedValue(undefined) })
+      .overrideProvider(RUNTIME_PROVIDER_AUDIT_SINK)
+      .useValue({ record: vi.fn().mockResolvedValue(undefined) })
+      .overrideProvider(RUNTIME_APPROVAL_VERIFIER)
+      .useValue({ consume: vi.fn().mockResolvedValue(false) })
+      .overrideProvider(DurableSessionService)
+      .useValue({})
+      .overrideProvider(DurableSessionRepository)
+      .useValue({})
+      .overrideProvider(AgentService)
+      .useValue({})
+      .overrideProvider(ProviderService)
+      .useValue({})
+      .overrideProvider(ProviderCredentialsService)
+      .useValue({})
+      .overrideProvider(RoutingService)
+      .useValue({})
+      .overrideProvider(RoutingEngineService)
+      .useValue({})
+      .overrideProvider(SkillLoaderService)
+      .useValue({})
+      .compile();
+
+    app = moduleRef.createNestApplication<NestFastifyApplication>(new FastifyAdapter());
+    await app.init();
+    await app.getHttpAdapter().getInstance().ready();
+  });
+
+  afterAll(async (): Promise<void> => {
+    await app?.close();
+  });
+
+  it('returns gateway denial responses from the actual guarded interaction routes', async (): Promise<void> => {
+    if (!app) throw new Error('Nest application did not initialize');
+
+    const attachDenied = await app.inject({
+      method: 'POST',
+      url: '/api/interaction/Nova/sessions/session-1/attach',
+      headers: { 'x-correlation-id': 'correlation-1' },
+      payload: { mode: 'read' },
+    });
+    expect(attachDenied.statusCode).toBe(401);
+
+    const sendDenied = await app.inject({
+      method: 'POST',
+      url: '/api/interaction/Nova/sessions/session-1/send',
+      headers: { cookie: 'session=trusted', 'x-correlation-id': 'correlation-1' },
+      payload: {},
+    });
+    expect(sendDenied.statusCode).toBe(403);
+    expect(sendDenied.json()).toMatchObject({
+      message: 'Content and idempotency key are required',
+    });
+
+    const stopDenied = await app.inject({
+      method: 'POST',
+      url: '/api/interaction/Nova/sessions/session-1/stop',
+      headers: { cookie: 'session=trusted', 'x-correlation-id': 'correlation-1' },
+      payload: {},
+    });
+    expect(stopDenied.statusCode).toBe(403);
+    expect(stopDenied.json()).toMatchObject({ message: 'Exact-action approval is required' });
+  });
+
+  it('requires authentication and reaches the Hermes provider registered by AgentModule', async (): Promise<void> => {
+    if (!app) throw new Error('Nest application did not initialize');
+    const registry = app.get(AGENT_RUNTIME_PROVIDER_REGISTRY);
+    expect(registry.get('runtime.hermes')).toBeInstanceOf(HermesRuntimeProvider);
+
+    const denied = await app.inject({
+      method: 'GET',
+      url: '/api/interaction/Nova/transitional-capabilities?provider=runtime.hermes',
+      headers: { 'x-correlation-id': 'correlation-1' },
+    });
+    expect(denied.statusCode).toBe(401);
+
+    const response = await app.inject({
+      method: 'GET',
+      url: '/api/interaction/Nova/transitional-capabilities?provider=runtime.hermes',
+      headers: { cookie: 'session=trusted', 'x-correlation-id': 'correlation-1' },
+    });
+    expect(response.statusCode).toBe(200);
+    expect(response.json()).toEqual([
+      { capability: 'kanban', status: 'unsupported' },
+      { capability: 'skills', status: 'unsupported' },
+      { capability: 'memory', status: 'unsupported' },
+      { capability: 'tools', status: 'unsupported' },
+      { capability: 'cron', status: 'unsupported' },
+    ]);
+  });
+});

apps/gateway/src/agent/hermes-runtime.transport.test.ts

@@ -0,0 +1,46 @@
+import { describe, expect, it, vi } from 'vitest';
+import { GatewayHermesRuntimeTransport } from './hermes-runtime.transport.js';
+
+const scope = {
+  actorId: 'owner-1',
+  tenantId: 'tenant-1',
+  channelId: 'cli',
+  correlationId: 'correlation-1',
+};
+
+describe('GatewayHermesRuntimeTransport', () => {
+  it('preserves a configured path prefix and authenticates the concrete runtime request', async () => {
+    const fetchFn = vi
+      .fn()
+      .mockResolvedValue(new Response(JSON.stringify(['session.list']), { status: 200 }));
+    const transport = new GatewayHermesRuntimeTransport(
+      'https://runtime.example.test/hermes',
+      'test-service-token',
+      fetchFn,
+    );
+
+    await expect(transport.capabilities(scope)).resolves.toEqual(['session.list']);
+
+    expect(fetchFn).toHaveBeenCalledWith(
+      new URL('https://runtime.example.test/hermes/capabilities'),
+      expect.objectContaining({
+        headers: expect.objectContaining({
+          authorization: 'Bearer test-service-token',
+          'x-mosaic-channel-id': 'cli',
+        }),
+      }),
+    );
+  });
+
+  it('rejects non-loopback HTTP runtime endpoints before sending identity headers', async () => {
+    const fetchFn = vi.fn();
+    const transport = new GatewayHermesRuntimeTransport(
+      'http://runtime.example.test/hermes',
+      'test-service-token',
+      fetchFn,
+    );
+
+    await expect(transport.capabilities(scope)).rejects.toThrow('requires HTTPS');
+    expect(fetchFn).not.toHaveBeenCalled();
+  });
+});

apps/gateway/src/agent/hermes-runtime.transport.ts

@@ -0,0 +1,121 @@
+import type { HermesLegacySession, HermesRuntimeTransport } from '@mosaicstack/agent';
+import type {
+  RuntimeAttachHandle,
+  RuntimeAttachMode,
+  RuntimeMessage,
+  RuntimeScope,
+  RuntimeStreamEvent,
+} from '@mosaicstack/types';
+
+/** Concrete HTTP transport for a configured legacy Hermes runtime endpoint. */
+export class GatewayHermesRuntimeTransport implements HermesRuntimeTransport {
+  constructor(
+    private readonly baseUrl = process.env['MOSAIC_HERMES_RUNTIME_URL']?.trim(),
+    private readonly serviceToken = process.env['MOSAIC_HERMES_RUNTIME_TOKEN']?.trim(),
+    private readonly fetchFn: typeof fetch = fetch,
+  ) {}
+
+  async capabilities(scope: RuntimeScope): Promise<string[]> {
+    return this.request<string[]>('/capabilities', scope);
+  }
+
+  async health(scope: RuntimeScope): Promise<{ status: string; detail?: string }> {
+    return this.request<{ status: string; detail?: string }>('/health', scope);
+  }
+
+  async sessions(scope: RuntimeScope): Promise<HermesLegacySession[]> {
+    return this.request<HermesLegacySession[]>('/sessions', scope);
+  }
+
+  async *stream(
+    sessionId: string,
+    cursor: string | undefined,
+    scope: RuntimeScope,
+  ): AsyncIterable<RuntimeStreamEvent> {
+    const params = new URLSearchParams(cursor ? { cursor } : {});
+    const events = await this.request<RuntimeStreamEvent[]>(
+      `/sessions/${encodeURIComponent(sessionId)}/stream?${params.toString()}`,
+      scope,
+    );
+    yield* events;
+  }
+
+  async send(sessionId: string, message: RuntimeMessage, scope: RuntimeScope): Promise<void> {
+    await this.request(`/sessions/${encodeURIComponent(sessionId)}/messages`, scope, {
+      method: 'POST',
+      body: message,
+    });
+  }
+
+  async attach(
+    sessionId: string,
+    mode: RuntimeAttachMode,
+    scope: RuntimeScope,
+  ): Promise<RuntimeAttachHandle> {
+    return this.request<RuntimeAttachHandle>(
+      `/sessions/${encodeURIComponent(sessionId)}/attach`,
+      scope,
+      {
+        method: 'POST',
+        body: { mode },
+      },
+    );
+  }
+
+  async detach(attachmentId: string, scope: RuntimeScope): Promise<void> {
+    await this.request(`/attachments/${encodeURIComponent(attachmentId)}`, scope, {
+      method: 'DELETE',
+    });
+  }
+
+  async terminate(sessionId: string, approvalRef: string, scope: RuntimeScope): Promise<void> {
+    await this.request(`/sessions/${encodeURIComponent(sessionId)}/terminate`, scope, {
+      method: 'POST',
+      body: { approvalRef },
+    });
+  }
+
+  private async request<T>(
+    path: string,
+    scope: RuntimeScope,
+    init: { method?: string; body?: unknown } = {},
+  ): Promise<T> {
+    if (!this.baseUrl || !this.serviceToken) {
+      throw new Error(
+        'MOSAIC_HERMES_RUNTIME_URL and MOSAIC_HERMES_RUNTIME_TOKEN must configure Hermes transport',
+      );
+    }
+    const endpoint = new URL(this.baseUrl);
+    if (endpoint.protocol !== 'https:' && !isLoopbackHttp(endpoint)) {
+      throw new Error('Hermes runtime transport requires HTTPS outside loopback');
+    }
+    const response = await this.fetchFn(
+      new URL(path.replace(/^\//, ''), `${endpoint.toString().replace(/\/$/, '')}/`),
+      {
+        method: init.method ?? 'GET',
+        headers: {
+          accept: 'application/json',
+          authorization: `Bearer ${this.serviceToken}`,
+          'x-mosaic-actor-id': scope.actorId,
+          'x-mosaic-tenant-id': scope.tenantId,
+          'x-mosaic-channel-id': scope.channelId,
+          'x-correlation-id': scope.correlationId,
+          ...(init.body ? { 'content-type': 'application/json' } : {}),
+        },
+        ...(init.body ? { body: JSON.stringify(init.body) } : {}),
+      },
+    );
+    if (!response.ok) throw new Error(`Hermes runtime request failed: ${response.status}`);
+    if (response.status === 204) return undefined as T;
+    return (await response.json()) as T;
+  }
+}
+
+function isLoopbackHttp(endpoint: URL): boolean {
+  return (
+    endpoint.protocol === 'http:' &&
+    (endpoint.hostname === 'localhost' ||
+      endpoint.hostname === '127.0.0.1' ||
+      endpoint.hostname === '::1')
+  );
+}

apps/gateway/src/agent/interaction.controller.test.ts

@@ -0,0 +1,263 @@
+import { createGatewayRuntimeProviderRegistry } from './agent.module.js';
+import { firstValueFrom } from 'rxjs';
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import {
+  RuntimeApprovalDeniedError,
+  RuntimeProviderService,
+} from './runtime-provider-registry.service.js';
+import { RuntimeApprovalDeniedFilter } from './runtime-approval-denied.filter.js';
+import { InteractionController } from './interaction.controller.js';
+
+describe('InteractionController', (): void => {
+  afterEach(() => vi.restoreAllMocks());
+
+  it('maps a denied runtime approval to Fastify HTTP 403', () => {
+    const send = vi.fn();
+    const status = vi.fn().mockReturnValue({ send });
+    const response = { status };
+    const host = { switchToHttp: () => ({ getResponse: () => response }) };
+
+    new RuntimeApprovalDeniedFilter().catch(new RuntimeApprovalDeniedError(), host as never);
+
+    expect(status).toHaveBeenCalledWith(403);
+    expect(send).toHaveBeenCalledWith({
+      statusCode: 403,
+      message: 'Runtime termination approval denied',
+    });
+  });
+
+  it('honors a differently named configured instance without a code change', async () => {
+    const prior = process.env['MOSAIC_AGENT_NAME'];
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const runtime = { listSessions: vi.fn().mockResolvedValue([]) };
+    const controller = new InteractionController(runtime as never, {} as never);
+
+    await expect(
+      controller.sessions('Nova', 'fleet', { id: 'owner', tenantId: 'team' }, 'corr-1'),
+    ).resolves.toEqual([]);
+    await expect(
+      controller.sessions('Other', 'fleet', { id: 'owner', tenantId: 'team' }, 'corr-1'),
+    ).rejects.toThrow('Interaction agent is not configured');
+
+    if (prior === undefined) delete process.env['MOSAIC_AGENT_NAME'];
+    else process.env['MOSAIC_AGENT_NAME'] = prior;
+  });
+
+  it('reaches the registered Hermes provider through the authenticated transitional matrix route', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const registry = createGatewayRuntimeProviderRegistry();
+    const runtime = new RuntimeProviderService(
+      registry,
+      { record: vi.fn().mockResolvedValue(undefined) },
+      { consume: vi.fn().mockResolvedValue(false) },
+    );
+    const controller = new InteractionController(runtime, {} as never);
+
+    await expect(
+      controller.transitionalCapabilities(
+        'Nova',
+        'runtime.hermes',
+        { id: 'owner', tenantId: 'team' },
+        'corr-1',
+      ),
+    ).resolves.toEqual([
+      { capability: 'kanban', status: 'unsupported' },
+      { capability: 'skills', status: 'unsupported' },
+      { capability: 'memory', status: 'unsupported' },
+      { capability: 'tools', status: 'unsupported' },
+      { capability: 'cron', status: 'unsupported' },
+    ]);
+  });
+
+  it('rejects a request without the non-simple correlation header', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const controller = new InteractionController({ listSessions: vi.fn() } as never, {} as never);
+
+    await expect(controller.sessions('Nova', 'fleet', { id: 'owner' })).rejects.toThrow(
+      'X-Correlation-Id is required',
+    );
+  });
+
+  it('enrolls a visible runtime session under the cross-surface conversation handle', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const runtime = {
+      listSessions: vi.fn().mockResolvedValue([{ id: 'runtime-1' }]),
+    };
+    const durable = { enroll: vi.fn().mockResolvedValue(undefined) };
+    const controller = new InteractionController(runtime as never, durable as never);
+
+    await expect(
+      controller.enroll(
+        'Nova',
+        'conversation-1',
+        { providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+        { id: 'owner', tenantId: 'team' },
+        'corr-1',
+      ),
+    ).resolves.toEqual({ status: 'enrolled', sessionId: 'conversation-1' });
+    expect(durable.enroll).toHaveBeenCalledWith(
+      {
+        agentName: 'Nova',
+        sessionId: 'conversation-1',
+        tenantId: 'team',
+        ownerId: 'owner',
+        providerId: 'fleet',
+        runtimeSessionId: 'runtime-1',
+      },
+      expect.objectContaining({ correlationId: 'corr-1' }),
+    );
+  });
+
+  it('rejects an invalid attach mode before invoking a provider', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const controller = new InteractionController({ attach: vi.fn() } as never, {} as never);
+
+    await expect(
+      controller.attach('Nova', 'durable-1', { mode: 'write' as never }, { id: 'owner' }, 'corr-1'),
+    ).rejects.toThrow('Interaction attach mode is invalid');
+  });
+
+  it('resumes a durable session by attaching and streaming its runtime events', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const runtimeEvent = {
+      type: 'message.delta' as const,
+      sessionId: 'runtime-1',
+      cursor: 'cursor-1',
+      occurredAt: '2026-07-13T00:00:00.000Z',
+      content: 'resumed',
+    };
+    const runtime = {
+      attach: vi.fn().mockResolvedValue({ attachmentId: 'attach-1', sessionId: 'runtime-1' }),
+      streamSession: vi.fn(async function* () {
+        yield runtimeEvent;
+      }),
+    };
+    const durable = {
+      getSnapshot: vi.fn().mockResolvedValue({
+        identity: { agentName: 'Nova', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+      }),
+    };
+    const controller = new InteractionController(runtime as never, durable as never);
+
+    await controller.attach('Nova', 'conversation-1', { mode: 'read' }, { id: 'owner' }, 'corr-1');
+    await expect(
+      firstValueFrom(
+        controller.stream('Nova', 'conversation-1', undefined, { id: 'owner' }, 'corr-1'),
+      ),
+    ).resolves.toEqual({ data: runtimeEvent });
+
+    expect(runtime.attach).toHaveBeenCalledWith(
+      'fleet',
+      'runtime-1',
+      'read',
+      expect.objectContaining({ correlationId: 'corr-1' }),
+    );
+    expect(runtime.streamSession).toHaveBeenCalledWith(
+      'fleet',
+      'runtime-1',
+      undefined,
+      expect.objectContaining({ correlationId: 'corr-1' }),
+    );
+  });
+
+  it('does not create a runtime stream after the SSE subscriber disconnects during snapshot lookup', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    let resolveSnapshot!: (value: { identity: Record<string, string> }) => void;
+    const snapshot = new Promise<{ identity: Record<string, string> }>((resolve) => {
+      resolveSnapshot = resolve;
+    });
+    const runtime = { streamSession: vi.fn() };
+    const durable = { getSnapshot: vi.fn().mockReturnValue(snapshot) };
+    const controller = new InteractionController(runtime as never, durable as never);
+
+    const subscription = controller
+      .stream('Nova', 'conversation-1', undefined, { id: 'owner' }, 'corr-1')
+      .subscribe();
+    subscription.unsubscribe();
+    resolveSnapshot({
+      identity: { agentName: 'Nova', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+    });
+    await new Promise((resolve) => setTimeout(resolve, 0));
+
+    expect(runtime.streamSession).not.toHaveBeenCalled();
+  });
+
+  it.each([
+    ['wrong actor', { getSnapshot: vi.fn().mockRejectedValue(new Error('scope mismatch')) }],
+    [
+      'session-agent mismatch',
+      {
+        getSnapshot: vi.fn().mockResolvedValue({
+          identity: { agentName: 'Other', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+        }),
+      },
+    ],
+  ])('denies a CLI stop for %s', async (_reason, durable) => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const runtime = { terminate: vi.fn().mockResolvedValue(undefined) };
+    const controller = new InteractionController(runtime as never, durable as never);
+
+    await expect(
+      controller.stop(
+        'Nova',
+        'durable-1',
+        { approvalRef: 'approval-1' },
+        { id: 'owner' },
+        'corr-1',
+      ),
+    ).rejects.toBeDefined();
+    expect(runtime.terminate).not.toHaveBeenCalled();
+  });
+
+  it('surfaces a denied runtime approval to the CLI interaction surface', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const runtime = {
+      terminate: vi.fn().mockRejectedValue(new Error('Runtime termination approval denied')),
+    };
+    const durable = {
+      getSnapshot: vi.fn().mockResolvedValue({
+        identity: { agentName: 'Nova', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+      }),
+    };
+    const controller = new InteractionController(runtime as never, durable as never);
+
+    await expect(
+      controller.stop(
+        'Nova',
+        'durable-1',
+        { approvalRef: 'approval-1' },
+        { id: 'owner' },
+        'corr-1',
+      ),
+    ).rejects.toThrow('Runtime termination approval denied');
+  });
+
+  it('uses the durable session identity and runtime registry for an approved stop', async () => {
+    process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+    const runtime = { terminate: vi.fn().mockResolvedValue(undefined) };
+    const durable = {
+      getSnapshot: vi.fn().mockResolvedValue({
+        identity: { agentName: 'Nova', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+      }),
+    };
+    const controller = new InteractionController(runtime as never, durable as never);
+
+    await controller.stop(
+      'Nova',
+      'durable-1',
+      { approvalRef: 'approval-1' },
+      { id: 'owner' },
+      'corr-1',
+    );
+
+    expect(runtime.terminate).toHaveBeenCalledWith(
+      'fleet',
+      'runtime-1',
+      'approval-1',
+      expect.objectContaining({
+        correlationId: 'corr-1',
+        actorScope: { userId: 'owner', tenantId: 'owner' },
+      }),
+    );
+  });
+});

apps/gateway/src/agent/interaction.controller.ts

@@ -0,0 +1,293 @@
+import {
+  Body,
+  Controller,
+  ForbiddenException,
+  Get,
+  Headers,
+  Sse,
+  Inject,
+  Param,
+  Post,
+  Query,
+  UseGuards,
+  UseFilters,
+} from '@nestjs/common';
+import type { RuntimeAttachMode, RuntimeStreamEvent } from '@mosaicstack/types';
+import { Observable } from 'rxjs';
+import { AuthGuard } from '../auth/auth.guard.js';
+import { CurrentUser } from '../auth/current-user.decorator.js';
+import { scopeFromUser, type AuthenticatedUserLike } from '../auth/session-scope.js';
+import { DurableSessionService } from './durable-session.service.js';
+import { RuntimeApprovalDeniedFilter } from './runtime-approval-denied.filter.js';
+import {
+  RuntimeProviderService,
+  type RuntimeProviderRequestContext,
+} from './runtime-provider-registry.service.js';
+
+/**
+ * Authenticated HTTP boundary for operator interaction clients. Identity is
+ * selected from deployment configuration, never a client-side command name.
+ */
+@Controller('api/interaction/:agentName')
+@UseGuards(AuthGuard)
+@UseFilters(RuntimeApprovalDeniedFilter)
+export class InteractionController {
+  constructor(
+    @Inject(RuntimeProviderService) private readonly runtime: RuntimeProviderService,
+    @Inject(DurableSessionService) private readonly durable: DurableSessionService,
+  ) {}
+
+  @Get('sessions')
+  async sessions(
+    @Param('agentName') agentName: string,
+    @Query('provider') providerId: string,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    return this.runtime.listSessions(
+      this.requiredProvider(providerId),
+      this.context(user, correlationId),
+    );
+  }
+
+  @Get('transitional-capabilities')
+  async transitionalCapabilities(
+    @Param('agentName') agentName: string,
+    @Query('provider') providerId: string,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    return this.runtime.transitionalCapabilityMatrix(
+      this.requiredProvider(providerId),
+      this.context(user, correlationId),
+    );
+  }
+
+  @Get('tree')
+  async tree(
+    @Param('agentName') agentName: string,
+    @Query('provider') providerId: string,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    return this.runtime.getSessionTree(
+      this.requiredProvider(providerId),
+      this.context(user, correlationId),
+    );
+  }
+
+  /**
+   * Bind an existing, authorized runtime session to the stable conversation ID.
+   * This is the lifecycle boundary where both runtime identifiers are known.
+   */
+  @Post('sessions/:sessionId/enroll')
+  async enroll(
+    @Param('agentName') agentName: string,
+    @Param('sessionId') sessionId: string,
+    @Body() body: { providerId?: string; runtimeSessionId?: string } = {},
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    const providerId = this.requiredProvider(body.providerId ?? '');
+    const runtimeSessionId = body.runtimeSessionId?.trim();
+    if (!runtimeSessionId) throw new ForbiddenException('Runtime session identity is required');
+    const context = this.context(user, correlationId);
+    const sessions = await this.runtime.listSessions(providerId, context);
+    if (!sessions.some((session): boolean => session.id === runtimeSessionId)) {
+      throw new ForbiddenException('Runtime session is not visible to this actor');
+    }
+    await this.durable.enroll(
+      {
+        agentName,
+        sessionId,
+        tenantId: context.actorScope.tenantId,
+        ownerId: context.actorScope.userId,
+        providerId,
+        runtimeSessionId,
+      },
+      context,
+    );
+    return { status: 'enrolled', sessionId };
+  }
+
+  @Post('sessions/:sessionId/attach')
+  async attach(
+    @Param('agentName') agentName: string,
+    @Param('sessionId') sessionId: string,
+    @Body() body: { mode?: RuntimeAttachMode } = {},
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    const context = this.context(user, correlationId);
+    const mode = body.mode ?? 'read';
+    if (mode !== 'read' && mode !== 'control') {
+      throw new ForbiddenException('Interaction attach mode is invalid');
+    }
+    const snapshot = await this.durable.getSnapshot(sessionId, context);
+    this.assertSessionAgent(snapshot.identity.agentName, agentName);
+    return this.runtime.attach(
+      snapshot.identity.providerId,
+      snapshot.identity.runtimeSessionId,
+      mode,
+      context,
+    );
+  }
+
+  @Sse('sessions/:sessionId/stream')
+  stream(
+    @Param('agentName') agentName: string,
+    @Param('sessionId') sessionId: string,
+    @Query('cursor') cursor: string | undefined,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ): Observable<{ data: RuntimeStreamEvent }> {
+    this.assertConfiguredAgent(agentName);
+    const context = this.context(user, correlationId);
+    return new Observable((subscriber) => {
+      let iterator: AsyncIterator<RuntimeStreamEvent> | undefined;
+      let cancelled = false;
+      void (async (): Promise<void> => {
+        try {
+          const snapshot = await this.durable.getSnapshot(sessionId, context);
+          if (cancelled || subscriber.closed) return;
+          this.assertSessionAgent(snapshot.identity.agentName, agentName);
+          iterator = this.runtime
+            .streamSession(
+              snapshot.identity.providerId,
+              snapshot.identity.runtimeSessionId,
+              cursor?.trim() || undefined,
+              context,
+            )
+            [Symbol.asyncIterator]();
+          if (cancelled || subscriber.closed) {
+            await iterator.return?.();
+            return;
+          }
+          while (!cancelled && !subscriber.closed) {
+            const next = await iterator.next();
+            if (next.done || cancelled || subscriber.closed) break;
+            subscriber.next({ data: next.value });
+          }
+          if (!subscriber.closed) subscriber.complete();
+        } catch (error: unknown) {
+          if (!subscriber.closed) subscriber.error(error);
+        }
+      })();
+      return (): void => {
+        cancelled = true;
+        void iterator?.return?.();
+      };
+    });
+  }
+
+  @Post('sessions/:sessionId/send')
+  async send(
+    @Param('agentName') agentName: string,
+    @Param('sessionId') sessionId: string,
+    @Body() body: { content?: string; idempotencyKey?: string } = {},
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    if (!body.content?.trim() || !body.idempotencyKey?.trim()) {
+      throw new ForbiddenException('Content and idempotency key are required');
+    }
+    const context = this.context(user, correlationId);
+    const snapshot = await this.durable.getSnapshot(sessionId, context);
+    this.assertSessionAgent(snapshot.identity.agentName, agentName);
+    const input = {
+      sessionId,
+      content: body.content,
+      idempotencyKey: body.idempotencyKey,
+      correlationId: context.correlationId,
+      context,
+    };
+    await this.durable.queueProviderSend(input);
+    await this.durable.dispatchProviderOutbox(sessionId, input);
+    return { status: 'queued', sessionId };
+  }
+
+  @Post('sessions/:sessionId/stop')
+  async stop(
+    @Param('agentName') agentName: string,
+    @Param('sessionId') sessionId: string,
+    @Body() body: { approvalRef?: string } = {},
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    if (!body.approvalRef?.trim())
+      throw new ForbiddenException('Exact-action approval is required');
+    const context = this.context(user, correlationId);
+    const snapshot = await this.durable.getSnapshot(sessionId, context);
+    this.assertSessionAgent(snapshot.identity.agentName, agentName);
+    await this.runtime.terminate(
+      snapshot.identity.providerId,
+      snapshot.identity.runtimeSessionId,
+      body.approvalRef,
+      context,
+    );
+    return { status: 'stopped', sessionId };
+  }
+
+  @Post('sessions/:sessionId/recover')
+  async recover(
+    @Param('agentName') agentName: string,
+    @Param('sessionId') sessionId: string,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ) {
+    this.assertConfiguredAgent(agentName);
+    const context = this.context(user, correlationId);
+    const snapshot = await this.durable.getSnapshot(sessionId, context);
+    this.assertSessionAgent(snapshot.identity.agentName, agentName);
+    await this.durable.recoverProviderSession(sessionId, {
+      sessionId,
+      content: '',
+      idempotencyKey: `recovery:${context.correlationId}`,
+      correlationId: context.correlationId,
+      context,
+    });
+    return { status: 'recovered', sessionId };
+  }
+
+  private context(
+    user: AuthenticatedUserLike,
+    correlationId?: string,
+  ): RuntimeProviderRequestContext {
+    const requestCorrelationId = correlationId?.trim();
+    // This non-simple request header is mandatory for mutations. Browser
+    // cross-origin requests cannot set it without a CORS preflight, and the
+    // gateway's allowlist rejects untrusted origins before the handler runs.
+    if (!requestCorrelationId) {
+      throw new ForbiddenException('X-Correlation-Id is required');
+    }
+    return {
+      actorScope: scopeFromUser(user),
+      channelId: 'cli',
+      correlationId: requestCorrelationId,
+    };
+  }
+
+  private assertConfiguredAgent(agentName: string): void {
+    const configured = process.env['MOSAIC_AGENT_NAME']?.trim();
+    if (!configured || configured !== agentName) {
+      throw new ForbiddenException('Interaction agent is not configured for this request');
+    }
+  }
+
+  private assertSessionAgent(sessionAgentName: string, agentName: string): void {
+    if (sessionAgentName !== agentName)
+      throw new ForbiddenException('Interaction session identity mismatch');
+  }
+
+  private requiredProvider(providerId: string): string {
+    if (!providerId?.trim()) throw new ForbiddenException('Runtime provider is required');
+    return providerId;
+  }
+}

apps/gateway/src/agent/provider.service.ts

@@ -107,8 +107,7 @@ export class ProviderService implements OnModuleInit, OnModuleDestroy {
    * Interval is configurable via PROVIDER_HEALTH_INTERVAL env (seconds, default 60).
    */
   private startHealthCheckScheduler(): void {
-    const intervalSecs =
-      parseInt(process.env['PROVIDER_HEALTH_INTERVAL'] ?? '', 10) || DEFAULT_HEALTH_INTERVAL_SECS;
+    const intervalSecs = this.effectiveHealthCheckIntervalSecs();
     const intervalMs = intervalSecs * 1000;
 
     // Run an initial check immediately (non-blocking)
@@ -176,6 +175,28 @@ export class ProviderService implements OnModuleInit, OnModuleDestroy {
     });
   }
 
+  /**
+   * Returns the effective provider operational policy without credentials,
+   * endpoints, request content, or provider error details.
+   */
+  getEffectivePolicyStatus(): {
+    healthCheckIntervalSecs: number;
+    configuredProviders: string[];
+    availableModelCount: number;
+  } {
+    return {
+      healthCheckIntervalSecs: this.effectiveHealthCheckIntervalSecs(),
+      configuredProviders: this.adapters.map((adapter) => adapter.name),
+      availableModelCount: this.registry?.getAvailable().length ?? 0,
+    };
+  }
+
+  private effectiveHealthCheckIntervalSecs(): number {
+    return (
+      parseInt(process.env['PROVIDER_HEALTH_INTERVAL'] ?? '', 10) || DEFAULT_HEALTH_INTERVAL_SECS
+    );
+  }
+
   // ---------------------------------------------------------------------------
   // Adapter-pattern API
   // ---------------------------------------------------------------------------

apps/gateway/src/agent/providers.controller.test.ts

@@ -0,0 +1,46 @@
+import { describe, expect, it, vi } from 'vitest';
+import { ProvidersController } from './providers.controller.js';
+
+describe('ProvidersController operational status', (): void => {
+  it('reports provider latency and effective policy without exposing provider error details', (): void => {
+    const providerService = {
+      getProvidersHealth: vi.fn(() => [
+        {
+          name: 'fleet',
+          status: 'down',
+          latencyMs: 42,
+          lastChecked: '2026-07-12T00:00:00.000Z',
+          modelCount: 0,
+          error: 'credential-canary=secret-value',
+        },
+      ]),
+      getEffectivePolicyStatus: vi.fn(() => ({
+        healthCheckIntervalSecs: 60,
+        configuredProviders: ['fleet'],
+        availableModelCount: 0,
+      })),
+    };
+    const controller = new ProvidersController(providerService as never, {} as never, {} as never);
+
+    const status = controller.status();
+
+    expect(status).toEqual({
+      providers: [
+        {
+          name: 'fleet',
+          status: 'down',
+          latencyMs: 42,
+          lastChecked: '2026-07-12T00:00:00.000Z',
+          modelCount: 0,
+          errorCode: 'provider_unavailable',
+        },
+      ],
+      effectivePolicy: {
+        healthCheckIntervalSecs: 60,
+        configuredProviders: ['fleet'],
+        availableModelCount: 0,
+      },
+    });
+    expect(JSON.stringify(status)).not.toContain('secret-value');
+  });
+});

apps/gateway/src/agent/providers.controller.ts

@@ -33,7 +33,20 @@ export class ProvidersController {
 
   @Get('health')
   health() {
-    return { providers: this.providerService.getProvidersHealth() };
+    return { providers: this.safeProviderHealth() };
+  }
+
+  /**
+   * Safe operational status for troubleshooting and readiness checks. Provider
+   * errors are reduced to a stable code so credentials and remote responses
+   * cannot leak through this endpoint.
+   */
+  @Get('status')
+  status() {
+    return {
+      providers: this.safeProviderHealth(),
+      effectivePolicy: this.providerService.getEffectivePolicyStatus(),
+    };
   }
 
   @Post('test')
@@ -51,6 +64,13 @@ export class ProvidersController {
     return this.routingService.rank(criteria);
   }
 
+  private safeProviderHealth() {
+    return this.providerService.getProvidersHealth().map(({ error, ...provider }) => ({
+      ...provider,
+      ...(error ? { errorCode: 'provider_unavailable' } : {}),
+    }));
+  }
+
   // ── Credential CRUD ──────────────────────────────────────────────────────
 
   /**

apps/gateway/src/agent/runtime-approval-denied.filter.ts

@@ -0,0 +1,13 @@
+import { Catch, type ArgumentsHost, type ExceptionFilter } from '@nestjs/common';
+import { RuntimeApprovalDeniedError } from './runtime-provider-registry.service.js';
+
+/** Maps a consumed/missing runtime approval to a stable HTTP authorization response. */
+@Catch(RuntimeApprovalDeniedError)
+export class RuntimeApprovalDeniedFilter implements ExceptionFilter {
+  catch(_exception: RuntimeApprovalDeniedError, host: ArgumentsHost): void {
+    const response = host.switchToHttp().getResponse<{
+      status(code: number): { send(body: { statusCode: number; message: string }): void };
+    }>();
+    response.status(403).send({ statusCode: 403, message: 'Runtime termination approval denied' });
+  }
+}

apps/gateway/src/agent/runtime-provider-registry.service.ts

@@ -0,0 +1,500 @@
+import { ForbiddenException, Inject, Injectable, Logger, NotFoundException } from '@nestjs/common';
+import { AgentRuntimeProviderRegistry } from '@mosaicstack/agent';
+import {
+  createRuntimeAuditLogEntry,
+  type LogService,
+  type RuntimeAuditErrorCode,
+} from '@mosaicstack/log';
+import type {
+  AgentRuntimeProvider,
+  RuntimeAttachHandle,
+  RuntimeAttachMode,
+  RuntimeCapability,
+  RuntimeCapabilitySet,
+  RuntimeHealth,
+  RuntimeMessage,
+  RuntimeScope,
+  RuntimeSession,
+  RuntimeSessionTree,
+  RuntimeStreamEvent,
+  TransitionalCapabilityInventoryEntry,
+  TransitionalCapabilityInventoryProvider,
+} from '@mosaicstack/types';
+import type { ActorTenantScope } from '../auth/session-scope.js';
+import { LOG_SERVICE } from '../log/log.tokens.js';
+
+export const AGENT_RUNTIME_PROVIDER_REGISTRY = Symbol('AGENT_RUNTIME_PROVIDER_REGISTRY');
+export const RUNTIME_PROVIDER_AUDIT_SINK = Symbol('RUNTIME_PROVIDER_AUDIT_SINK');
+export const RUNTIME_APPROVAL_VERIFIER = Symbol('RUNTIME_APPROVAL_VERIFIER');
+
+export type RuntimeProviderOperation =
+  | RuntimeCapability
+  | 'runtime.capabilities'
+  | 'runtime.health'
+  | 'runtime.transitional-capabilities';
+export type RuntimeProviderAuditOutcome = 'requested' | 'succeeded' | 'denied' | 'failed';
+
+/** Trusted server-side context only; it intentionally excludes client-provided identity fields. */
+export interface RuntimeProviderRequestContext {
+  actorScope: ActorTenantScope;
+  channelId: string;
+  correlationId: string;
+}
+
+/** Metadata-only audit record. Message bodies, idempotency keys, and approval refs are excluded. */
+export interface RuntimeAuditEvent {
+  providerId: string;
+  operation: RuntimeProviderOperation;
+  outcome: RuntimeProviderAuditOutcome;
+  actorId: string;
+  tenantId: string;
+  channelId: string;
+  correlationId: string;
+  resourceId?: string;
+  durationMs?: number;
+  errorCode?: RuntimeAuditErrorCode;
+}
+
+export interface RuntimeAuditSink {
+  record(event: RuntimeAuditEvent): Promise<void>;
+}
+
+/** Exact action shape that a durable approval implementation must consume once. */
+export interface RuntimeTerminationAction {
+  providerId: string;
+  sessionId: string;
+  actorId: string;
+  tenantId: string;
+  channelId: string;
+  correlationId: string;
+  agentName: string;
+}
+
+export interface RuntimeApprovalVerifier {
+  consume(approvalRef: string, action: RuntimeTerminationAction): Promise<boolean>;
+}
+
+function isTransitionalInventoryProvider(
+  provider: AgentRuntimeProvider,
+): provider is AgentRuntimeProvider & TransitionalCapabilityInventoryProvider {
+  return (
+    typeof (provider as Partial<TransitionalCapabilityInventoryProvider>)
+      .transitionalCapabilityMatrix === 'function'
+  );
+}
+
+function configuredAgentName(): string {
+  const agentName = process.env['MOSAIC_AGENT_NAME']?.trim();
+  if (!agentName) throw new RuntimeApprovalDeniedError();
+  return agentName;
+}
+
+export class RuntimeApprovalDeniedError extends Error {
+  constructor() {
+    super('Runtime termination approval denied');
+  }
+}
+
+/**
+ * The default denies all runtime termination until a durable, exact-action
+ * approval implementation is configured. This is safer than a permissive stub.
+ */
+@Injectable()
+export class DenyRuntimeApprovalVerifier implements RuntimeApprovalVerifier {
+  async consume(_approvalRef: string, _action: RuntimeTerminationAction): Promise<boolean> {
+    return false;
+  }
+}
+
+/**
+ * Temporary metadata-only audit sink. M1 observability can replace this token
+ * with a durable audit writer without changing provider call sites.
+ */
+@Injectable()
+export class RuntimeProviderAuditService implements RuntimeAuditSink {
+  private readonly logger = new Logger(RuntimeProviderAuditService.name);
+
+  constructor(@Inject(LOG_SERVICE) private readonly logService: LogService) {}
+
+  async record(event: RuntimeAuditEvent): Promise<void> {
+    const entry = createRuntimeAuditLogEntry(event);
+    await this.logService.logs.ingest(entry);
+    this.logger.log(JSON.stringify({ event: entry.content, metadata: entry.metadata }));
+  }
+}
+
+@Injectable()
+export class RuntimeProviderService {
+  private readonly logger = new Logger(RuntimeProviderService.name);
+
+  constructor(
+    @Inject(AGENT_RUNTIME_PROVIDER_REGISTRY)
+    private readonly registry: AgentRuntimeProviderRegistry,
+    @Inject(RUNTIME_PROVIDER_AUDIT_SINK)
+    private readonly audit: RuntimeAuditSink,
+    @Inject(RUNTIME_APPROVAL_VERIFIER)
+    private readonly approvals: RuntimeApprovalVerifier,
+  ) {}
+
+  async capabilities(
+    providerId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<RuntimeCapabilitySet> {
+    return this.execute(
+      providerId,
+      'runtime.capabilities',
+      undefined,
+      undefined,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<RuntimeCapabilitySet> =>
+        provider.capabilities(scope),
+    );
+  }
+
+  async health(providerId: string, context: RuntimeProviderRequestContext): Promise<RuntimeHealth> {
+    return this.execute(
+      providerId,
+      'runtime.health',
+      undefined,
+      undefined,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<RuntimeHealth> =>
+        provider.health(scope),
+    );
+  }
+
+  async transitionalCapabilityMatrix(
+    providerId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<TransitionalCapabilityInventoryEntry[]> {
+    return this.execute(
+      providerId,
+      'runtime.transitional-capabilities',
+      undefined,
+      undefined,
+      context,
+      async (provider: AgentRuntimeProvider, scope: RuntimeScope) => {
+        if (!isTransitionalInventoryProvider(provider)) {
+          throw new NotFoundException('Runtime provider has no transitional capability inventory');
+        }
+        return provider.transitionalCapabilityMatrix(scope);
+      },
+    );
+  }
+
+  async listSessions(
+    providerId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<RuntimeSession[]> {
+    return this.execute(
+      providerId,
+      'session.list',
+      'session.list',
+      undefined,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<RuntimeSession[]> =>
+        provider.listSessions(scope),
+    );
+  }
+
+  async getSessionTree(
+    providerId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<RuntimeSessionTree[]> {
+    return this.execute(
+      providerId,
+      'session.tree',
+      'session.tree',
+      undefined,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<RuntimeSessionTree[]> =>
+        provider.getSessionTree(scope),
+    );
+  }
+
+  streamSession(
+    providerId: string,
+    sessionId: string,
+    cursor: string | undefined,
+    context: RuntimeProviderRequestContext,
+  ): AsyncIterable<RuntimeStreamEvent> {
+    return this.stream(
+      providerId,
+      'session.stream',
+      'session.stream',
+      sessionId,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): AsyncIterable<RuntimeStreamEvent> =>
+        provider.streamSession(sessionId, cursor, scope),
+    );
+  }
+
+  async sendMessage(
+    providerId: string,
+    sessionId: string,
+    message: RuntimeMessage,
+    context: RuntimeProviderRequestContext,
+  ): Promise<void> {
+    await this.execute(
+      providerId,
+      'session.send',
+      'session.send',
+      sessionId,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<void> =>
+        provider.sendMessage(sessionId, message, scope),
+    );
+  }
+
+  async attach(
+    providerId: string,
+    sessionId: string,
+    mode: RuntimeAttachMode,
+    context: RuntimeProviderRequestContext,
+  ): Promise<RuntimeAttachHandle> {
+    return this.execute(
+      providerId,
+      'session.attach',
+      'session.attach',
+      sessionId,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<RuntimeAttachHandle> =>
+        provider.attach(sessionId, mode, scope),
+    );
+  }
+
+  async detach(
+    providerId: string,
+    attachmentId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<void> {
+    await this.execute(
+      providerId,
+      'session.attach',
+      'session.attach',
+      attachmentId,
+      context,
+      (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<void> =>
+        provider.detach(attachmentId, scope),
+    );
+  }
+
+  async terminate(
+    providerId: string,
+    sessionId: string,
+    approvalRef: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<void> {
+    await this.execute(
+      providerId,
+      'session.terminate',
+      'session.terminate',
+      sessionId,
+      context,
+      async (provider: AgentRuntimeProvider, scope: RuntimeScope): Promise<void> => {
+        const approved = await this.approvals.consume(approvalRef, {
+          providerId,
+          sessionId,
+          actorId: scope.actorId,
+          tenantId: scope.tenantId,
+          channelId: scope.channelId,
+          correlationId: scope.correlationId,
+          agentName: configuredAgentName(),
+        });
+        if (!approved) {
+          throw new RuntimeApprovalDeniedError();
+        }
+        await provider.terminate(sessionId, approvalRef, scope);
+      },
+    );
+  }
+
+  private async execute<T>(
+    providerId: string,
+    operation: RuntimeProviderOperation,
+    requiredCapability: RuntimeCapability | undefined,
+    resourceId: string | undefined,
+    context: RuntimeProviderRequestContext,
+    invoke: (provider: AgentRuntimeProvider, scope: RuntimeScope) => Promise<T>,
+  ): Promise<T> {
+    const scope = this.deriveScope(context);
+    const startedAt = Date.now();
+    await this.record(providerId, operation, 'requested', scope, resourceId);
+    let invocationStarted = false;
+    try {
+      const provider = this.provider(providerId);
+      if (requiredCapability) {
+        await this.assertCapability(provider, requiredCapability, scope);
+      }
+      invocationStarted = true;
+      const result = await invoke(provider, scope);
+      await this.recordCompletion(providerId, operation, scope, resourceId, Date.now() - startedAt);
+      return result;
+    } catch (error: unknown) {
+      const durationMs = Date.now() - startedAt;
+      if (invocationStarted && !this.isAuthorizationDenied(error)) {
+        await this.recordFailure(providerId, operation, scope, resourceId, durationMs);
+      } else {
+        await this.record(
+          providerId,
+          operation,
+          'denied',
+          scope,
+          resourceId,
+          durationMs,
+          'policy_denied',
+        );
+      }
+      throw error;
+    }
+  }
+
+  private async *stream(
+    providerId: string,
+    operation: RuntimeProviderOperation,
+    requiredCapability: RuntimeCapability,
+    resourceId: string,
+    context: RuntimeProviderRequestContext,
+    invoke: (
+      provider: AgentRuntimeProvider,
+      scope: RuntimeScope,
+    ) => AsyncIterable<RuntimeStreamEvent>,
+  ): AsyncIterable<RuntimeStreamEvent> {
+    const scope = this.deriveScope(context);
+    const startedAt = Date.now();
+    await this.record(providerId, operation, 'requested', scope, resourceId);
+    let invocationStarted = false;
+    try {
+      const provider = this.provider(providerId);
+      await this.assertCapability(provider, requiredCapability, scope);
+      invocationStarted = true;
+      for await (const event of invoke(provider, scope)) {
+        yield event;
+      }
+      await this.recordCompletion(providerId, operation, scope, resourceId, Date.now() - startedAt);
+    } catch (error: unknown) {
+      const durationMs = Date.now() - startedAt;
+      if (invocationStarted) {
+        await this.recordFailure(providerId, operation, scope, resourceId, durationMs);
+      } else {
+        await this.record(
+          providerId,
+          operation,
+          'denied',
+          scope,
+          resourceId,
+          durationMs,
+          'policy_denied',
+        );
+      }
+      throw error;
+    }
+  }
+
+  private isAuthorizationDenied(error: unknown): boolean {
+    return (
+      error instanceof RuntimeApprovalDeniedError ||
+      error instanceof ForbiddenException ||
+      (typeof error === 'object' &&
+        error !== null &&
+        'code' in error &&
+        (error as { code?: unknown }).code === 'forbidden')
+    );
+  }
+
+  private provider(providerId: string): AgentRuntimeProvider {
+    try {
+      return this.registry.require(providerId);
+    } catch (error: unknown) {
+      const message = error instanceof Error ? error.message : 'Runtime provider is not registered';
+      throw new NotFoundException(message);
+    }
+  }
+
+  private async assertCapability(
+    provider: AgentRuntimeProvider,
+    requiredCapability: RuntimeCapability,
+    scope: RuntimeScope,
+  ): Promise<void> {
+    const capabilities = await provider.capabilities(scope);
+    if (!capabilities.supported.includes(requiredCapability)) {
+      throw new ForbiddenException(`Runtime provider capability denied: ${requiredCapability}`);
+    }
+  }
+
+  private deriveScope(context: RuntimeProviderRequestContext): RuntimeScope {
+    const actorId = context.actorScope.userId.trim();
+    const tenantId = context.actorScope.tenantId.trim();
+    const channelId = context.channelId.trim();
+    const correlationId = context.correlationId.trim();
+    if (!actorId || !tenantId || !channelId || !correlationId) {
+      throw new ForbiddenException(
+        'Authenticated runtime actor scope and correlation are required',
+      );
+    }
+    return Object.freeze({ actorId, tenantId, channelId, correlationId });
+  }
+
+  private async recordFailure(
+    providerId: string,
+    operation: RuntimeProviderOperation,
+    scope: RuntimeScope,
+    resourceId: string | undefined,
+    durationMs: number,
+  ): Promise<void> {
+    try {
+      await this.record(
+        providerId,
+        operation,
+        'failed',
+        scope,
+        resourceId,
+        durationMs,
+        'provider_error',
+      );
+    } catch {
+      this.logger.error(
+        `Runtime provider failure audit failed provider=${providerId} operation=${operation} correlation=${scope.correlationId}`,
+      );
+    }
+  }
+
+  private async recordCompletion(
+    providerId: string,
+    operation: RuntimeProviderOperation,
+    scope: RuntimeScope,
+    resourceId: string | undefined,
+    durationMs: number,
+  ): Promise<void> {
+    try {
+      await this.record(providerId, operation, 'succeeded', scope, resourceId, durationMs);
+    } catch {
+      this.logger.error(
+        `Runtime provider completion audit failed provider=${providerId} operation=${operation} correlation=${scope.correlationId}`,
+      );
+    }
+  }
+
+  private async record(
+    providerId: string,
+    operation: RuntimeProviderOperation,
+    outcome: RuntimeProviderAuditOutcome,
+    scope: RuntimeScope,
+    resourceId: string | undefined,
+    durationMs?: number,
+    errorCode?: RuntimeAuditErrorCode,
+  ): Promise<void> {
+    await this.audit.record({
+      providerId,
+      operation,
+      outcome,
+      actorId: scope.actorId,
+      tenantId: scope.tenantId,
+      channelId: scope.channelId,
+      correlationId: scope.correlationId,
+      ...(resourceId ? { resourceId } : {}),
+      ...(durationMs !== undefined ? { durationMs } : {}),
+      ...(errorCode ? { errorCode } : {}),
+    });
+  }
+}

apps/gateway/src/agent/sessions.controller.ts

@@ -10,6 +10,8 @@ import {
   UseGuards,
 } from '@nestjs/common';
 import { AuthGuard } from '../auth/auth.guard.js';
+import { CurrentUser } from '../auth/current-user.decorator.js';
+import { scopeFromUser, type AuthenticatedUserLike } from '../auth/session-scope.js';
 import { AgentService } from './agent.service.js';
 
 @Controller('api/sessions')
@@ -18,23 +20,24 @@ export class SessionsController {
   constructor(@Inject(AgentService) private readonly agentService: AgentService) {}
 
   @Get()
-  list() {
-    const sessions = this.agentService.listSessions();
+  list(@CurrentUser() user: AuthenticatedUserLike) {
+    const sessions = this.agentService.listSessions(scopeFromUser(user));
     return { sessions, total: sessions.length };
   }
 
   @Get(':id')
-  findOne(@Param('id') id: string) {
-    const info = this.agentService.getSessionInfo(id);
+  findOne(@Param('id') id: string, @CurrentUser() user: AuthenticatedUserLike) {
+    const info = this.agentService.getSessionInfo(id, scopeFromUser(user));
     if (!info) throw new NotFoundException('Session not found');
     return info;
   }
 
   @Delete(':id')
   @HttpCode(HttpStatus.NO_CONTENT)
-  async destroy(@Param('id') id: string) {
-    const info = this.agentService.getSessionInfo(id);
+  async destroy(@Param('id') id: string, @CurrentUser() user: AuthenticatedUserLike) {
+    const scope = scopeFromUser(user);
+    const info = this.agentService.getSessionInfo(id, scope);
     if (!info) throw new NotFoundException('Session not found');
-    await this.agentService.destroySession(id);
+    await this.agentService.destroySession(id, scope);
   }
 }

apps/gateway/src/agent/tools/memory-tools.test.ts

@@ -0,0 +1,41 @@
+import { describe, expect, it, vi } from 'vitest';
+import { createMemoryTools } from './memory-tools.js';
+
+describe('createMemoryTools operator retrieval binding', () => {
+  const memory = {
+    insights: { searchByEmbedding: vi.fn(), create: vi.fn() },
+    preferences: { findByUserAndCategory: vi.fn(), findByUser: vi.fn(), upsert: vi.fn() },
+  };
+  const scope = { tenantId: 'tenant-a', ownerId: 'owner-a', sessionId: 'session-a' };
+
+  it('uses the configured plugin with the server-derived scope for retrieval and capture', async () => {
+    const plugin = {
+      search: vi.fn(async () => []),
+      capture: vi.fn(async () => ({ id: 'insight-1' })),
+    };
+    const tools = createMemoryTools(memory as never, null, 'owner-a', {
+      plugin: plugin as never,
+      scope,
+    });
+
+    await tools
+      .find((tool) => tool.name === 'memory_search')!
+      .execute('call-1', { query: 'plans' }, undefined, undefined, {} as never);
+    await tools
+      .find((tool) => tool.name === 'memory_save_insight')!
+      .execute(
+        'call-2',
+        { content: 'secret', category: 'decision' },
+        undefined,
+        undefined,
+        {} as never,
+      );
+
+    expect(plugin.search).toHaveBeenCalledWith(scope, 'plans', 5);
+    expect(plugin.capture).toHaveBeenCalledWith(scope, {
+      content: 'secret',
+      source: 'agent',
+      category: 'decision',
+    });
+  });
+});

apps/gateway/src/agent/tools/memory-tools.ts

@@ -1,7 +1,11 @@
 import { Type } from '@sinclair/typebox';
 import type { ToolDefinition } from '@mariozechner/pi-coding-agent';
-import type { Memory } from '@mosaicstack/memory';
-import type { EmbeddingProvider } from '@mosaicstack/memory';
+import type {
+  EmbeddingProvider,
+  Memory,
+  OperatorMemoryPlugin,
+  OperatorMemoryScope,
+} from '@mosaicstack/memory';
 
 /**
  * Create memory tools bound to the session's authenticated userId.
@@ -13,8 +17,10 @@ import type { EmbeddingProvider } from '@mosaicstack/memory';
 export function createMemoryTools(
   memory: Memory,
   embeddingProvider: EmbeddingProvider | null,
-  /** Authenticated user ID from the session. All memory operations are scoped to this user. */
+  /** Authenticated user ID from the session. All preference operations are scoped to this user. */
   sessionUserId: string | undefined,
+  /** Optional configured retrieval plugin, bound to a server-derived session scope. */
+  operatorMemory?: { plugin: OperatorMemoryPlugin; scope: OperatorMemoryScope },
 ): ToolDefinition[] {
   /** Return an error result when no session user is bound. */
   function noUserError() {
@@ -46,6 +52,14 @@ export function createMemoryTools(
         limit?: number;
       };
 
+      if (operatorMemory) {
+        const results = await operatorMemory.plugin.search(operatorMemory.scope, query, limit ?? 5);
+        return {
+          content: [{ type: 'text' as const, text: JSON.stringify(results, null, 2) }],
+          details: undefined,
+        };
+      }
+
       if (!embeddingProvider) {
         return {
           content: [
@@ -158,6 +172,18 @@ export function createMemoryTools(
       };
       type Cat = 'decision' | 'learning' | 'preference' | 'fact' | 'pattern' | 'general';
 
+      if (operatorMemory) {
+        const insight = await operatorMemory.plugin.capture(operatorMemory.scope, {
+          content,
+          source: 'agent',
+          category: category ?? 'learning',
+        });
+        return {
+          content: [{ type: 'text' as const, text: JSON.stringify(insight, null, 2) }],
+          details: undefined,
+        };
+      }
+
       let embedding: number[] | null = null;
       if (embeddingProvider) {
         embedding = await embeddingProvider.embed(content);

apps/gateway/src/auth/session-scope.ts

@@ -0,0 +1,24 @@
+export interface AuthenticatedUserLike {
+  id: string;
+  tenantId?: string | null;
+  teamId?: string | null;
+  organizationId?: string | null;
+  orgId?: string | null;
+}
+
+export interface ActorTenantScope {
+  userId: string;
+  tenantId: string;
+}
+
+/**
+ * Build the immutable server-derived scope used for Tess session operations.
+ * Current Mosaic auth is user-scoped; future org/team claims can populate one
+ * of the tenant fields without allowing clients to choose another tenant.
+ */
+export function scopeFromUser(user: AuthenticatedUserLike): ActorTenantScope {
+  return {
+    userId: user.id,
+    tenantId: user.tenantId ?? user.teamId ?? user.organizationId ?? user.orgId ?? user.id,
+  };
+}

apps/gateway/src/chat/__tests__/chat-security.test.ts

@@ -12,7 +12,8 @@ describe('Chat controller source hardening', () => {
     const source = readFileSync(resolve('src/chat/chat.controller.ts'), 'utf8');
 
     expect(source).toContain('@UseGuards(AuthGuard)');
-    expect(source).toContain('@CurrentUser() user: { id: string }');
+    expect(source).toContain('@CurrentUser() user: AuthenticatedUserLike');
+    expect(source).toContain('const scope = scopeFromUser(user);');
   });
 });
 

apps/gateway/src/chat/chat.controller.ts

@@ -3,8 +3,10 @@ import {
   Post,
   Body,
   Logger,
+  ForbiddenException,
   HttpException,
   HttpStatus,
+  NotFoundException,
   Inject,
   UseGuards,
 } from '@nestjs/common';
@@ -13,6 +15,7 @@ import { Throttle } from '@nestjs/throttler';
 import { AgentService } from '../agent/agent.service.js';
 import { AuthGuard } from '../auth/auth.guard.js';
 import { CurrentUser } from '../auth/current-user.decorator.js';
+import { scopeFromUser, type AuthenticatedUserLike } from '../auth/session-scope.js';
 import { v4 as uuid } from 'uuid';
 import { ChatRequestDto } from './chat.dto.js';
 
@@ -32,16 +35,23 @@ export class ChatController {
   @Throttle({ default: { limit: 10, ttl: 60_000 } })
   async chat(
     @Body() body: ChatRequestDto,
-    @CurrentUser() user: { id: string },
+    @CurrentUser() user: AuthenticatedUserLike,
   ): Promise<ChatResponse> {
     const conversationId = body.conversationId ?? uuid();
+    const scope = scopeFromUser(user);
 
     try {
-      let agentSession = this.agentService.getSession(conversationId);
+      let agentSession = this.agentService.getSession(conversationId, scope);
       if (!agentSession) {
-        agentSession = await this.agentService.createSession(conversationId);
+        agentSession = await this.agentService.createSession(conversationId, {
+          userId: scope.userId,
+          tenantId: scope.tenantId,
+        });
       }
     } catch (err) {
+      if (err instanceof ForbiddenException) {
+        throw new NotFoundException('Session not found');
+      }
       this.logger.error(
         `Session creation failed for conversation=${conversationId}`,
         err instanceof Error ? err.stack : String(err),
@@ -60,20 +70,27 @@ export class ChatController {
         reject(new Error('Agent response timed out'));
       }, 120_000);
 
-      const cleanup = this.agentService.onEvent(conversationId, (event: AgentSessionEvent) => {
-        if (event.type === 'message_update' && event.assistantMessageEvent.type === 'text_delta') {
-          responseText += event.assistantMessageEvent.delta;
-        }
-        if (event.type === 'agent_end') {
-          clearTimeout(timer);
-          cleanup();
-          resolve();
-        }
-      });
+      const cleanup = this.agentService.onEvent(
+        conversationId,
+        (event: AgentSessionEvent) => {
+          if (
+            event.type === 'message_update' &&
+            event.assistantMessageEvent.type === 'text_delta'
+          ) {
+            responseText += event.assistantMessageEvent.delta;
+          }
+          if (event.type === 'agent_end') {
+            clearTimeout(timer);
+            cleanup();
+            resolve();
+          }
+        },
+        scope,
+      );
     });
 
     try {
-      await this.agentService.prompt(conversationId, body.content);
+      await this.agentService.prompt(conversationId, body.content, scope);
       await done;
     } catch (err) {
       if (err instanceof HttpException) throw err;

apps/gateway/src/chat/chat.gateway-auth.ts

@@ -1,3 +1,4 @@
+import { timingSafeEqual } from 'node:crypto';
 import type { IncomingHttpHeaders } from 'node:http';
 import { fromNodeHeaders } from 'better-auth/node';
 
@@ -12,6 +13,19 @@ export interface SessionAuth {
   };
 }
 
+export function validateDiscordServiceToken(
+  candidate: unknown,
+  expected: string | undefined,
+): boolean {
+  if (typeof candidate !== 'string' || !expected) return false;
+  const candidateBuffer = Buffer.from(candidate);
+  const expectedBuffer = Buffer.from(expected);
+  return (
+    candidateBuffer.length === expectedBuffer.length &&
+    timingSafeEqual(candidateBuffer, expectedBuffer)
+  );
+}
+
 export async function validateSocketSession(
   headers: IncomingHttpHeaders,
   auth: SessionAuth,

apps/gateway/src/chat/chat.gateway-command-approval.spec.ts

@@ -0,0 +1,74 @@
+import { describe, expect, it, vi } from 'vitest';
+import type { SlashCommandPayload } from '@mosaicstack/types';
+import { ChatGateway } from './chat.gateway.js';
+
+const payload: SlashCommandPayload = {
+  command: 'gc',
+  conversationId: 'conversation-1',
+  approvalId: 'approval-1',
+};
+
+function buildGateway(commandExecutor: {
+  execute: ReturnType<typeof vi.fn>;
+  createApproval: ReturnType<typeof vi.fn>;
+}): ChatGateway {
+  return new ChatGateway(
+    {} as never,
+    {} as never,
+    {} as never,
+    {} as never,
+    commandExecutor as never,
+    {} as never,
+  );
+}
+
+describe('ChatGateway command approval ingress', () => {
+  it('passes the client approval ID through to command execution while deriving the actor server-side', async (): Promise<void> => {
+    const commandExecutor = {
+      execute: vi.fn().mockResolvedValue({ ...payload, success: true }),
+      createApproval: vi.fn(),
+    };
+    const gateway = buildGateway(commandExecutor);
+    const client = { data: { user: { id: 'admin-1' } }, emit: vi.fn() };
+
+    await gateway.handleCommandExecute(client as never, payload);
+
+    expect(commandExecutor.execute).toHaveBeenCalledWith(payload, {
+      userId: 'admin-1',
+      tenantId: 'admin-1',
+    });
+    expect(client.emit).toHaveBeenCalledWith(
+      'command:result',
+      expect.objectContaining({ success: true }),
+    );
+  });
+
+  it('issues a durable approval only for the authenticated actor', async (): Promise<void> => {
+    const commandExecutor = {
+      execute: vi.fn(),
+      createApproval: vi.fn().mockResolvedValue({
+        approvalId: 'approval-1',
+        expiresAt: '2026-07-12T00:05:00.000Z',
+      }),
+    };
+    const gateway = buildGateway(commandExecutor);
+    const client = { data: { user: { id: 'admin-1' } }, emit: vi.fn() };
+
+    await gateway.handleCommandApproval(client as never, {
+      command: 'gc',
+      conversationId: 'conversation-1',
+    });
+
+    expect(commandExecutor.createApproval).toHaveBeenCalledWith(
+      { command: 'gc', conversationId: 'conversation-1' },
+      { userId: 'admin-1', tenantId: 'admin-1' },
+    );
+    expect(client.emit).toHaveBeenCalledWith('command:approval', {
+      command: 'gc',
+      conversationId: 'conversation-1',
+      success: true,
+      approvalId: 'approval-1',
+      expiresAt: '2026-07-12T00:05:00.000Z',
+    });
+  });
+});

apps/gateway/src/chat/chat.gateway-redaction.spec.ts

@@ -0,0 +1,170 @@
+import { describe, expect, it, vi } from 'vitest';
+import { ChatGateway } from './chat.gateway.js';
+
+const CONVERSATION_ID = 'conversation-1';
+const CANARY = 'sk_canary12345678';
+
+type GatewayInternals = {
+  clientSessions: Map<string, unknown>;
+  relayEvent(client: unknown, conversationId: string, event: unknown): void;
+};
+
+function buildGateway() {
+  const brain = {
+    conversations: {
+      addMessage: vi.fn().mockResolvedValue(undefined),
+    },
+  };
+  const agentService = {
+    getSession: vi.fn().mockReturnValue(undefined),
+  };
+  const gateway = new ChatGateway(
+    agentService as never,
+    {} as never,
+    brain as never,
+    {} as never,
+    {} as never,
+    {} as never,
+  );
+
+  return { gateway: gateway as unknown as GatewayInternals, brain };
+}
+
+describe('ChatGateway redaction boundary', (): void => {
+  it('redacts a secret split across assistant deltas before egress and persistence', (): void => {
+    const { gateway } = buildGateway();
+    const client = {
+      connected: true,
+      id: 'client-1',
+      data: { user: { id: 'user-1' } },
+      emit: vi.fn(),
+    };
+    const session = {
+      conversationId: CONVERSATION_ID,
+      cleanup: vi.fn(),
+      assistantText: '',
+      toolCalls: [],
+      pendingToolCalls: new Map(),
+      scope: { userId: 'user-1', tenantId: 'tenant-1' },
+    };
+    gateway.clientSessions.set(client.id, session);
+
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: 'sk_canary' },
+    });
+
+    expect(JSON.stringify(client.emit.mock.calls)).not.toContain('sk_canary');
+
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: '12345678 ' },
+    });
+
+    expect(client.emit).toHaveBeenCalledWith('agent:text', {
+      conversationId: CONVERSATION_ID,
+      text: '[REDACTED_SECRET] ',
+    });
+    expect(session.assistantText).toBe(`${CANARY} `);
+    expect(JSON.stringify(client.emit.mock.calls)).not.toContain(CANARY);
+  });
+
+  it('retains a split secret label until its value can be redacted', (): void => {
+    const { gateway } = buildGateway();
+    const client = {
+      connected: true,
+      id: 'client-1',
+      data: { user: { id: 'user-1' } },
+      emit: vi.fn(),
+    };
+
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: 'token ' },
+    });
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: '=canaryvalue123 ' },
+    });
+
+    expect(client.emit).toHaveBeenCalledWith('agent:text', {
+      conversationId: CONVERSATION_ID,
+      text: '[REDACTED_SECRET] ',
+    });
+    expect(JSON.stringify(client.emit.mock.calls)).not.toContain('canaryvalue123');
+  });
+
+  it('holds a streamed private key until it can be redacted', (): void => {
+    const { gateway } = buildGateway();
+    const client = {
+      connected: true,
+      id: 'client-1',
+      data: { user: { id: 'user-1' } },
+      emit: vi.fn(),
+    };
+
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: '-----BEGIN PRIVATE KEY-----\ncanary' },
+    });
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: '\n-----END PRIVATE KEY-----' },
+    });
+
+    expect(client.emit).toHaveBeenCalledWith('agent:text', {
+      conversationId: CONVERSATION_ID,
+      text: '[REDACTED_SECRET]',
+    });
+    expect(JSON.stringify(client.emit.mock.calls)).not.toContain('canary');
+  });
+
+  it('drops an oversized unterminated stream fragment rather than retaining it', (): void => {
+    const { gateway } = buildGateway();
+    const client = {
+      connected: true,
+      id: 'client-1',
+      data: { user: { id: 'user-1' } },
+      emit: vi.fn(),
+    };
+
+    gateway.relayEvent(client, CONVERSATION_ID, {
+      type: 'message_update',
+      assistantMessageEvent: { type: 'text_delta', delta: 'x'.repeat(8_193) },
+    });
+
+    expect(client.emit).toHaveBeenCalledWith('agent:text', {
+      conversationId: CONVERSATION_ID,
+      text: '[REDACTED_STREAM_OVERFLOW]',
+    });
+  });
+
+  it('persists only redacted assistant content with classifications', (): void => {
+    const { gateway, brain } = buildGateway();
+    const client = {
+      connected: true,
+      id: 'client-1',
+      data: { user: { id: 'user-1' } },
+      emit: vi.fn(),
+    };
+    gateway.clientSessions.set(client.id, {
+      conversationId: CONVERSATION_ID,
+      cleanup: vi.fn(),
+      assistantText: CANARY,
+      toolCalls: [],
+      pendingToolCalls: new Map(),
+      scope: { userId: 'user-1', tenantId: 'tenant-1' },
+    });
+
+    gateway.relayEvent(client, CONVERSATION_ID, { type: 'agent_end' });
+
+    expect(brain.conversations.addMessage).toHaveBeenCalledWith(
+      expect.objectContaining({
+        content: '[REDACTED_SECRET]',
+        metadata: expect.objectContaining({ classifications: ['secret'] }),
+      }),
+      'user-1',
+    );
+    expect(JSON.stringify(brain.conversations.addMessage.mock.calls)).not.toContain(CANARY);
+  });
+});

apps/gateway/src/chat/chat.gateway.ts

@@ -1,4 +1,5 @@
-import { Inject, Logger } from '@nestjs/common';
+import { createHash } from 'node:crypto';
+import { Inject, Logger, Optional } from '@nestjs/common';
 import {
   WebSocketGateway,
   WebSocketServer,
@@ -11,24 +12,47 @@ import {
 } from '@nestjs/websockets';
 import { Server, Socket } from 'socket.io';
 import type { AgentSessionEvent } from '@mariozechner/pi-coding-agent';
+import {
+  verifyDiscordIngressEnvelope,
+  parseDiscordInteractionBindings,
+  resolveDiscordInteractionActorId,
+  resolveDiscordInteractionBinding,
+  type DiscordIngressEnvelope,
+  type DiscordIngressPayload,
+} from '@mosaicstack/discord-plugin';
 import type { Auth } from '@mosaicstack/auth';
 import type { Brain } from '@mosaicstack/brain';
+import { redactSensitiveContent } from '@mosaicstack/log';
 import type {
   SetThinkingPayload,
+  SlashCommandApprovalResultPayload,
   SlashCommandPayload,
   SystemReloadPayload,
   RoutingDecisionInfo,
   AbortPayload,
 } from '@mosaicstack/types';
 import { AgentService, type ConversationHistoryMessage } from '../agent/agent.service.js';
+import {
+  RUNTIME_PROVIDER_AUDIT_SINK,
+  RuntimeProviderService,
+  type RuntimeAuditSink,
+} from '../agent/runtime-provider-registry.service.js';
+import { DurableSessionService } from '../agent/durable-session.service.js';
 import { AUTH } from '../auth/auth.tokens.js';
+import {
+  scopeFromUser,
+  type ActorTenantScope,
+  type AuthenticatedUserLike,
+} from '../auth/session-scope.js';
 import { BRAIN } from '../brain/brain.tokens.js';
 import { CommandRegistryService } from '../commands/command-registry.service.js';
 import { CommandExecutorService } from '../commands/command-executor.service.js';
+import { CommandAuthorizationService } from '../commands/command-authorization.service.js';
 import { RoutingEngineService } from '../agent/routing/routing-engine.service.js';
 import { v4 as uuid } from 'uuid';
 import { ChatSocketMessageDto } from './chat.dto.js';
-import { validateSocketSession } from './chat.gateway-auth.js';
+import { validateDiscordServiceToken, validateSocketSession } from './chat.gateway-auth.js';
+import { DiscordReplayProtector } from '../plugin/discord-replay-protector.js';
 
 /** Per-client state tracking streaming accumulation for persistence. */
 interface ClientSession {
@@ -40,6 +64,8 @@ interface ClientSession {
   toolCalls: Array<{ toolCallId: string; toolName: string; args: unknown; isError: boolean }>;
   /** Tool calls in-flight (started but not ended yet). */
   pendingToolCalls: Map<string, { toolName: string; args: unknown }>;
+  /** Server-derived owner/tenant scope for this socket's conversation attachment. */
+  scope: ActorTenantScope;
   /** Last routing decision made for this session (M4-008) */
   lastRoutingDecision?: RoutingDecisionInfo;
 }
@@ -49,6 +75,38 @@ interface ClientSession {
  * Keyed by conversationId, value is the model name to use.
  */
 const modelOverrides = new Map<string, string>();
+const MAX_REDACTION_BUFFER_LENGTH = 8_192;
+
+function isDiscordIngressEnvelope(value: unknown): value is DiscordIngressEnvelope {
+  if (typeof value !== 'object' || value === null) return false;
+  const envelope = value as { payload?: unknown; signature?: unknown };
+  if (
+    typeof envelope.signature !== 'string' ||
+    typeof envelope.payload !== 'object' ||
+    envelope.payload === null
+  ) {
+    return false;
+  }
+  const payload = envelope.payload as Record<string, unknown>;
+  return [
+    payload['correlationId'],
+    payload['messageId'],
+    payload['guildId'],
+    payload['channelId'],
+    payload['userId'],
+    payload['conversationId'],
+    payload['content'],
+  ].every((field: unknown): boolean => typeof field === 'string');
+}
+
+function isChatSocketMessage(value: unknown): value is ChatSocketMessageDto {
+  if (typeof value !== 'object' || value === null) return false;
+  const payload = value as { content?: unknown; conversationId?: unknown };
+  return (
+    typeof payload.content === 'string' &&
+    (payload.conversationId === undefined || typeof payload.conversationId === 'string')
+  );
+}
 
 @WebSocketGateway({
   cors: {
@@ -62,6 +120,11 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
 
   private readonly logger = new Logger(ChatGateway.name);
   private readonly clientSessions = new Map<string, ClientSession>();
+  /** Raw stream fragments are kept in memory only until they are safe to redact and emit. */
+  private readonly textEgressBuffers = new Map<string, string>();
+  private readonly thinkingEgressBuffers = new Map<string, string>();
+  private readonly overflowedEgress = new Set<string>();
+  private readonly discordReplayProtector = new DiscordReplayProtector();
 
   constructor(
     @Inject(AgentService) private readonly agentService: AgentService,
@@ -70,6 +133,18 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     @Inject(CommandRegistryService) private readonly commandRegistry: CommandRegistryService,
     @Inject(CommandExecutorService) private readonly commandExecutor: CommandExecutorService,
     @Inject(RoutingEngineService) private readonly routingEngine: RoutingEngineService,
+    @Optional()
+    @Inject(CommandAuthorizationService)
+    private readonly commandAuthorization: CommandAuthorizationService | null = null,
+    @Optional()
+    @Inject(RuntimeProviderService)
+    private readonly runtimeRegistry: RuntimeProviderService | null = null,
+    @Optional()
+    @Inject(DurableSessionService)
+    private readonly durableSessions: DurableSessionService | null = null,
+    @Optional()
+    @Inject(RUNTIME_PROVIDER_AUDIT_SINK)
+    private readonly runtimeAudit: RuntimeAuditSink | null = null,
   ) {}
 
   afterInit(): void {
@@ -77,6 +152,13 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
   }
 
   async handleConnection(client: Socket): Promise<void> {
+    const serviceToken = client.handshake.auth['discordServiceToken'];
+    if (validateDiscordServiceToken(serviceToken, process.env['DISCORD_SERVICE_TOKEN'])) {
+      client.data.discordService = true;
+      this.logger.log(`Authenticated Discord service connected: ${client.id}`);
+      return;
+    }
+
     const session = await validateSocketSession(client.handshake.headers, this.auth);
     if (!session) {
       this.logger.warn(`Rejected unauthenticated WebSocket client: ${client.id}`);
@@ -87,8 +169,6 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     client.data.user = session.user;
     client.data.session = session.session;
     this.logger.log(`Client connected: ${client.id}`);
-
-    // Broadcast command manifest to the newly connected client
     client.emit('commands:manifest', { manifest: this.commandRegistry.getManifest() });
   }
 
@@ -97,25 +177,84 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     const session = this.clientSessions.get(client.id);
     if (session) {
       session.cleanup();
-      this.agentService.removeChannel(session.conversationId, `websocket:${client.id}`);
+      this.agentService.removeChannel(
+        session.conversationId,
+        `websocket:${client.id}`,
+        session.scope,
+      );
       this.clientSessions.delete(client.id);
     }
+    this.textEgressBuffers.delete(client.id);
+    this.thinkingEgressBuffers.delete(client.id);
+    this.overflowedEgress.delete(this.egressKey(client, 'agent:text'));
+    this.overflowedEgress.delete(this.egressKey(client, 'agent:thinking'));
+  }
+
+  private getClientScope(client: Socket): ActorTenantScope | null {
+    const user = client.data.user as AuthenticatedUserLike | undefined;
+    if (!user?.id) return null;
+    return scopeFromUser(user);
+  }
+
+  private modelOverrideKey(conversationId: string, scope: ActorTenantScope): string {
+    return `${scope.tenantId}:${scope.userId}:${conversationId}`;
+  }
+
+  private scopesEqual(a: ActorTenantScope, b: ActorTenantScope): boolean {
+    return a.userId === b.userId && a.tenantId === b.tenantId;
   }
 
   @SubscribeMessage('message')
   async handleMessage(
     @ConnectedSocket() client: Socket,
-    @MessageBody() data: ChatSocketMessageDto,
+    @MessageBody() rawData: unknown,
   ): Promise<void> {
+    let discordIngress: DiscordIngressPayload | null = null;
+    let data: ChatSocketMessageDto;
+    if (client.data.discordService) {
+      if (!isDiscordIngressEnvelope(rawData)) {
+        this.logger.warn(`Rejected malformed Discord ingress from ${client.id}`);
+        return;
+      }
+      discordIngress = this.resolveDiscordIngress(client, rawData);
+      if (!discordIngress) return;
+      data = { conversationId: discordIngress.conversationId, content: discordIngress.content };
+    } else {
+      if (!isChatSocketMessage(rawData)) {
+        this.logger.warn(`Rejected malformed chat message from ${client.id}`);
+        return;
+      }
+      data = rawData;
+    }
     const conversationId = data.conversationId ?? uuid();
-    const userId = (client.data.user as { id: string } | undefined)?.id;
+    const discordServiceUserId = process.env['DISCORD_SERVICE_USER_ID'];
+    if (discordIngress && !discordServiceUserId) {
+      this.logger.warn(
+        `Rejected Discord ingress without configured service owner from ${client.id}`,
+      );
+      return;
+    }
+    const scope = discordIngress
+      ? {
+          userId: discordServiceUserId!,
+          tenantId: process.env['DISCORD_SERVICE_TENANT_ID'] ?? discordServiceUserId!,
+        }
+      : this.getClientScope(client);
+    if (!scope) {
+      client.emit('error', { conversationId, error: 'Authenticated user scope is required.' });
+      return;
+    }
+    const userId = scope.userId;
+    const correlationId = discordIngress?.correlationId;
 
-    this.logger.log(`Message from ${client.id} in conversation ${conversationId}`);
+    this.logger.log(
+      `Message from ${client.id} in conversation ${conversationId}${correlationId ? ` correlation=${correlationId}` : ''}`,
+    );
 
     // Ensure agent session exists for this conversation
     let sessionRoutingDecision: RoutingDecisionInfo | undefined;
     try {
-      let agentSession = this.agentService.getSession(conversationId);
+      let agentSession = this.agentService.getSession(conversationId, scope);
       if (!agentSession) {
         // When resuming an existing conversation, load prior messages to inject as context (M1-004)
         const conversationHistory = await this.loadConversationHistory(conversationId, userId);
@@ -135,7 +274,7 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
         let resolvedProvider = data.provider;
         let resolvedModelId = data.modelId;
 
-        const modelOverride = modelOverrides.get(conversationId);
+        const modelOverride = modelOverrides.get(this.modelOverrideKey(conversationId, scope));
         if (modelOverride) {
           // /model override bypasses routing engine (M4-007)
           resolvedModelId = modelOverride;
@@ -172,6 +311,7 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
           modelId: resolvedModelId,
           agentConfigId: data.agentId,
           userId,
+          tenantId: scope.tenantId,
           conversationHistory: conversationHistory.length > 0 ? conversationHistory : undefined,
         });
 
@@ -210,9 +350,17 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
           {
             conversationId,
             role: 'user',
-            content: data.content,
+            content: redactSensitiveContent(data.content).content,
             metadata: {
               timestamp: new Date().toISOString(),
+              ...(correlationId
+                ? {
+                    correlationId,
+                    discordMessageId: discordIngress?.messageId,
+                    discordUserId: discordIngress?.userId,
+                  }
+                : {}),
+              classifications: redactSensitiveContent(data.content).classifications,
             },
           },
           userId,
@@ -232,9 +380,13 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     }
 
     // Subscribe to agent events and relay to client
-    const cleanup = this.agentService.onEvent(conversationId, (event: AgentSessionEvent) => {
-      this.relayEvent(client, conversationId, event);
-    });
+    const cleanup = this.agentService.onEvent(
+      conversationId,
+      (event: AgentSessionEvent) => {
+        this.relayEvent(client, conversationId, event);
+      },
+      scope,
+    );
 
     // Preserve routing decision from the existing client session if we didn't get a new one
     const prevClientSession = this.clientSessions.get(client.id);
@@ -246,16 +398,17 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
       assistantText: '',
       toolCalls: [],
       pendingToolCalls: new Map(),
+      scope,
       lastRoutingDecision: routingDecisionToStore,
     });
 
     // Track channel connection
-    this.agentService.addChannel(conversationId, `websocket:${client.id}`);
+    this.agentService.addChannel(conversationId, `websocket:${client.id}`, scope);
 
     // Send session info so the client knows the model/provider (M4-008: include routing decision)
     // Include agentName when a named agent config is active (M5-001)
     {
-      const agentSession = this.agentService.getSession(conversationId);
+      const agentSession = this.agentService.getSession(conversationId, scope);
       if (agentSession) {
         const piSession = agentSession.piSession;
         client.emit('session:info', {
@@ -271,11 +424,21 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     }
 
     // Send acknowledgment
-    client.emit('message:ack', { conversationId, messageId: uuid() });
+    client.emit('message:ack', {
+      conversationId,
+      messageId: uuid(),
+      ...(correlationId
+        ? {
+            correlationId,
+            discordMessageId: discordIngress?.messageId,
+            discordUserId: discordIngress?.userId,
+          }
+        : {}),
+    });
 
     // Dispatch to agent
     try {
-      await this.agentService.prompt(conversationId, data.content);
+      await this.agentService.prompt(conversationId, data.content, scope);
     } catch (err) {
       this.logger.error(
         `Agent prompt failed for client=${client.id}, conversation=${conversationId}`,
@@ -293,7 +456,16 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     @ConnectedSocket() client: Socket,
     @MessageBody() data: SetThinkingPayload,
   ): void {
-    const session = this.agentService.getSession(data.conversationId);
+    const scope = this.getClientScope(client);
+    if (!scope) {
+      client.emit('error', {
+        conversationId: data.conversationId,
+        error: 'Authenticated user scope is required.',
+      });
+      return;
+    }
+
+    const session = this.agentService.getSession(data.conversationId, scope);
     if (!session) {
       client.emit('error', {
         conversationId: data.conversationId,
@@ -334,7 +506,13 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     const conversationId = data.conversationId;
     this.logger.log(`Abort requested by ${client.id} for conversation ${conversationId}`);
 
-    const session = this.agentService.getSession(conversationId);
+    const scope = this.getClientScope(client);
+    if (!scope) {
+      client.emit('error', { conversationId, error: 'Authenticated user scope is required.' });
+      return;
+    }
+
+    const session = this.agentService.getSession(conversationId, scope);
     if (!session) {
       client.emit('error', {
         conversationId,
@@ -363,11 +541,45 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     @ConnectedSocket() client: Socket,
     @MessageBody() payload: SlashCommandPayload,
   ): Promise<void> {
-    const userId = (client.data.user as { id: string } | undefined)?.id ?? 'unknown';
-    const result = await this.commandExecutor.execute(payload, userId);
+    const scope = this.getClientScope(client);
+    if (!scope) {
+      client.emit('command:result', {
+        command: payload.command,
+        conversationId: payload.conversationId,
+        success: false,
+        message: 'Authenticated user scope is required.',
+      });
+      return;
+    }
+
+    const result = await this.commandExecutor.execute(payload, scope);
     client.emit('command:result', result);
   }
 
+  @SubscribeMessage('command:approve')
+  async handleCommandApproval(
+    @ConnectedSocket() client: Socket,
+    @MessageBody() payload: SlashCommandPayload,
+  ): Promise<void> {
+    const scope = this.getClientScope(client);
+    const approval = scope ? await this.commandExecutor.createApproval(payload, scope) : null;
+    const result: SlashCommandApprovalResultPayload = approval
+      ? {
+          command: payload.command,
+          conversationId: payload.conversationId,
+          success: true,
+          approvalId: approval.approvalId,
+          expiresAt: approval.expiresAt,
+        }
+      : {
+          command: payload.command,
+          conversationId: payload.conversationId,
+          success: false,
+          message: 'Not authorized to approve this command.',
+        };
+    client.emit('command:approval', result);
+  }
+
   broadcastReload(payload: SystemReloadPayload): void {
     this.server.emit('system:reload', payload);
     this.logger.log('Broadcasted system:reload to all connected clients');
@@ -380,18 +592,23 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
    * M5-005: Emits session:info to clients subscribed to this conversation when a model is set.
    * M5-007: Records a model switch in session metrics.
    */
-  setModelOverride(conversationId: string, modelName: string | null): void {
+  setModelOverride(
+    conversationId: string,
+    modelName: string | null,
+    scope: ActorTenantScope,
+  ): void {
+    const key = this.modelOverrideKey(conversationId, scope);
     if (modelName) {
-      modelOverrides.set(conversationId, modelName);
+      modelOverrides.set(key, modelName);
       this.logger.log(`Model override set: conversation=${conversationId} model="${modelName}"`);
 
       // M5-002: Update the live session's modelId so session:info reflects the new model immediately
-      this.agentService.updateSessionModel(conversationId, modelName);
+      this.agentService.updateSessionModel(conversationId, modelName, scope);
 
       // M5-005: Broadcast session:info to all clients subscribed to this conversation
-      this.broadcastSessionInfo(conversationId);
+      this.broadcastSessionInfo(conversationId, scope);
     } else {
-      modelOverrides.delete(conversationId);
+      modelOverrides.delete(key);
       this.logger.log(`Model override cleared: conversation=${conversationId}`);
     }
   }
@@ -399,8 +616,8 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
   /**
    * Return the active model override for a conversation, or undefined if none.
    */
-  getModelOverride(conversationId: string): string | undefined {
-    return modelOverrides.get(conversationId);
+  getModelOverride(conversationId: string, scope: ActorTenantScope): string | undefined {
+    return modelOverrides.get(this.modelOverrideKey(conversationId, scope));
   }
 
   /**
@@ -409,9 +626,10 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
    */
   broadcastSessionInfo(
     conversationId: string,
+    scope: ActorTenantScope,
     extra?: { agentName?: string; routingDecision?: RoutingDecisionInfo },
   ): void {
-    const agentSession = this.agentService.getSession(conversationId);
+    const agentSession = this.agentService.getSession(conversationId, scope);
     if (!agentSession) return;
 
     const piSession = agentSession.piSession;
@@ -428,7 +646,7 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
 
     // Emit to all clients currently subscribed to this conversation
     for (const [clientId, session] of this.clientSessions) {
-      if (session.conversationId === conversationId) {
+      if (session.conversationId === conversationId && this.scopesEqual(session.scope, scope)) {
         const socket = this.server.sockets.sockets.get(clientId);
         if (socket?.connected) {
           socket.emit('session:info', payload);
@@ -442,6 +660,233 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
    * Creates it if absent — safe to call concurrently since a duplicate insert
    * would fail on the PK constraint and be caught here.
    */
+  @SubscribeMessage('discord:approve')
+  async handleDiscordApproval(
+    @ConnectedSocket() client: Socket,
+    @MessageBody() envelope: DiscordIngressEnvelope,
+  ): Promise<void> {
+    if (!client.data.discordService) return;
+    const ingress = this.resolveDiscordIngress(client, envelope, 'approve');
+    const isApprovalCommand = /^\/approve\s*$/i.test(ingress?.content ?? '');
+    const tenantId = process.env['DISCORD_SERVICE_TENANT_ID']?.trim();
+    if (
+      !ingress ||
+      !isApprovalCommand ||
+      !tenantId ||
+      !this.commandAuthorization ||
+      !this.durableSessions
+    )
+      return;
+    const binding = resolveDiscordInteractionBinding(
+      parseDiscordInteractionBindings(process.env['DISCORD_INTERACTION_BINDINGS']),
+      ingress.guildId,
+      ingress.channelId,
+      ingress.userId,
+      'approve',
+    );
+    const actorId = binding && resolveDiscordInteractionActorId(binding, ingress.userId);
+    const agentName = process.env['MOSAIC_AGENT_NAME']?.trim();
+    if (!actorId || !agentName || binding.instanceId !== agentName) {
+      this.logger.warn(
+        `Rejected Discord approval without a matching runtime agent from ${client.id}`,
+      );
+      client.emit('discord:approval', {
+        correlationId: ingress.correlationId,
+        success: false,
+        approvalId: undefined,
+        expiresAt: undefined,
+      });
+      return;
+    }
+    let snapshot;
+    try {
+      snapshot = await this.durableSessions.getSnapshot(ingress.conversationId, {
+        actorScope: { userId: actorId, tenantId },
+        channelId: ingress.channelId,
+        correlationId: ingress.correlationId,
+      });
+    } catch {
+      client.emit('discord:approval', {
+        correlationId: ingress.correlationId,
+        success: false,
+        approvalId: undefined,
+        expiresAt: undefined,
+      });
+      return;
+    }
+    if (snapshot.identity.agentName !== agentName) {
+      client.emit('discord:approval', {
+        correlationId: ingress.correlationId,
+        success: false,
+        approvalId: undefined,
+        expiresAt: undefined,
+      });
+      return;
+    }
+    const approval = await this.commandAuthorization.createRuntimeTerminationApproval({
+      providerId: snapshot.identity.providerId,
+      sessionId: snapshot.identity.runtimeSessionId,
+      actorId,
+      tenantId,
+      channelId: ingress.channelId,
+      correlationId: this.discordRuntimeActionCorrelation(
+        binding.instanceId,
+        ingress,
+        snapshot.identity.providerId,
+        snapshot.identity.runtimeSessionId,
+      ),
+      agentName,
+    });
+    if (!approval) {
+      await this.runtimeAudit?.record({
+        providerId: snapshot.identity.providerId,
+        operation: 'session.terminate',
+        outcome: 'denied',
+        actorId,
+        tenantId,
+        channelId: ingress.channelId,
+        correlationId: this.discordRuntimeActionCorrelation(
+          binding.instanceId,
+          ingress,
+          snapshot.identity.providerId,
+          snapshot.identity.runtimeSessionId,
+        ),
+        resourceId: snapshot.identity.runtimeSessionId,
+        errorCode: 'policy_denied',
+      });
+    }
+    client.emit('discord:approval', {
+      correlationId: ingress.correlationId,
+      success: approval !== null,
+      approvalId: approval?.approvalId,
+      expiresAt: approval?.expiresAt,
+    });
+  }
+
+  @SubscribeMessage('discord:stop')
+  async handleDiscordStop(
+    @ConnectedSocket() client: Socket,
+    @MessageBody() envelope: DiscordIngressEnvelope,
+  ): Promise<void> {
+    if (!client.data.discordService) return;
+    const ingress = this.resolveDiscordIngress(client, envelope, 'stop');
+    const approvalRef = /^\/stop\s+([^\s]+)$/i.exec(ingress?.content ?? '')?.[1];
+    const tenantId = process.env['DISCORD_SERVICE_TENANT_ID']?.trim();
+    if (!ingress || !approvalRef || !tenantId || !this.runtimeRegistry || !this.durableSessions)
+      return;
+    const binding = resolveDiscordInteractionBinding(
+      parseDiscordInteractionBindings(process.env['DISCORD_INTERACTION_BINDINGS']),
+      ingress.guildId,
+      ingress.channelId,
+      ingress.userId,
+      'stop',
+    );
+    const actorId = binding && resolveDiscordInteractionActorId(binding, ingress.userId);
+    if (!actorId) return;
+
+    try {
+      const context = {
+        actorScope: { userId: actorId, tenantId },
+        channelId: ingress.channelId,
+        correlationId: ingress.correlationId,
+      };
+      const snapshot = await this.durableSessions.getSnapshot(ingress.conversationId, context);
+      if (snapshot.identity.agentName !== binding.instanceId) throw new Error('agent mismatch');
+      // RuntimeProviderService consumes the durable approval exactly once using the
+      // provisioned approving-admin identity, never the Discord service account.
+      await this.runtimeRegistry.terminate(
+        snapshot.identity.providerId,
+        snapshot.identity.runtimeSessionId,
+        approvalRef,
+        {
+          ...context,
+          correlationId: this.discordRuntimeActionCorrelation(
+            binding.instanceId,
+            ingress,
+            snapshot.identity.providerId,
+            snapshot.identity.runtimeSessionId,
+          ),
+        },
+      );
+      client.emit('discord:stop', { correlationId: ingress.correlationId, success: true });
+    } catch {
+      client.emit('discord:stop', { correlationId: ingress.correlationId, success: false });
+    }
+  }
+
+  /**
+   * Correlates the immutable termination target rather than either Discord message.
+   * Approval and stop are distinct ingress events, but must consume the same seven-field action.
+   */
+  private discordRuntimeActionCorrelation(
+    instanceId: string,
+    ingress: DiscordIngressPayload,
+    providerId: string,
+    sessionId: string,
+  ): string {
+    const target = [
+      instanceId,
+      ingress.guildId,
+      ingress.channelId,
+      ingress.conversationId,
+      providerId,
+      sessionId,
+    ];
+    return `discord-action:v1:${createHash('sha256').update(JSON.stringify(target)).digest('hex')}`;
+  }
+
+  private resolveDiscordIngress(
+    client: Socket,
+    envelope: DiscordIngressEnvelope,
+    operation: 'send' | 'approve' | 'stop' = 'send',
+  ): DiscordIngressPayload | null {
+    const payload = verifyDiscordIngressEnvelope(
+      envelope,
+      process.env['DISCORD_SERVICE_TOKEN'] ?? '',
+      {
+        guildIds: this.readDiscordAllowlist('DISCORD_ALLOWED_GUILD_IDS'),
+        channelIds: this.readDiscordAllowlist('DISCORD_ALLOWED_CHANNEL_IDS'),
+        userIds: this.readDiscordAllowlist('DISCORD_ALLOWED_USER_IDS'),
+      },
+    );
+    if (!payload) {
+      this.logger.warn(`Rejected invalid Discord ingress envelope from ${client.id}`);
+      return null;
+    }
+    try {
+      const binding = resolveDiscordInteractionBinding(
+        parseDiscordInteractionBindings(process.env['DISCORD_INTERACTION_BINDINGS']),
+        payload.guildId,
+        payload.channelId,
+        payload.userId,
+        operation,
+      );
+      if (!binding) {
+        this.logger.warn(`Rejected unpaired Discord ingress from ${client.id}`);
+        return null;
+      }
+    } catch {
+      this.logger.warn(
+        `Rejected Discord ingress without valid binding configuration from ${client.id}`,
+      );
+      return null;
+    }
+    if (!this.discordReplayProtector.claim(payload.messageId)) {
+      this.logger.warn(
+        `Rejected replayed Discord message=${payload.messageId} correlation=${payload.correlationId}`,
+      );
+      return null;
+    }
+    return payload;
+  }
+
+  private readDiscordAllowlist(name: string): string[] {
+    return (process.env[name] ?? '')
+      .split(',')
+      .map((id: string): string => id.trim())
+      .filter((id: string): boolean => id.length > 0);
+  }
+
   private async ensureConversation(conversationId: string, userId: string): Promise<void> {
     try {
       const existing = await this.brain.conversations.findById(conversationId, userId);
@@ -527,6 +972,127 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
     }
   }
 
+  private appendAndFlushRedactedEgress(
+    client: Socket,
+    conversationId: string,
+    eventName: 'agent:text' | 'agent:thinking',
+    buffers: Map<string, string>,
+    delta: string,
+  ): void {
+    const key = this.egressKey(client, eventName);
+    if (this.overflowedEgress.has(key)) return;
+
+    const buffered = `${buffers.get(client.id) ?? ''}${delta}`;
+    if (buffered.length > MAX_REDACTION_BUFFER_LENGTH) {
+      buffers.delete(client.id);
+      this.overflowedEgress.add(key);
+      client.emit(eventName, { conversationId, text: '[REDACTED_STREAM_OVERFLOW]' });
+      return;
+    }
+
+    buffers.set(client.id, buffered);
+    this.flushRedactedEgress(client, conversationId, eventName, buffers, false);
+  }
+
+  /**
+   * Holds any suffix that could become a secret, email, or phone number after a
+   * later stream chunk. This avoids relying on downstream redaction after data
+   * has already reached the socket.
+   */
+  private flushRedactedEgress(
+    client: Socket,
+    conversationId: string,
+    eventName: 'agent:text' | 'agent:thinking',
+    buffers: Map<string, string>,
+    final: boolean,
+  ): void {
+    const key = this.egressKey(client, eventName);
+    if (this.overflowedEgress.has(key)) {
+      if (final) this.overflowedEgress.delete(key);
+      return;
+    }
+
+    const buffered = buffers.get(client.id) ?? '';
+    const releaseLength = final ? buffered.length : this.safeRedactionPrefixLength(buffered);
+    const released = buffered.slice(0, releaseLength);
+    const pending = buffered.slice(releaseLength);
+
+    if (pending) {
+      buffers.set(client.id, pending);
+    } else {
+      buffers.delete(client.id);
+    }
+
+    if (released) {
+      client.emit(eventName, {
+        conversationId,
+        text: redactSensitiveContent(released).content,
+      });
+    }
+  }
+
+  private safeRedactionPrefixLength(content: string): number {
+    let retainedFrom = content.length;
+
+    // Retain the current token because it may become a split secret or email.
+    const token = /(?:^|\s)(\S*)$/.exec(content);
+    if (token) {
+      const matched = token[0] ?? '';
+      const trailingToken = token[1] ?? '';
+      retainedFrom = token.index + matched.length - trailingToken.length;
+    }
+
+    // The secret classifier accepts whitespace around ':' and '=', so preserve
+    // a pending label until its value and delimiter are both complete.
+    const pendingSecretLabel =
+      /(?:^|[^A-Za-z0-9_])((?:api[_-]?key|token|password|secret|bearer|authorization)\s*)$/i.exec(
+        content,
+      );
+    if (pendingSecretLabel) {
+      const label = pendingSecretLabel[1] ?? '';
+      retainedFrom = Math.min(
+        retainedFrom,
+        pendingSecretLabel.index + pendingSecretLabel[0].length - label.length,
+      );
+    }
+
+    const secretLabel = /(?:api[_-]?key|token|password|secret|authorization)\s*[:=]\s*$/i.exec(
+      content,
+    );
+    if (secretLabel) {
+      retainedFrom = Math.min(retainedFrom, secretLabel.index);
+    }
+
+    // Phone numbers can contain whitespace and punctuation; preserve the full
+    // trailing numeric candidate until a non-phone character establishes a boundary.
+    const phone = /(?:^|[^A-Za-z0-9_])(\+?\d[\d(). -]*)$/.exec(content);
+    if (phone) {
+      const matched = phone[0] ?? '';
+      const trailingPhoneCandidate = phone[1] ?? '';
+      retainedFrom = Math.min(
+        retainedFrom,
+        phone.index + matched.length - trailingPhoneCandidate.length,
+      );
+    }
+
+    const privateKeyStart = content.lastIndexOf('-----BEGIN');
+    if (privateKeyStart >= 0) {
+      const privateKey = content.slice(privateKeyStart);
+      if (/-----END(?: [A-Z]+)* KEY-----/.test(privateKey)) {
+        // Release the complete block in one pass so the full-block classifier can redact it.
+        retainedFrom = content.length;
+      } else {
+        retainedFrom = Math.min(retainedFrom, privateKeyStart);
+      }
+    }
+
+    return retainedFrom;
+  }
+
+  private egressKey(client: Socket, eventName: 'agent:text' | 'agent:thinking'): string {
+    return `${client.id}:${eventName}`;
+  }
+
   private relayEvent(client: Socket, conversationId: string, event: AgentSessionEvent): void {
     if (!client.connected) {
       this.logger.warn(
@@ -544,13 +1110,20 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
           cs.toolCalls = [];
           cs.pendingToolCalls.clear();
         }
+        this.textEgressBuffers.set(client.id, '');
+        this.thinkingEgressBuffers.set(client.id, '');
+        this.overflowedEgress.delete(this.egressKey(client, 'agent:text'));
+        this.overflowedEgress.delete(this.egressKey(client, 'agent:thinking'));
         client.emit('agent:start', { conversationId });
         break;
       }
 
       case 'agent_end': {
         // Gather usage stats from the Pi session
-        const agentSession = this.agentService.getSession(conversationId);
+        const activeClientSession = this.clientSessions.get(client.id);
+        const agentSession = activeClientSession
+          ? this.agentService.getSession(conversationId, activeClientSession.scope)
+          : undefined;
         const piSession = agentSession?.piSession;
         const stats = piSession?.getSessionStats();
         const contextUsage = piSession?.getContextUsage();
@@ -569,6 +1142,20 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
             }
           : undefined;
 
+        this.flushRedactedEgress(
+          client,
+          conversationId,
+          'agent:text',
+          this.textEgressBuffers,
+          true,
+        );
+        this.flushRedactedEgress(
+          client,
+          conversationId,
+          'agent:thinking',
+          this.thinkingEgressBuffers,
+          true,
+        );
         client.emit('agent:end', {
           conversationId,
           usage: usagePayload,
@@ -611,8 +1198,11 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
               {
                 conversationId,
                 role: 'assistant',
-                content: cs.assistantText,
-                metadata,
+                content: redactSensitiveContent(cs.assistantText).content,
+                metadata: {
+                  ...metadata,
+                  classifications: redactSensitiveContent(cs.assistantText).classifications,
+                },
               },
               userId,
             )
@@ -634,20 +1224,26 @@ export class ChatGateway implements OnGatewayInit, OnGatewayConnection, OnGatewa
       case 'message_update': {
         const assistantEvent = event.assistantMessageEvent;
         if (assistantEvent.type === 'text_delta') {
-          // Accumulate assistant text for persistence
+          // Keep raw stream material in memory only; persist and emit only redacted text.
           const cs = this.clientSessions.get(client.id);
           if (cs) {
             cs.assistantText += assistantEvent.delta;
           }
-          client.emit('agent:text', {
+          this.appendAndFlushRedactedEgress(
+            client,
             conversationId,
-            text: assistantEvent.delta,
-          });
+            'agent:text',
+            this.textEgressBuffers,
+            assistantEvent.delta,
+          );
         } else if (assistantEvent.type === 'thinking_delta') {
-          client.emit('agent:thinking', {
+          this.appendAndFlushRedactedEgress(
+            client,
             conversationId,
-            text: assistantEvent.delta,
-          });
+            'agent:thinking',
+            this.thinkingEgressBuffers,
+            assistantEvent.delta,
+          );
         }
         break;
       }

apps/gateway/src/commands/command-authorization.service.spec.ts

@@ -0,0 +1,116 @@
+import { describe, expect, it } from 'vitest';
+import type { CommandDef, SlashCommandPayload } from '@mosaicstack/types';
+import { CommandAuthorizationService } from './command-authorization.service.js';
+
+const adminCommand: CommandDef = {
+  name: 'gc',
+  description: 'GC',
+  aliases: [],
+  scope: 'admin',
+  execution: 'socket',
+  available: true,
+};
+const payload: SlashCommandPayload = { command: 'gc', conversationId: 'conversation-1' };
+
+function createService(
+  role: string,
+  entries: Map<string, string> = new Map<string, string>(),
+): CommandAuthorizationService {
+  const db = {
+    select: () => ({ from: () => ({ where: () => ({ limit: async () => [{ role }] }) }) }),
+  };
+  const redis = {
+    get: async (key: string) => entries.get(key) ?? null,
+    set: async (key: string, value: string) => {
+      entries.set(key, value);
+    },
+    del: async (key: string) => Number(entries.delete(key)),
+  };
+  return new CommandAuthorizationService(db as never, redis);
+}
+
+describe('CommandAuthorizationService', () => {
+  it('consumes one exact actor-bound approval once', async (): Promise<void> => {
+    const service = createService('admin');
+    const approval = await service.createApproval(adminCommand, payload, 'admin-1');
+    expect(approval).not.toBeNull();
+    expect(
+      (await service.authorize(adminCommand, payload, 'admin-1', approval!.approvalId)).allowed,
+    ).toBe(true);
+    expect(
+      (await service.authorize(adminCommand, payload, 'admin-1', approval!.approvalId)).allowed,
+    ).toBe(false);
+  });
+
+  it('rejects an approval when the structured action is mutated', async (): Promise<void> => {
+    const service = createService('admin');
+    const approval = await service.createApproval(adminCommand, payload, 'admin-1');
+    const mutated = { ...payload, conversationId: 'other-conversation' };
+    expect(approval).not.toBeNull();
+    expect(
+      (await service.authorize(adminCommand, mutated, 'admin-1', approval!.approvalId)).allowed,
+    ).toBe(false);
+  });
+
+  it('denies an admin command to a member before approval is considered', async (): Promise<void> => {
+    const service = createService('member');
+    const approval = await service.createApproval(adminCommand, payload, 'member-1');
+    expect(approval).toBeNull();
+    expect(
+      (await service.authorize(adminCommand, payload, 'member-1', 'forged-approval-id')).allowed,
+    ).toBe(false);
+  });
+
+  it('denies a malformed durable approval expiry instead of treating it as unexpired', async (): Promise<void> => {
+    const entries = new Map<string, string>();
+    const action = {
+      providerId: 'fleet',
+      sessionId: 'nova',
+      actorId: 'admin-1',
+      tenantId: 'tenant-1',
+      channelId: 'discord:operator',
+      correlationId: 'correlation-malformed-expiry',
+      agentName: 'Nova',
+    };
+    const service = createService('admin', entries);
+    const approval = await service.createRuntimeTerminationApproval(action);
+    expect(approval).not.toBeNull();
+    const key = `agent:Nova:command-approval:${approval!.approvalId}`;
+    const stored = entries.get(key);
+    expect(stored).toBeDefined();
+    entries.set(key, JSON.stringify({ ...JSON.parse(stored!), expiresAt: 'not-a-date' }));
+
+    expect(await service.consumeRuntimeTerminationApproval(approval!.approvalId, action)).toBe(
+      false,
+    );
+  });
+
+  it('persists and consumes one exact runtime termination approval across a service restart', async (): Promise<void> => {
+    const entries = new Map<string, string>();
+    const action = {
+      providerId: 'fleet',
+      sessionId: 'nova',
+      actorId: 'admin-1',
+      tenantId: 'tenant-1',
+      channelId: 'discord:operator',
+      correlationId: 'correlation-1',
+      agentName: 'Nova',
+    };
+    const beforeRestart = createService('admin', entries);
+    const approval = await beforeRestart.createRuntimeTerminationApproval(action);
+
+    const afterRestart = createService('admin', entries);
+    expect(
+      await afterRestart.consumeRuntimeTerminationApproval(approval!.approvalId, {
+        ...action,
+        sessionId: 'forged-session',
+      }),
+    ).toBe(false);
+    expect(await afterRestart.consumeRuntimeTerminationApproval(approval!.approvalId, action)).toBe(
+      true,
+    );
+    expect(await afterRestart.consumeRuntimeTerminationApproval(approval!.approvalId, action)).toBe(
+      false,
+    );
+  });
+});

apps/gateway/src/commands/command-authorization.service.ts

@@ -0,0 +1,268 @@
+import { createHash, randomUUID } from 'node:crypto';
+import { Inject, Injectable } from '@nestjs/common';
+import { eq, users as usersTable, type Db } from '@mosaicstack/db';
+import type { CommandDef, SlashCommandPayload } from '@mosaicstack/types';
+import { DB } from '../database/database.module.js';
+import { COMMANDS_REDIS } from './commands.tokens.js';
+
+export type CommandRole = 'admin' | 'member' | 'viewer';
+
+export interface CommandApproval {
+  approvalId: string;
+  actionDigest: string;
+  actorId: string;
+  command: string;
+  expiresAt: string;
+}
+
+/** Exact immutable binding for a privileged runtime termination. */
+export interface RuntimeTerminationApprovalAction {
+  providerId: string;
+  sessionId: string;
+  actorId: string;
+  tenantId: string;
+  channelId: string;
+  correlationId: string;
+  /** Provisioned roster identity; isolates approvals between interaction agents. */
+  agentName: string;
+}
+
+export interface RuntimeTerminationApproval extends RuntimeTerminationApprovalAction {
+  approvalId: string;
+  actionDigest: string;
+  expiresAt: string;
+}
+
+export interface CommandAuthorizationResult {
+  allowed: boolean;
+  reason?: string;
+}
+
+@Injectable()
+export class CommandAuthorizationService {
+  constructor(
+    @Inject(DB) private readonly db: Db,
+    @Inject(COMMANDS_REDIS)
+    private readonly redis: {
+      get(key: string): Promise<string | null>;
+      set(key: string, value: string, ...args: string[]): Promise<unknown>;
+      del(key: string): Promise<number>;
+    },
+  ) {}
+
+  async authorize(
+    command: CommandDef,
+    payload: SlashCommandPayload,
+    actorId: string,
+    approvalId?: string,
+  ): Promise<CommandAuthorizationResult> {
+    const role = await this.resolveRole(actorId);
+    if (!role || !this.hasScope(role, command.scope)) {
+      return { allowed: false, reason: 'not authorized for this command scope' };
+    }
+    if (command.scope !== 'admin') return { allowed: true };
+    if (!approvalId) return { allowed: false, reason: 'durable approval is required' };
+    const actionDigest = this.actionDigest(command.name, payload);
+    const approved = await this.consumeApproval(approvalId, actorId, actionDigest);
+    return approved
+      ? { allowed: true }
+      : {
+          allowed: false,
+          reason: 'approval is invalid, expired, replayed, or does not match this action',
+        };
+  }
+
+  async createApproval(
+    command: CommandDef,
+    payload: SlashCommandPayload,
+    actorId: string,
+  ): Promise<CommandApproval | null> {
+    const role = await this.resolveRole(actorId);
+    if (!role || command.scope !== 'admin' || !this.hasScope(role, command.scope)) return null;
+
+    const approvalId = randomUUID();
+    const expiresAt = new Date(Date.now() + 5 * 60_000).toISOString();
+    const approval: CommandApproval = {
+      approvalId,
+      actionDigest: this.actionDigest(command.name, payload),
+      actorId,
+      command: command.name,
+      expiresAt,
+    };
+    await this.redis.set(this.key(approvalId), JSON.stringify(approval), 'EX', '300');
+    return approval;
+  }
+
+  /**
+   * Uses the same `interaction:command-approval:*` store and one-time deletion rule as
+   * command approvals. This deliberately avoids a parallel approval database.
+   */
+  async createRuntimeTerminationApproval(
+    action: RuntimeTerminationApprovalAction,
+  ): Promise<RuntimeTerminationApproval | null> {
+    if (!this.hasRuntimeTerminationAction(action)) return null;
+    const role = await this.resolveRole(action.actorId);
+    if (role !== 'admin') return null;
+
+    const approval: RuntimeTerminationApproval = {
+      approvalId: randomUUID(),
+      actionDigest: this.runtimeActionDigest(action),
+      ...action,
+      expiresAt: new Date(Date.now() + 5 * 60_000).toISOString(),
+    };
+    await this.redis.set(
+      this.runtimeKey(action.agentName, approval.approvalId),
+      JSON.stringify(approval),
+      'EX',
+      '300',
+    );
+    return approval;
+  }
+
+  async consumeRuntimeTerminationApproval(
+    approvalId: string,
+    action: RuntimeTerminationApprovalAction,
+  ): Promise<boolean> {
+    const encoded = await this.redis.get(this.runtimeKey(action.agentName, approvalId));
+    if (!encoded) return false;
+    let approval: unknown;
+    try {
+      approval = JSON.parse(encoded);
+    } catch {
+      return false;
+    }
+    if (
+      !this.isRuntimeTerminationApproval(approval) ||
+      approval.actionDigest !== this.runtimeActionDigest(action) ||
+      approval.actorId !== action.actorId ||
+      approval.tenantId !== action.tenantId ||
+      !this.isUnexpired(approval.expiresAt)
+    ) {
+      return false;
+    }
+    if ((await this.resolveRole(approval.actorId)) !== 'admin') return false;
+    return (await this.redis.del(this.runtimeKey(action.agentName, approvalId))) === 1;
+  }
+
+  private async resolveRole(actorId: string): Promise<CommandRole | null> {
+    const [user] = await this.db
+      .select({ role: usersTable.role })
+      .from(usersTable)
+      .where(eq(usersTable.id, actorId))
+      .limit(1);
+    const role = user?.role;
+    return role === 'admin' || role === 'member' || role === 'viewer' ? role : null;
+  }
+
+  private hasScope(role: CommandRole, scope: CommandDef['scope']): boolean {
+    if (role === 'admin') return true;
+    return role === 'member' && (scope === 'core' || scope === 'agent');
+  }
+
+  private async consumeApproval(
+    approvalId: string,
+    actorId: string,
+    actionDigest: string,
+  ): Promise<boolean> {
+    const key = this.key(approvalId);
+    const encoded = await this.redis.get(key);
+    if (!encoded) return false;
+    let parsed: unknown;
+    try {
+      parsed = JSON.parse(encoded);
+    } catch {
+      return false;
+    }
+    if (
+      !this.isCommandApproval(parsed) ||
+      parsed.actorId !== actorId ||
+      parsed.actionDigest !== actionDigest ||
+      !this.isUnexpired(parsed.expiresAt)
+    )
+      return false;
+    return (await this.redis.del(key)) === 1;
+  }
+
+  private actionDigest(command: string, payload: SlashCommandPayload): string {
+    return createHash('sha256')
+      .update(
+        JSON.stringify({
+          command,
+          args: payload.args?.trim() ?? '',
+          conversationId: payload.conversationId,
+        }),
+      )
+      .digest('hex');
+  }
+
+  private hasRuntimeTerminationAction(action: RuntimeTerminationApprovalAction): boolean {
+    return [
+      action.providerId,
+      action.sessionId,
+      action.actorId,
+      action.tenantId,
+      action.channelId,
+      action.correlationId,
+      action.agentName,
+    ].every((value: string): boolean => value.trim().length > 0);
+  }
+
+  private runtimeActionDigest(action: RuntimeTerminationApprovalAction): string {
+    return createHash('sha256')
+      .update(
+        JSON.stringify({
+          providerId: action.providerId,
+          sessionId: action.sessionId,
+          actorId: action.actorId,
+          tenantId: action.tenantId,
+          channelId: action.channelId,
+          correlationId: action.correlationId,
+          agentName: action.agentName,
+        }),
+      )
+      .digest('hex');
+  }
+
+  private isUnexpired(expiresAt: unknown): expiresAt is string {
+    if (typeof expiresAt !== 'string') return false;
+    const expiresAtMs = Date.parse(expiresAt);
+    return Number.isFinite(expiresAtMs) && expiresAtMs > Date.now();
+  }
+
+  private isCommandApproval(value: unknown): value is CommandApproval {
+    return (
+      typeof value === 'object' &&
+      value !== null &&
+      'approvalId' in value &&
+      'actionDigest' in value &&
+      'actorId' in value &&
+      'expiresAt' in value &&
+      'command' in value
+    );
+  }
+
+  private isRuntimeTerminationApproval(value: unknown): value is RuntimeTerminationApproval {
+    return (
+      typeof value === 'object' &&
+      value !== null &&
+      'approvalId' in value &&
+      'actionDigest' in value &&
+      'actorId' in value &&
+      'tenantId' in value &&
+      'providerId' in value &&
+      'sessionId' in value &&
+      'channelId' in value &&
+      'correlationId' in value &&
+      'agentName' in value &&
+      'expiresAt' in value
+    );
+  }
+
+  private key(approvalId: string): string {
+    return `interaction:command-approval:${approvalId}`;
+  }
+
+  private runtimeKey(agentName: string, approvalId: string): string {
+    return `agent:${encodeURIComponent(agentName)}:command-approval:${approvalId}`;
+  }
+}

apps/gateway/src/commands/command-executor-p8012.spec.ts

@@ -89,6 +89,7 @@ function buildService(): CommandExecutorService {
 describe('CommandExecutorService — P8-012 commands', () => {
   let service: CommandExecutorService;
   const userId = 'user-123';
+  const userScope = { userId, tenantId: userId };
   const conversationId = 'conv-456';
 
   beforeEach(() => {
@@ -99,31 +100,26 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /provider login — missing provider name
   it('/provider login with no provider name returns usage error', async () => {
     const payload: SlashCommandPayload = { command: 'provider', args: 'login', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(false);
     expect(result.message).toContain('Usage: /provider login');
     expect(result.command).toBe('provider');
   });
 
-  // /provider login anthropic — success with URL containing poll token
-  it('/provider login <name> returns success with URL and poll token', async () => {
+  // /provider login anthropic — no bearer token or auth URL reaches chat output
+  it('/provider login <name> keeps its one-time token out of chat output', async () => {
     const payload: SlashCommandPayload = {
       command: 'provider',
       args: 'login anthropic',
       conversationId,
     };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('provider');
     expect(result.message).toContain('anthropic');
-    expect(result.message).toContain('http');
-    // data should contain loginUrl and pollToken
-    expect(result.data).toBeDefined();
-    const data = result.data as Record<string, unknown>;
-    expect(typeof data['loginUrl']).toBe('string');
-    expect(typeof data['pollToken']).toBe('string');
-    expect(data['loginUrl'] as string).toContain('anthropic');
-    expect(data['loginUrl'] as string).toContain(data['pollToken'] as string);
+    expect(result.message).not.toContain('http');
+    expect(result.message).not.toContain('token=');
+    expect(result.data).toEqual({ provider: 'anthropic' });
     // Verify Valkey was called
     expect(mockRedis.set).toHaveBeenCalledOnce();
     const [key, value, , ttl] = mockRedis.set.mock.calls[0] as [string, string, string, number];
@@ -138,7 +134,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /provider with no args — returns usage
   it('/provider with no args returns usage message', async () => {
     const payload: SlashCommandPayload = { command: 'provider', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('Usage: /provider');
   });
@@ -146,7 +142,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /provider list
   it('/provider list returns success', async () => {
     const payload: SlashCommandPayload = { command: 'provider', args: 'list', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('provider');
   });
@@ -154,7 +150,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /provider logout with no name — usage error
   it('/provider logout with no name returns error', async () => {
     const payload: SlashCommandPayload = { command: 'provider', args: 'logout', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(false);
     expect(result.message).toContain('Usage: /provider logout');
   });
@@ -166,7 +162,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
       args: 'unknown',
       conversationId,
     };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(false);
     expect(result.message).toContain('Unknown subcommand');
   });
@@ -174,7 +170,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /mission status
   it('/mission status returns stub message', async () => {
     const payload: SlashCommandPayload = { command: 'mission', args: 'status', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('mission');
     expect(result.message).toContain('Mission status');
@@ -183,7 +179,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /mission with no args
   it('/mission with no args returns status stub', async () => {
     const payload: SlashCommandPayload = { command: 'mission', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('Mission status');
   });
@@ -195,7 +191,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
       args: 'set my-mission-123',
       conversationId,
     };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('my-mission-123');
   });
@@ -203,7 +199,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /agent list
   it('/agent list returns stub message', async () => {
     const payload: SlashCommandPayload = { command: 'agent', args: 'list', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('agent');
     expect(result.message).toContain('agent');
@@ -212,7 +208,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /agent with no args
   it('/agent with no args returns usage', async () => {
     const payload: SlashCommandPayload = { command: 'agent', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('Usage: /agent');
   });
@@ -224,7 +220,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
       args: 'my-agent-id',
       conversationId,
     };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('my-agent-id');
   });
@@ -232,7 +228,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /prdy
   it('/prdy returns PRD wizard message', async () => {
     const payload: SlashCommandPayload = { command: 'prdy', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('prdy');
     expect(result.message).toContain('mosaic prdy');
@@ -241,7 +237,7 @@ describe('CommandExecutorService — P8-012 commands', () => {
   // /tools
   it('/tools returns tools stub message', async () => {
     const payload: SlashCommandPayload = { command: 'tools', conversationId };
-    const result = await service.execute(payload, userId);
+    const result = await service.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('tools');
     expect(result.message).toContain('tools');

apps/gateway/src/commands/command-executor-tess-security.spec.ts

@@ -0,0 +1,112 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest';
+import type { SlashCommandPayload } from '@mosaicstack/types';
+import { CommandAuthorizationService } from './command-authorization.service.js';
+import { CommandExecutorService } from './command-executor.service.js';
+
+const registry = {
+  getManifest: vi.fn(() => ({
+    version: 1,
+    commands: [
+      {
+        name: 'gc',
+        description: 'System-wide garbage collection',
+        aliases: [],
+        scope: 'admin' as const,
+        execution: 'socket' as const,
+        available: true,
+      },
+    ],
+    skills: [],
+  })),
+};
+
+const sessionGc = {
+  sweepOrphans: vi.fn().mockResolvedValue({ orphanedSessions: 1, totalCleaned: [], duration: 1 }),
+};
+
+const scope = (userId: string) => ({ userId, tenantId: 'tenant-1' });
+
+const authorization = {
+  authorize: vi.fn((_command: unknown, _payload: unknown, actorId: string) =>
+    Promise.resolve(
+      actorId === 'member-1'
+        ? { allowed: false, reason: 'durable approval is required' }
+        : { allowed: false, reason: 'not authorized for this command scope' },
+    ),
+  ),
+};
+
+function buildExecutor(authorizationService: unknown = authorization): CommandExecutorService {
+  return new CommandExecutorService(
+    registry as never,
+    { getSession: vi.fn() } as never,
+    { clear: vi.fn(), set: vi.fn() } as never,
+    sessionGc as never,
+    { set: vi.fn() } as never,
+    { agents: {} } as never,
+    null,
+    null,
+    null,
+    authorizationService as never,
+  );
+}
+
+function createDurableAuthorization(): CommandAuthorizationService {
+  const entries = new Map<string, string>();
+  const db = {
+    select: () => ({ from: () => ({ where: () => ({ limit: async () => [{ role: 'admin' }] }) }) }),
+  };
+  const redis = {
+    get: async (key: string) => entries.get(key) ?? null,
+    set: async (key: string, value: string) => {
+      entries.set(key, value);
+    },
+    del: async (key: string) => Number(entries.delete(key)),
+  };
+  return new CommandAuthorizationService(db as never, redis);
+}
+
+describe('TESS-M1-SEC-001 command authorization abuse cases', () => {
+  const payload: SlashCommandPayload = { command: 'gc', conversationId: 'conversation-1' };
+
+  beforeEach((): void => {
+    vi.clearAllMocks();
+  });
+
+  it('denies a forged admin identity and does not execute a system-wide command', async (): Promise<void> => {
+    const result = await buildExecutor().execute(payload, scope('admin-forged-by-client'));
+
+    expect(result.success).toBe(false);
+    expect(result.message).toContain('not authorized');
+    expect(sessionGc.sweepOrphans).not.toHaveBeenCalled();
+  });
+
+  it('denies a privileged command without a server-bound durable approval', async (): Promise<void> => {
+    const result = await buildExecutor().execute(payload, scope('member-1'));
+
+    expect(result.success).toBe(false);
+    expect(result.message).toContain('approval');
+    expect(sessionGc.sweepOrphans).not.toHaveBeenCalled();
+  });
+
+  it('executes an admin command only after a valid durable approval is issued and supplied', async (): Promise<void> => {
+    const executor = buildExecutor(createDurableAuthorization());
+
+    const adminScope = scope('admin-1');
+    const denied = await executor.execute(payload, adminScope);
+    const approval = await executor.createApproval(payload, adminScope);
+    const approved = await executor.execute(
+      { ...payload, approvalId: approval?.approvalId },
+      adminScope,
+    );
+
+    expect(denied.success).toBe(false);
+    expect(denied.message).toContain('approval');
+    expect(approval).not.toBeNull();
+    // A valid durable approval is consumed, but cannot authorize an unimplemented
+    // global retention operation. Session-scoped cleanup remains lifecycle-only.
+    expect(approved.success).toBe(false);
+    expect(approved.message).toContain('Global GC is disabled');
+    expect(sessionGc.sweepOrphans).not.toHaveBeenCalled();
+  });
+});

apps/gateway/src/commands/command-executor.service.ts

@@ -3,6 +3,7 @@ import type { QueueHandle } from '@mosaicstack/queue';
 import type { Brain } from '@mosaicstack/brain';
 import type { SlashCommandPayload, SlashCommandResultPayload } from '@mosaicstack/types';
 import { AgentService } from '../agent/agent.service.js';
+import type { ActorTenantScope } from '../auth/session-scope.js';
 import { ChatGateway } from '../chat/chat.gateway.js';
 import { SessionGCService } from '../gc/session-gc.service.js';
 import { SystemOverrideService } from '../preferences/system-override.service.js';
@@ -10,6 +11,7 @@ import { ReloadService } from '../reload/reload.service.js';
 import { McpClientService } from '../mcp-client/mcp-client.service.js';
 import { BRAIN } from '../brain/brain.tokens.js';
 import { COMMANDS_REDIS } from './commands.tokens.js';
+import { CommandAuthorizationService } from './command-authorization.service.js';
 import { CommandRegistryService } from './command-registry.service.js';
 
 @Injectable()
@@ -32,10 +34,17 @@ export class CommandExecutorService {
     @Optional()
     @Inject(McpClientService)
     private readonly mcpClient: McpClientService | null,
+    @Optional()
+    @Inject(CommandAuthorizationService)
+    private readonly authorization: CommandAuthorizationService | null = null,
   ) {}
 
-  async execute(payload: SlashCommandPayload, userId: string): Promise<SlashCommandResultPayload> {
+  async execute(
+    payload: SlashCommandPayload,
+    scope: ActorTenantScope,
+  ): Promise<SlashCommandResultPayload> {
     const { command, args, conversationId } = payload;
+    const userId = scope.userId;
 
     const def = this.registry.getManifest().commands.find((c) => c.name === command);
     if (!def) {
@@ -47,14 +56,24 @@ export class CommandExecutorService {
       };
     }
 
+    const authorization = await this.authorization?.authorize(
+      def,
+      payload,
+      userId,
+      payload.approvalId,
+    );
+    if (authorization && !authorization.allowed) {
+      return { command, conversationId, success: false, message: authorization.reason };
+    }
+
     try {
       switch (command) {
         case 'model':
-          return await this.handleModel(args ?? null, conversationId);
+          return await this.handleModel(args ?? null, conversationId, scope);
         case 'thinking':
           return await this.handleThinking(args ?? null, conversationId);
         case 'system':
-          return await this.handleSystem(args ?? null, conversationId);
+          return await this.handleSystem(args ?? null, conversationId, scope);
         case 'new':
           return {
             command,
@@ -83,18 +102,17 @@ export class CommandExecutorService {
             success: true,
             message: 'Retry last message requested.',
           };
-        case 'gc': {
-          // Admin-only: system-wide GC sweep across all sessions
-          const result = await this.sessionGC.sweepOrphans();
+        case 'gc':
+          // Global retention requires a separate, authorized and audited job.
+          // Session cleanup is performed only through the session lifecycle.
           return {
             command: 'gc',
-            success: true,
-            message: `GC sweep complete: ${result.orphanedSessions} orphaned sessions cleaned in ${result.duration}ms.`,
+            success: false,
+            message: 'Global GC is disabled pending an authorized retention job.',
             conversationId,
           };
-        }
         case 'agent':
-          return await this.handleAgent(args ?? null, conversationId, userId);
+          return await this.handleAgent(args ?? null, conversationId, scope);
         case 'provider':
           return await this.handleProvider(args ?? null, userId, conversationId);
         case 'mission':
@@ -143,13 +161,22 @@ export class CommandExecutorService {
     }
   }
 
+  async createApproval(payload: SlashCommandPayload, scope: ActorTenantScope) {
+    const def = this.registry
+      .getManifest()
+      .commands.find((command) => command.name === payload.command);
+    if (!def || !this.authorization) return null;
+    return this.authorization.createApproval(def, payload, scope.userId);
+  }
+
   private async handleModel(
     args: string | null,
     conversationId: string,
+    scope: ActorTenantScope,
   ): Promise<SlashCommandResultPayload> {
     if (!args || args.trim().length === 0) {
       // Show current override or usage hint
-      const currentOverride = this.chatGateway?.getModelOverride(conversationId);
+      const currentOverride = this.chatGateway?.getModelOverride(conversationId, scope);
       if (currentOverride) {
         return {
           command: 'model',
@@ -171,7 +198,7 @@ export class CommandExecutorService {
 
     // /model clear removes the override and re-enables automatic routing
     if (modelName === 'clear') {
-      this.chatGateway?.setModelOverride(conversationId, null);
+      this.chatGateway?.setModelOverride(conversationId, null, scope);
       return {
         command: 'model',
         conversationId,
@@ -181,9 +208,9 @@ export class CommandExecutorService {
     }
 
     // Set the sticky per-session override (M4-007)
-    this.chatGateway?.setModelOverride(conversationId, modelName);
+    this.chatGateway?.setModelOverride(conversationId, modelName, scope);
 
-    const session = this.agentService.getSession(conversationId);
+    const session = this.agentService.getSession(conversationId, scope);
     if (!session) {
       return {
         command: 'model',
@@ -224,10 +251,11 @@ export class CommandExecutorService {
   private async handleSystem(
     args: string | null,
     conversationId: string,
+    scope: ActorTenantScope,
   ): Promise<SlashCommandResultPayload> {
     if (!args || args.trim().length === 0) {
       // Clear the override when called with no args
-      await this.systemOverride.clear(conversationId);
+      await this.systemOverride.clear(conversationId, scope);
       return {
         command: 'system',
         conversationId,
@@ -236,7 +264,7 @@ export class CommandExecutorService {
       };
     }
 
-    await this.systemOverride.set(conversationId, args.trim());
+    await this.systemOverride.set(conversationId, args.trim(), scope);
     return {
       command: 'system',
       conversationId,
@@ -248,8 +276,9 @@ export class CommandExecutorService {
   private async handleAgent(
     args: string | null,
     conversationId: string,
-    userId: string,
+    scope: ActorTenantScope,
   ): Promise<SlashCommandResultPayload> {
+    const userId = scope.userId;
     if (!args) {
       return {
         command: 'agent',
@@ -338,11 +367,14 @@ export class CommandExecutorService {
         conversationId,
         agentConfig.id,
         agentConfig.name,
+        scope,
         agentConfig.model ?? undefined,
       );
 
       // Broadcast updated session:info so TUI TopBar reflects new agent/model
-      this.chatGateway?.broadcastSessionInfo(conversationId, { agentName: agentConfig.name });
+      this.chatGateway?.broadcastSessionInfo(conversationId, scope, {
+        agentName: agentConfig.name,
+      });
 
       this.logger.log(
         `Agent switched to "${agentConfig.name}" (${agentConfig.id}) for conversation ${conversationId} (M5-003)`,
@@ -403,22 +435,28 @@ export class CommandExecutorService {
           };
         }
         const pollToken = crypto.randomUUID();
-        const key = `mosaic:auth:poll:${pollToken}`;
-        // Store pending state in Valkey (TTL 5 minutes)
+        const tokenDigest = await crypto.subtle.digest(
+          'SHA-256',
+          new TextEncoder().encode(pollToken),
+        );
+        const tokenHash = Array.from(new Uint8Array(tokenDigest), (byte: number): string =>
+          byte.toString(16).padStart(2, '0'),
+        ).join('');
+        const key = `mosaic:auth:poll:${tokenHash}`;
+        // Persist only a short-lived token digest. The raw token is delivered only by
+        // the authenticated dashboard flow, never in chat output or command metadata.
         await this.redis.set(
           key,
           JSON.stringify({ status: 'pending', provider: providerName, userId }),
           'EX',
           300,
         );
-        // In production this would construct an OAuth URL
-        const loginUrl = `${process.env['MOSAIC_BASE_URL'] ?? 'http://localhost:3000'}/auth/provider/${providerName}?token=${pollToken}`;
         return {
           command: 'provider',
           success: true,
-          message: `Open this URL to authenticate with ${providerName}:\n${loginUrl}`,
+          message: `Provider login for ${providerName} is ready. Continue in the authenticated dashboard.`,
           conversationId,
-          data: { loginUrl, pollToken, provider: providerName },
+          data: { provider: providerName },
         };
       }
 

apps/gateway/src/commands/commands.integration.spec.ts

@@ -159,6 +159,7 @@ describe('CommandExecutorService — integration', () => {
   let registry: CommandRegistryService;
   let executor: CommandExecutorService;
   const userId = 'user-integ-001';
+  const userScope = { userId, tenantId: userId };
   const conversationId = 'conv-integ-001';
 
   beforeEach(() => {
@@ -170,28 +171,26 @@ describe('CommandExecutorService — integration', () => {
   // Unknown command returns error
   it('unknown command returns success:false with descriptive message', async () => {
     const payload: SlashCommandPayload = { command: 'nonexistent', conversationId };
-    const result = await executor.execute(payload, userId);
+    const result = await executor.execute(payload, userScope);
     expect(result.success).toBe(false);
     expect(result.message).toContain('nonexistent');
     expect(result.command).toBe('nonexistent');
   });
 
-  // /gc handler calls SessionGCService.sweepOrphans (admin-only, no userId arg)
-  it('/gc calls SessionGCService.sweepOrphans without arguments', async () => {
+  it('/gc refuses an unaudited global sweep', async () => {
     const payload: SlashCommandPayload = { command: 'gc', conversationId };
-    const result = await executor.execute(payload, userId);
-    expect(mockSessionGC.sweepOrphans).toHaveBeenCalledWith();
-    expect(result.success).toBe(true);
-    expect(result.message).toContain('GC sweep complete');
-    expect(result.message).toContain('3 orphaned sessions');
+    const result = await executor.execute(payload, userScope);
+    expect(mockSessionGC.sweepOrphans).not.toHaveBeenCalled();
+    expect(result.success).toBe(false);
+    expect(result.message).toContain('disabled pending an authorized retention job');
   });
 
   // /system with args calls SystemOverrideService.set
   it('/system with text calls SystemOverrideService.set', async () => {
     const override = 'You are a helpful assistant.';
     const payload: SlashCommandPayload = { command: 'system', args: override, conversationId };
-    const result = await executor.execute(payload, userId);
-    expect(mockSystemOverride.set).toHaveBeenCalledWith(conversationId, override);
+    const result = await executor.execute(payload, userScope);
+    expect(mockSystemOverride.set).toHaveBeenCalledWith(conversationId, override, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('override set');
   });
@@ -199,8 +198,8 @@ describe('CommandExecutorService — integration', () => {
   // /system with no args clears the override
   it('/system with no args calls SystemOverrideService.clear', async () => {
     const payload: SlashCommandPayload = { command: 'system', conversationId };
-    const result = await executor.execute(payload, userId);
-    expect(mockSystemOverride.clear).toHaveBeenCalledWith(conversationId);
+    const result = await executor.execute(payload, userScope);
+    expect(mockSystemOverride.clear).toHaveBeenCalledWith(conversationId, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('cleared');
   });
@@ -212,7 +211,7 @@ describe('CommandExecutorService — integration', () => {
       args: 'claude-3-opus',
       conversationId,
     };
-    const result = await executor.execute(payload, userId);
+    const result = await executor.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('model');
     expect(result.message).toContain('claude-3-opus');
@@ -221,7 +220,7 @@ describe('CommandExecutorService — integration', () => {
   // /thinking with valid level returns success
   it('/thinking with valid level returns success', async () => {
     const payload: SlashCommandPayload = { command: 'thinking', args: 'high', conversationId };
-    const result = await executor.execute(payload, userId);
+    const result = await executor.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('high');
   });
@@ -229,7 +228,7 @@ describe('CommandExecutorService — integration', () => {
   // /thinking with invalid level returns usage message
   it('/thinking with invalid level returns usage message', async () => {
     const payload: SlashCommandPayload = { command: 'thinking', args: 'invalid', conversationId };
-    const result = await executor.execute(payload, userId);
+    const result = await executor.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.message).toContain('Usage:');
   });
@@ -237,7 +236,7 @@ describe('CommandExecutorService — integration', () => {
   // /new command returns success
   it('/new returns success', async () => {
     const payload: SlashCommandPayload = { command: 'new', conversationId };
-    const result = await executor.execute(payload, userId);
+    const result = await executor.execute(payload, userScope);
     expect(result.success).toBe(true);
     expect(result.command).toBe('new');
   });
@@ -245,7 +244,7 @@ describe('CommandExecutorService — integration', () => {
   // /reload without reloadService returns failure
   it('/reload without ReloadService returns failure', async () => {
     const payload: SlashCommandPayload = { command: 'reload', conversationId };
-    const result = await executor.execute(payload, userId);
+    const result = await executor.execute(payload, userScope);
     expect(result.success).toBe(false);
     expect(result.message).toContain('ReloadService');
   });
@@ -255,7 +254,7 @@ describe('CommandExecutorService — integration', () => {
   for (const cmd of stubCommands) {
     it(`/${cmd} returns success (stub)`, async () => {
       const payload: SlashCommandPayload = { command: cmd, conversationId };
-      const result = await executor.execute(payload, userId);
+      const result = await executor.execute(payload, userScope);
       expect(result.success).toBe(true);
       expect(result.command).toBe(cmd);
     });

apps/gateway/src/commands/commands.module.ts

@@ -3,8 +3,10 @@ import { createQueue, type QueueHandle } from '@mosaicstack/queue';
 import { ChatModule } from '../chat/chat.module.js';
 import { GCModule } from '../gc/gc.module.js';
 import { ReloadModule } from '../reload/reload.module.js';
+import { CommandAuthorizationService } from './command-authorization.service.js';
 import { CommandExecutorService } from './command-executor.service.js';
 import { CommandRegistryService } from './command-registry.service.js';
+import { CommandRuntimeApprovalVerifier } from './runtime-approval-verifier.js';
 import { COMMANDS_REDIS } from './commands.tokens.js';
 
 const COMMANDS_QUEUE_HANDLE = 'COMMANDS_QUEUE_HANDLE';
@@ -24,9 +26,16 @@ const COMMANDS_QUEUE_HANDLE = 'COMMANDS_QUEUE_HANDLE';
       inject: [COMMANDS_QUEUE_HANDLE],
     },
     CommandRegistryService,
+    CommandAuthorizationService,
+    CommandRuntimeApprovalVerifier,
+    CommandExecutorService,
+  ],
+  exports: [
+    CommandRegistryService,
+    CommandAuthorizationService,
+    CommandRuntimeApprovalVerifier,
     CommandExecutorService,
   ],
-  exports: [CommandRegistryService, CommandExecutorService],
 })
 export class CommandsModule implements OnApplicationShutdown {
   constructor(@Inject(COMMANDS_QUEUE_HANDLE) private readonly handle: QueueHandle) {}

apps/gateway/src/commands/runtime-approval-verifier.ts

@@ -0,0 +1,23 @@
+import { Inject, Injectable } from '@nestjs/common';
+import type {
+  RuntimeApprovalVerifier,
+  RuntimeTerminationAction,
+} from '../agent/runtime-provider-registry.service.js';
+import { CommandAuthorizationService } from './command-authorization.service.js';
+
+/**
+ * Adapter from the provider registry's exact termination action to the shared,
+ * Redis-backed `interaction:command-approval:*` store. It has no separate approval
+ * persistence or replay semantics.
+ */
+@Injectable()
+export class CommandRuntimeApprovalVerifier implements RuntimeApprovalVerifier {
+  constructor(
+    @Inject(CommandAuthorizationService)
+    private readonly authorization: CommandAuthorizationService,
+  ) {}
+
+  async consume(approvalRef: string, action: RuntimeTerminationAction): Promise<boolean> {
+    return this.authorization.consumeRuntimeTerminationApproval(approvalRef, action);
+  }
+}

apps/gateway/src/coord/coord.module.ts

@@ -1,10 +1,32 @@
 import { Module } from '@nestjs/common';
+import { InMemoryInteractionCoordinationPort } from '@mosaicstack/coord';
 import { CoordService } from './coord.service.js';
 import { CoordController } from './coord.controller.js';
+import { InteractionCoordinationController } from './interaction-coordination.controller.js';
+import {
+  COORDINATION_CONFIG,
+  COORDINATION_PORT,
+  InteractionCoordinationService,
+} from './interaction-coordination.service.js';
 
 @Module({
-  providers: [CoordService],
-  controllers: [CoordController],
-  exports: [CoordService],
+  providers: [
+    CoordService,
+    {
+      provide: COORDINATION_PORT,
+      useFactory: (): InMemoryInteractionCoordinationPort =>
+        new InMemoryInteractionCoordinationPort(),
+    },
+    {
+      provide: COORDINATION_CONFIG,
+      useFactory: () => ({
+        interactionAgentId: process.env['MOSAIC_AGENT_NAME'],
+        orchestrationAgentId: process.env['MOSAIC_ORCHESTRATOR_AGENT_NAME'],
+      }),
+    },
+    InteractionCoordinationService,
+  ],
+  controllers: [CoordController, InteractionCoordinationController],
+  exports: [CoordService, InteractionCoordinationService],
 })
 export class CoordModule {}

apps/gateway/src/coord/interaction-coordination.controller.test.ts

@@ -0,0 +1,47 @@
+const PATH_METADATA = 'path';
+import { describe, expect, it, vi } from 'vitest';
+import { InteractionCoordinationController } from './interaction-coordination.controller.js';
+
+const user = { id: 'operator-1', tenantId: 'tenant-a' };
+
+describe('InteractionCoordinationController', () => {
+  it('exposes the neutral canonical route and Mos compatibility alias over identical handlers', () => {
+    expect(Reflect.getMetadata(PATH_METADATA, InteractionCoordinationController)).toEqual([
+      'api/coord/interaction',
+      'api/coord/mos',
+    ]);
+    expect(InteractionCoordinationController.prototype.handoff).toBeTypeOf('function');
+    expect(InteractionCoordinationController.prototype.observe).toBeTypeOf('function');
+    expect(InteractionCoordinationController.prototype.result).toBeTypeOf('function');
+  });
+
+  it('derives actor and tenant from the authenticated user rather than handoff input', async () => {
+    const coordination = {
+      handoff: vi.fn(async () => ({ handoffId: 'handoff-1' })),
+      observe: vi.fn(),
+      result: vi.fn(),
+    };
+    const controller = new InteractionCoordinationController(coordination as never);
+
+    await controller.handoff({ idempotencyKey: 'request-1', summary: 'Implement' }, user, 'corr-1');
+
+    expect(coordination.handoff).toHaveBeenCalledWith(
+      { idempotencyKey: 'request-1', summary: 'Implement' },
+      expect.objectContaining({
+        actorScope: { userId: 'operator-1', tenantId: 'tenant-a' },
+        channelId: 'cli',
+        correlationId: 'corr-1',
+      }),
+    );
+  });
+
+  it('requires a correlation header before invoking the coordination service', async () => {
+    const coordination = { handoff: vi.fn(), observe: vi.fn(), result: vi.fn() };
+    const controller = new InteractionCoordinationController(coordination as never);
+
+    await expect(
+      controller.handoff({ idempotencyKey: 'request-1', summary: 'Implement' }, user, undefined),
+    ).rejects.toThrow('X-Correlation-Id is required');
+    expect(coordination.handoff).not.toHaveBeenCalled();
+  });
+});

apps/gateway/src/coord/interaction-coordination.controller.ts

@@ -0,0 +1,75 @@
+import {
+  Body,
+  Controller,
+  ForbiddenException,
+  Get,
+  Headers,
+  Inject,
+  Param,
+  Post,
+  UseGuards,
+} from '@nestjs/common';
+import { AuthGuard } from '../auth/auth.guard.js';
+import { CurrentUser } from '../auth/current-user.decorator.js';
+import { scopeFromUser, type AuthenticatedUserLike } from '../auth/session-scope.js';
+import type { RuntimeProviderRequestContext } from '../agent/runtime-provider-registry.service.js';
+import type {
+  InteractionCoordinationObservationDto,
+  InteractionCoordinationResponseDto,
+  InteractionCoordinationResultDto,
+  CreateHandoffDto,
+} from './interaction-coordination.dto.js';
+import { InteractionCoordinationService } from './interaction-coordination.service.js';
+
+/** Authenticated interaction-plane boundary for the handoff/observe/result-only interaction coordination contract. */
+/** `api/coord/interaction` is canonical; the Mos path remains a compatibility alias. */
+@Controller(['api/coord/interaction', 'api/coord/mos'])
+@UseGuards(AuthGuard)
+export class InteractionCoordinationController {
+  constructor(
+    @Inject(InteractionCoordinationService)
+    private readonly coordination: InteractionCoordinationService,
+  ) {}
+
+  @Post('handoff')
+  async handoff(
+    @Body() request: CreateHandoffDto,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ): Promise<InteractionCoordinationResponseDto> {
+    return { receipt: await this.coordination.handoff(request, this.context(user, correlationId)) };
+  }
+
+  @Get(':handoffId/observe')
+  async observe(
+    @Param('handoffId') handoffId: string,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ): Promise<InteractionCoordinationObservationDto> {
+    return {
+      observation: await this.coordination.observe(handoffId, this.context(user, correlationId)),
+    };
+  }
+
+  @Get(':handoffId/result')
+  async result(
+    @Param('handoffId') handoffId: string,
+    @CurrentUser() user: AuthenticatedUserLike,
+    @Headers('x-correlation-id') correlationId?: string,
+  ): Promise<InteractionCoordinationResultDto> {
+    return { result: await this.coordination.result(handoffId, this.context(user, correlationId)) };
+  }
+
+  private context(
+    user: AuthenticatedUserLike,
+    correlationId?: string,
+  ): RuntimeProviderRequestContext {
+    const requestCorrelationId = correlationId?.trim();
+    if (!requestCorrelationId) throw new ForbiddenException('X-Correlation-Id is required');
+    return {
+      actorScope: scopeFromUser(user),
+      channelId: 'cli',
+      correlationId: requestCorrelationId,
+    };
+  }
+}

apps/gateway/src/coord/interaction-coordination.dto.ts

@@ -0,0 +1,25 @@
+import type {
+  CoordinationObservation,
+  CoordinationResult,
+  HandoffReceipt,
+} from '@mosaicstack/coord';
+
+/** Input accepted at the gateway coordination boundary. Agent identity is not caller-controlled. */
+export interface CreateHandoffDto {
+  idempotencyKey: string;
+  summary: string;
+  context?: string;
+  missionId?: string;
+}
+
+export interface InteractionCoordinationResponseDto {
+  receipt: HandoffReceipt;
+}
+
+export interface InteractionCoordinationObservationDto {
+  observation: CoordinationObservation;
+}
+
+export interface InteractionCoordinationResultDto {
+  result: CoordinationResult;
+}

apps/gateway/src/coord/interaction-coordination.routing.e2e.test.ts

@@ -0,0 +1,88 @@
+import 'reflect-metadata';
+import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest';
+import { Global, Module } from '@nestjs/common';
+import { Test } from '@nestjs/testing';
+import { FastifyAdapter, type NestFastifyApplication } from '@nestjs/platform-fastify';
+import { AUTH } from '../auth/auth.tokens.js';
+import { AuthGuard } from '../auth/auth.guard.js';
+import { InteractionCoordinationController } from './interaction-coordination.controller.js';
+import { InteractionCoordinationService } from './interaction-coordination.service.js';
+
+@Global()
+@Module({
+  providers: [
+    {
+      provide: AUTH,
+      useValue: {
+        api: {
+          getSession: vi.fn(async ({ headers }: { headers: Headers }) =>
+            headers.get('cookie') === 'session=trusted'
+              ? { user: { id: 'operator-1', tenantId: 'tenant-1' }, session: { id: 'session-1' } }
+              : null,
+          ),
+        },
+      },
+    },
+    AuthGuard,
+  ],
+  exports: [AUTH, AuthGuard],
+})
+class AuthenticatedRequestModule {}
+
+describe('InteractionCoordinationController route aliases', (): void => {
+  let app: NestFastifyApplication | undefined;
+  const coordination = {
+    handoff: vi.fn(async () => ({ handoffId: 'handoff-1' })),
+    observe: vi.fn(async () => ({ status: 'running' })),
+    result: vi.fn(async () => ({ status: 'completed' })),
+  };
+
+  beforeAll(async (): Promise<void> => {
+    const moduleRef = await Test.createTestingModule({
+      imports: [AuthenticatedRequestModule],
+      controllers: [InteractionCoordinationController],
+      providers: [{ provide: InteractionCoordinationService, useValue: coordination }],
+    }).compile();
+    app = moduleRef.createNestApplication<NestFastifyApplication>(new FastifyAdapter());
+    await app.init();
+    await app.getHttpAdapter().getInstance().ready();
+  });
+  afterAll(async (): Promise<void> => app?.close());
+
+  it('routes handoff, observe, and result through the same AuthGuard-protected service for both prefixes', async (): Promise<void> => {
+    if (!app) throw new Error('test app was not initialized');
+    for (const prefix of ['/api/coord/interaction', '/api/coord/mos']) {
+      const headers = { cookie: 'session=trusted', 'x-correlation-id': `corr-${prefix}` };
+      expect(
+        (
+          await app.inject({
+            method: 'POST',
+            url: `${prefix}/handoff`,
+            headers,
+            payload: { idempotencyKey: `key-${prefix}`, summary: 'handoff' },
+          })
+        ).statusCode,
+      ).toBe(201);
+      expect(
+        (await app.inject({ method: 'GET', url: `${prefix}/handoff-1/observe`, headers }))
+          .statusCode,
+      ).toBe(200);
+      expect(
+        (await app.inject({ method: 'GET', url: `${prefix}/handoff-1/result`, headers }))
+          .statusCode,
+      ).toBe(200);
+    }
+    expect(coordination.handoff).toHaveBeenCalledTimes(2);
+    expect(coordination.observe).toHaveBeenCalledTimes(2);
+    expect(coordination.result).toHaveBeenCalledTimes(2);
+    expect(
+      (
+        await app.inject({
+          method: 'POST',
+          url: '/api/coord/interaction/handoff',
+          payload: { idempotencyKey: 'denied', summary: 'x' },
+        })
+      ).statusCode,
+    ).toBe(401);
+  });
+});

apps/gateway/src/coord/interaction-coordination.service.test.ts

@@ -0,0 +1,217 @@
+import { describe, expect, it, vi } from 'vitest';
+import {
+  InMemoryInteractionCoordinationPort,
+  type InteractionCoordinationPort,
+  type Handoff,
+} from '@mosaicstack/coord';
+import type { RuntimeProviderRequestContext } from '../agent/runtime-provider-registry.service.js';
+import {
+  InteractionCoordinationService,
+  type InteractionCoordinationConfig,
+  type InteractionCoordinationGatewayError,
+} from './interaction-coordination.service.js';
+
+const context: RuntimeProviderRequestContext = {
+  actorScope: { userId: 'operator-1', tenantId: 'tenant-a' },
+  channelId: 'cli',
+  correlationId: 'corr-1',
+};
+
+const config: InteractionCoordinationConfig = {
+  interactionAgentId: 'Nova',
+  orchestrationAgentId: 'Conductor',
+};
+
+function service(
+  port: InteractionCoordinationPort = new InMemoryInteractionCoordinationPort(),
+  options: {
+    config?: InteractionCoordinationConfig;
+    handoffIdFactory?: () => string;
+  } = {},
+): InteractionCoordinationService {
+  return new InteractionCoordinationService(
+    port,
+    options.config ?? config,
+    options.handoffIdFactory ?? (() => 'handoff-1'),
+  );
+}
+
+describe('InteractionCoordinationService authority boundary', (): void => {
+  it('derives identity and actor/tenant scope server-side, then round-trips the native adapter', async (): Promise<void> => {
+    const adapter = new InMemoryInteractionCoordinationPort();
+    const coordination = service(adapter);
+
+    await expect(
+      coordination.handoff(
+        { idempotencyKey: 'request-1', summary: 'Implement the requested feature' },
+        context,
+      ),
+    ).resolves.toEqual({
+      handoffId: 'handoff-1',
+      targetAgentId: 'Conductor',
+      status: 'queued',
+      correlationId: 'corr-1',
+    });
+
+    adapter.recordActivity('handoff-1', 'running', 'Orchestrator accepted the request');
+    adapter.recordResult('handoff-1', 'completed', 'Merged by orchestrator');
+
+    const followUpContext = { ...context, correlationId: 'corr-2' };
+    await expect(coordination.observe('handoff-1', followUpContext)).resolves.toMatchObject({
+      targetAgentId: 'Conductor',
+      status: 'completed',
+    });
+    await expect(coordination.result('handoff-1', followUpContext)).resolves.toMatchObject({
+      targetAgentId: 'Conductor',
+      status: 'completed',
+      summary: 'Merged by orchestrator',
+    });
+  });
+
+  it('fails closed without calling a port when the interaction requester is unconfigured', async (): Promise<void> => {
+    const adapter = new InMemoryInteractionCoordinationPort();
+    const handoff = vi.spyOn(adapter, 'handoff');
+    const coordination = service(adapter, {
+      config: { interactionAgentId: '', orchestrationAgentId: 'Conductor' },
+    });
+
+    await expect(
+      coordination.handoff(
+        { idempotencyKey: 'request-1', summary: 'Implement the requested feature' },
+        context,
+      ),
+    ).rejects.toMatchObject({
+      code: 'unconfigured_requester',
+    } satisfies Partial<InteractionCoordinationGatewayError>);
+    expect(handoff).not.toHaveBeenCalled();
+  });
+
+  it('rejects self-delegation configuration before delivering work', async (): Promise<void> => {
+    const adapter = new InMemoryInteractionCoordinationPort();
+    const handoff = vi.spyOn(adapter, 'handoff');
+    const coordination = service(adapter, {
+      config: { interactionAgentId: 'Nova', orchestrationAgentId: 'Nova' },
+    });
+
+    await expect(
+      coordination.handoff(
+        { idempotencyKey: 'request-1', summary: 'Implement the requested feature' },
+        context,
+      ),
+    ).rejects.toThrow('Interaction and orchestration identities must differ');
+    expect(handoff).not.toHaveBeenCalled();
+  });
+
+  it('denies cross-tenant observe and result before calling the adapter', async (): Promise<void> => {
+    const adapter = new InMemoryInteractionCoordinationPort();
+    const observe = vi.spyOn(adapter, 'observe');
+    const result = vi.spyOn(adapter, 'result');
+    const coordination = service(adapter);
+    await coordination.handoff(
+      { idempotencyKey: 'request-1', summary: 'Implement the requested feature' },
+      context,
+    );
+
+    const otherTenant = {
+      ...context,
+      actorScope: { ...context.actorScope, tenantId: 'tenant-b' },
+    };
+    await expect(coordination.observe('handoff-1', otherTenant)).rejects.toMatchObject({
+      code: 'cross_tenant_forbidden',
+    } satisfies Partial<InteractionCoordinationGatewayError>);
+    await expect(coordination.result('handoff-1', otherTenant)).rejects.toMatchObject({
+      code: 'cross_tenant_forbidden',
+    } satisfies Partial<InteractionCoordinationGatewayError>);
+    expect(observe).not.toHaveBeenCalled();
+    expect(result).not.toHaveBeenCalled();
+  });
+
+  it('scopes idempotency by actor and joins concurrent retries without duplicate delivery', async (): Promise<void> => {
+    let handoffSequence = 0;
+    let release: (() => void) | undefined;
+    const delivered = new Promise<void>((resolve: () => void): void => {
+      release = resolve;
+    });
+    const adapter: InteractionCoordinationPort = {
+      handoff: vi.fn(async (handoff: Handoff) => {
+        await delivered;
+        return {
+          handoffId: handoff.handoffId,
+          targetAgentId: handoff.targetAgentId,
+          status: 'queued' as const,
+          correlationId: handoff.scope.correlationId,
+        };
+      }),
+      observe: vi.fn(),
+      result: vi.fn(),
+    };
+    const coordination = service(adapter, {
+      handoffIdFactory: (): string => `handoff-${++handoffSequence}`,
+    });
+    const request = { idempotencyKey: 'request-1', summary: 'Implement the requested feature' };
+
+    const first = coordination.handoff(request, context);
+    const retry = coordination.handoff(request, context);
+    expect(adapter.handoff).toHaveBeenCalledTimes(1);
+    release?.();
+    await expect(Promise.all([first, retry])).resolves.toEqual([
+      expect.objectContaining({ handoffId: 'handoff-1' }),
+      expect.objectContaining({ handoffId: 'handoff-1' }),
+    ]);
+
+    await expect(
+      coordination.handoff(request, {
+        ...context,
+        actorScope: { ...context.actorScope, userId: 'operator-2' },
+      }),
+    ).resolves.toMatchObject({ handoffId: 'handoff-2' });
+    expect(adapter.handoff).toHaveBeenCalledTimes(2);
+  });
+
+  it('rejects idempotency-key payload drift and malformed handoff input before delivery', async (): Promise<void> => {
+    const adapter = new InMemoryInteractionCoordinationPort();
+    const handoff = vi.spyOn(adapter, 'handoff');
+    const coordination = service(adapter);
+    await coordination.handoff(
+      { idempotencyKey: 'request-1', summary: 'Implement the requested feature' },
+      context,
+    );
+
+    await expect(
+      coordination.handoff({ idempotencyKey: 'request-1', summary: 'Different work' }, context),
+    ).rejects.toMatchObject({
+      code: 'handoff_conflict',
+    } satisfies Partial<InteractionCoordinationGatewayError>);
+    await expect(
+      coordination.handoff({ idempotencyKey: 'request-2', summary: '' }, context),
+    ).rejects.toMatchObject({
+      code: 'invalid_request',
+    } satisfies Partial<InteractionCoordinationGatewayError>);
+    await expect(
+      coordination.handoff({ idempotencyKey: 'request-3', summary: 'x'.repeat(2_049) }, context),
+    ).rejects.toMatchObject({
+      code: 'invalid_request',
+    } satisfies Partial<InteractionCoordinationGatewayError>);
+    expect(handoff).toHaveBeenCalledTimes(1);
+  });
+
+  it('fails closed when the port reports a target that drifts from configuration', async (): Promise<void> => {
+    const adapter: InteractionCoordinationPort = {
+      handoff: vi.fn(async (handoff: Handoff) => ({
+        handoffId: handoff.handoffId,
+        targetAgentId: 'Unexpected',
+        status: 'accepted' as const,
+        correlationId: handoff.scope.correlationId,
+      })),
+      observe: vi.fn(),
+      result: vi.fn(),
+    };
+
+    await expect(
+      service(adapter).handoff(
+        { idempotencyKey: 'request-1', summary: 'Implement the requested feature' },
+        context,
+      ),
+    ).rejects.toMatchObject({ code: 'target_drift' });
+  });
+});

apps/gateway/src/coord/interaction-coordination.service.ts

@@ -0,0 +1,303 @@
+import { Inject, Injectable } from '@nestjs/common';
+import {
+  InteractionCoordinationClient,
+  type CoordinationObservation,
+  type CoordinationResult,
+  type CoordinationScope,
+  type InteractionCoordinationIdentity,
+  type InteractionCoordinationPort,
+  type HandoffReceipt,
+} from '@mosaicstack/coord';
+import type { RuntimeProviderRequestContext } from '../agent/runtime-provider-registry.service.js';
+import type { CreateHandoffDto } from './interaction-coordination.dto.js';
+
+export const COORDINATION_PORT = Symbol('COORDINATION_PORT');
+export const COORDINATION_CONFIG = Symbol('COORDINATION_CONFIG');
+
+const HANDOFF_TRACKING_TTL_MS = 60 * 60 * 1_000;
+const MAX_TRACKED_HANDOFFS = 1_000;
+const MAX_IDEMPOTENCY_KEY_LENGTH = 128;
+const MAX_SUMMARY_LENGTH = 2_048;
+const MAX_CONTEXT_LENGTH = 8_192;
+const MAX_MISSION_ID_LENGTH = 128;
+
+export interface InteractionCoordinationConfig {
+  interactionAgentId?: string;
+  orchestrationAgentId?: string;
+}
+
+interface HandoffOwner {
+  actorId: string;
+  tenantId: string;
+  requesterAgentId: string;
+  correlationId: string;
+  expiresAt: number;
+}
+
+interface NormalizedHandoffRequest {
+  idempotencyKey: string;
+  summary: string;
+  context?: string;
+  missionId?: string;
+}
+
+interface TrackedHandoff {
+  request: NormalizedHandoffRequest;
+  receipt: Promise<HandoffReceipt>;
+  expiresAt: number;
+}
+
+/**
+ * Gateway authority boundary for the interaction agent. It derives requester,
+ * actor, and tenant from trusted server configuration and authentication; no
+ * channel request can name a target or gain orchestrator-owned orchestration verbs.
+ */
+@Injectable()
+export class InteractionCoordinationService {
+  private readonly owners = new Map<string, HandoffOwner>();
+  private readonly handoffsByIdempotencyKey = new Map<string, TrackedHandoff>();
+
+  constructor(
+    @Inject(COORDINATION_PORT) private readonly port: InteractionCoordinationPort,
+    @Inject(COORDINATION_CONFIG) private readonly config: InteractionCoordinationConfig,
+    private readonly handoffIdFactory: () => string = (): string => crypto.randomUUID(),
+  ) {}
+
+  async handoff(
+    request: CreateHandoffDto,
+    context: RuntimeProviderRequestContext,
+  ): Promise<HandoffReceipt> {
+    this.pruneExpiredTracking();
+    const normalized = this.normalizeRequest(request);
+    const scope = this.scope(context);
+    const idempotencyKey = this.idempotencyKey(normalized.idempotencyKey, scope);
+    const existing = this.handoffsByIdempotencyKey.get(idempotencyKey);
+    if (existing !== undefined) {
+      if (!sameRequest(existing.request, normalized)) {
+        throw new InteractionCoordinationGatewayError(
+          'handoff_conflict',
+          'Handoff idempotency key is already bound to different immutable input',
+        );
+      }
+      return existing.receipt;
+    }
+
+    const pending = this.deliverHandoff(this.handoffIdFactory(), normalized, scope);
+    const tracked: TrackedHandoff = {
+      request: normalized,
+      receipt: pending,
+      expiresAt: this.expiresAt(),
+    };
+    this.handoffsByIdempotencyKey.set(idempotencyKey, tracked);
+    this.enforceTrackingLimit(this.handoffsByIdempotencyKey);
+    try {
+      return await pending;
+    } catch (error: unknown) {
+      if (this.handoffsByIdempotencyKey.get(idempotencyKey) === tracked) {
+        this.handoffsByIdempotencyKey.delete(idempotencyKey);
+      }
+      throw error;
+    }
+  }
+
+  async observe(
+    handoffId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<CoordinationObservation> {
+    this.pruneExpiredTracking();
+    const scope = this.scope(context);
+    const owner = this.ownerFor(handoffId, scope);
+    return this.client().observe(handoffId, { ...scope, correlationId: owner.correlationId });
+  }
+
+  async result(
+    handoffId: string,
+    context: RuntimeProviderRequestContext,
+  ): Promise<CoordinationResult> {
+    this.pruneExpiredTracking();
+    const scope = this.scope(context);
+    const owner = this.ownerFor(handoffId, scope);
+    return this.client().result(handoffId, { ...scope, correlationId: owner.correlationId });
+  }
+
+  private async deliverHandoff(
+    handoffId: string,
+    request: NormalizedHandoffRequest,
+    scope: CoordinationScope,
+  ): Promise<HandoffReceipt> {
+    const receipt = await this.client((): string => handoffId).handoff(request, scope);
+    const owner: HandoffOwner = {
+      actorId: scope.actorId,
+      tenantId: scope.tenantId,
+      requesterAgentId: scope.requesterAgentId,
+      correlationId: scope.correlationId,
+      expiresAt: this.expiresAt(),
+    };
+    const existing = this.owners.get(receipt.handoffId);
+    if (existing !== undefined && !sameOwner(existing, owner)) {
+      throw new InteractionCoordinationGatewayError(
+        'handoff_conflict',
+        'Handoff ID is already bound to a different authenticated scope',
+      );
+    }
+    this.owners.set(receipt.handoffId, owner);
+    this.enforceTrackingLimit(this.owners);
+    return receipt;
+  }
+
+  private client(handoffIdFactory?: () => string): InteractionCoordinationClient {
+    return new InteractionCoordinationClient(this.identity(), this.port, handoffIdFactory);
+  }
+
+  private identity(): InteractionCoordinationIdentity {
+    const interactionAgentId = this.config.interactionAgentId?.trim();
+    const orchestrationAgentId = this.config.orchestrationAgentId?.trim();
+    if (!interactionAgentId) {
+      throw new InteractionCoordinationGatewayError(
+        'unconfigured_requester',
+        'Interaction agent identity is not configured',
+      );
+    }
+    if (!orchestrationAgentId) {
+      throw new InteractionCoordinationGatewayError(
+        'unconfigured_target',
+        'Orchestration agent identity is not configured',
+      );
+    }
+    return { interactionAgentId, orchestrationAgentId };
+  }
+
+  private scope(context: RuntimeProviderRequestContext): CoordinationScope {
+    const identity = this.identity();
+    return Object.freeze({
+      actorId: context.actorScope.userId,
+      tenantId: context.actorScope.tenantId,
+      correlationId: context.correlationId,
+      requesterAgentId: identity.interactionAgentId,
+    });
+  }
+
+  private normalizeRequest(request: CreateHandoffDto): NormalizedHandoffRequest {
+    if (typeof request !== 'object' || request === null) {
+      throw new InteractionCoordinationGatewayError(
+        'invalid_request',
+        'Handoff request is invalid',
+      );
+    }
+    const idempotencyKey = this.requiredString(
+      request.idempotencyKey,
+      'idempotency key',
+      MAX_IDEMPOTENCY_KEY_LENGTH,
+    );
+    const summary = this.requiredString(request.summary, 'summary', MAX_SUMMARY_LENGTH);
+    const context = this.optionalString(request.context, 'context', MAX_CONTEXT_LENGTH);
+    const missionId = this.optionalString(request.missionId, 'mission ID', MAX_MISSION_ID_LENGTH);
+    return Object.freeze({
+      idempotencyKey,
+      summary,
+      ...(context === undefined ? {} : { context }),
+      ...(missionId === undefined ? {} : { missionId }),
+    });
+  }
+
+  private idempotencyKey(requestKey: string, scope: CoordinationScope): string {
+    return `${scope.tenantId}\u0000${scope.actorId}\u0000${scope.requesterAgentId}\u0000${requestKey}`;
+  }
+
+  private requiredString(value: unknown, field: string, maximumLength: number): string {
+    if (typeof value !== 'string') {
+      throw new InteractionCoordinationGatewayError(
+        'invalid_request',
+        `Handoff ${field} must be a string`,
+      );
+    }
+    const normalized = value.trim();
+    if (normalized.length === 0 || normalized.length > maximumLength) {
+      throw new InteractionCoordinationGatewayError(
+        'invalid_request',
+        `Handoff ${field} is invalid`,
+      );
+    }
+    return normalized;
+  }
+
+  private optionalString(value: unknown, field: string, maximumLength: number): string | undefined {
+    if (value === undefined) return undefined;
+    return this.requiredString(value, field, maximumLength);
+  }
+
+  private expiresAt(): number {
+    return Date.now() + HANDOFF_TRACKING_TTL_MS;
+  }
+
+  private pruneExpiredTracking(): void {
+    const now = Date.now();
+    for (const [key, tracked] of this.handoffsByIdempotencyKey) {
+      if (tracked.expiresAt <= now) this.handoffsByIdempotencyKey.delete(key);
+    }
+    for (const [key, owner] of this.owners) {
+      if (owner.expiresAt <= now) this.owners.delete(key);
+    }
+  }
+
+  private enforceTrackingLimit<T>(entries: Map<string, T>): void {
+    while (entries.size > MAX_TRACKED_HANDOFFS) {
+      const oldest = entries.keys().next().value;
+      if (typeof oldest !== 'string') return;
+      entries.delete(oldest);
+    }
+  }
+
+  private ownerFor(handoffId: string, scope: CoordinationScope): HandoffOwner {
+    const owner = this.owners.get(handoffId);
+    if (owner === undefined) {
+      throw new InteractionCoordinationGatewayError('not_found', 'Handoff was not found');
+    }
+    if (
+      owner.tenantId !== scope.tenantId ||
+      owner.actorId !== scope.actorId ||
+      owner.requesterAgentId !== scope.requesterAgentId
+    ) {
+      throw new InteractionCoordinationGatewayError(
+        'cross_tenant_forbidden',
+        'Handoff is outside the authenticated scope',
+      );
+    }
+    return owner;
+  }
+}
+
+export type InteractionCoordinationGatewayErrorCode =
+  | 'cross_tenant_forbidden'
+  | 'handoff_conflict'
+  | 'invalid_request'
+  | 'not_found'
+  | 'unconfigured_requester'
+  | 'unconfigured_target';
+
+function sameOwner(left: HandoffOwner, right: HandoffOwner): boolean {
+  return (
+    left.actorId === right.actorId &&
+    left.tenantId === right.tenantId &&
+    left.requesterAgentId === right.requesterAgentId
+  );
+}
+
+function sameRequest(left: NormalizedHandoffRequest, right: NormalizedHandoffRequest): boolean {
+  return (
+    left.idempotencyKey === right.idempotencyKey &&
+    left.summary === right.summary &&
+    left.context === right.context &&
+    left.missionId === right.missionId
+  );
+}
+
+export class InteractionCoordinationGatewayError extends Error {
+  constructor(
+    readonly code: InteractionCoordinationGatewayErrorCode,
+    message: string,
+  ) {
+    super(message);
+    this.name = InteractionCoordinationGatewayError.name;
+  }
+}

apps/gateway/src/federation/client/__tests__/query-source.service.spec.ts

@@ -0,0 +1,255 @@
+import 'reflect-metadata';
+import { describe, expect, it, vi } from 'vitest';
+import type { Db } from '@mosaicstack/db';
+import type { FederationListResponse } from '@mosaicstack/types';
+import {
+  FederationClientError,
+  type FederationClientService,
+} from '../federation-client.service.js';
+import { type QuerySourceError, QuerySourceService } from '../query-source.service.js';
+
+interface TestRow {
+  id: string;
+  title: string;
+}
+
+interface PeerRow {
+  id: string;
+  commonName: string;
+  endpointUrl: string | null;
+  clientKeyPem: string | null;
+  state: 'active' | 'pending' | 'suspended' | 'revoked';
+}
+
+const LOCAL_ROWS: TestRow[] = [
+  { id: 'local-1', title: 'Local One' },
+  { id: 'local-2', title: 'Local Two' },
+];
+
+const PEER_A: PeerRow = {
+  id: 'aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa',
+  commonName: 'peer-a',
+  endpointUrl: 'https://peer-a.example.com',
+  clientKeyPem: 'sealed-key-a',
+  state: 'active',
+};
+
+const PEER_B: PeerRow = {
+  id: 'bbbbbbbb-bbbb-bbbb-bbbb-bbbbbbbbbbbb',
+  commonName: 'peer-b',
+  endpointUrl: 'https://peer-b.example.com',
+  clientKeyPem: 'sealed-key-b',
+  state: 'active',
+};
+
+const PEER_LOCALHOST: PeerRow = {
+  id: 'cccccccc-cccc-cccc-cccc-cccccccccccc',
+  commonName: 'peer-localhost',
+  endpointUrl: 'https://localhost:3001',
+  clientKeyPem: 'sealed-key-c',
+  state: 'active',
+};
+
+function makeDb(activePeers: PeerRow[]): Db {
+  const orderBy = vi.fn().mockResolvedValue(activePeers);
+  const where = vi.fn().mockReturnValue({ orderBy });
+  const from = vi.fn().mockReturnValue({ where });
+  const select = vi.fn().mockReturnValue({ from });
+
+  return {
+    select,
+    insert: vi.fn(),
+    update: vi.fn(),
+    delete: vi.fn(),
+    transaction: vi.fn(),
+  } as unknown as Db;
+}
+
+function makeFederationClient(
+  list: (
+    peerId: string,
+    resource: string,
+    request: Record<string, unknown>,
+  ) => Promise<FederationListResponse<TestRow>>,
+): FederationClientService {
+  return {
+    list: list as unknown as FederationClientService['list'],
+  } as FederationClientService;
+}
+
+function makeLocalResponse(rows: TestRow[] = LOCAL_ROWS): Promise<FederationListResponse<TestRow>> {
+  return Promise.resolve({ items: rows });
+}
+
+describe('QuerySourceService', () => {
+  it('routes source="local" to the local executor and tags rows as local', async () => {
+    const list = vi.fn(async (): Promise<FederationListResponse<TestRow>> => ({ items: [] }));
+    const service = new QuerySourceService(makeDb([PEER_A]), makeFederationClient(list));
+
+    const result = await service.list<TestRow>({
+      source: 'local',
+      resource: 'tasks',
+      request: { cursor: 'ignored-for-local-test' },
+      local: () => makeLocalResponse(),
+    });
+
+    expect(result).toEqual({
+      items: [
+        { id: 'local-1', title: 'Local One', _source: 'local' },
+        { id: 'local-2', title: 'Local Two', _source: 'local' },
+      ],
+    });
+    expect(list).not.toHaveBeenCalled();
+  });
+
+  it('routes source="federated:<host>" to the matching active peer and tags rows with peer commonName', async () => {
+    const list = vi.fn(
+      async (): Promise<FederationListResponse<TestRow>> => ({
+        items: [{ id: 'remote-1', title: 'Remote One' }],
+      }),
+    );
+    const service = new QuerySourceService(makeDb([PEER_A, PEER_B]), makeFederationClient(list));
+
+    const result = await service.list<TestRow>({
+      source: 'federated:peer-b.example.com',
+      resource: 'tasks',
+      request: { status: 'open' },
+      local: () => makeLocalResponse(),
+    });
+
+    expect(result).toEqual({
+      items: [{ id: 'remote-1', title: 'Remote One', _source: 'peer-b' }],
+    });
+    expect(list).toHaveBeenCalledWith(PEER_B.id, 'tasks', { status: 'open' });
+  });
+
+  it('matches federated hosts by endpoint host including non-default port', async () => {
+    const list = vi.fn(
+      async (): Promise<FederationListResponse<TestRow>> => ({
+        items: [{ id: 'remote-port', title: 'Remote Port' }],
+      }),
+    );
+    const service = new QuerySourceService(makeDb([PEER_LOCALHOST]), makeFederationClient(list));
+
+    const result = await service.list<TestRow>({
+      source: 'federated:localhost:3001',
+      resource: 'tasks',
+      request: {},
+      local: () => makeLocalResponse(),
+    });
+
+    expect(result).toEqual({
+      items: [{ id: 'remote-port', title: 'Remote Port', _source: 'peer-localhost' }],
+    });
+    expect(list).toHaveBeenCalledWith(PEER_LOCALHOST.id, 'tasks', {});
+  });
+
+  it('fans out source="all" to local plus every active outbound peer in parallel and merges tagged rows', async () => {
+    const callOrder: string[] = [];
+    const list = vi.fn(async (peerId: string): Promise<FederationListResponse<TestRow>> => {
+      callOrder.push(`remote-start:${peerId}`);
+      await Promise.resolve();
+      return {
+        items: [{ id: `remote-${peerId.slice(0, 1)}`, title: `Remote ${peerId.slice(0, 1)}` }],
+      };
+    });
+    const service = new QuerySourceService(makeDb([PEER_A, PEER_B]), makeFederationClient(list));
+
+    const result = await service.list<TestRow>({
+      source: 'all',
+      resource: 'tasks',
+      request: { limit: 25 },
+      local: async () => {
+        callOrder.push('local-start');
+        await Promise.resolve();
+        return { items: [{ id: 'local-1', title: 'Local One' }] };
+      },
+    });
+
+    expect(result).toEqual({
+      items: [
+        { id: 'local-1', title: 'Local One', _source: 'local' },
+        { id: 'remote-a', title: 'Remote a', _source: 'peer-a' },
+        { id: 'remote-b', title: 'Remote b', _source: 'peer-b' },
+      ],
+    });
+    expect(list).toHaveBeenCalledTimes(2);
+    expect(callOrder).toEqual([
+      'local-start',
+      `remote-start:${PEER_A.id}`,
+      `remote-start:${PEER_B.id}`,
+    ]);
+  });
+
+  it('marks source="all" as partial and truncated when any subquery returns a cursor', async () => {
+    const list = vi.fn(
+      async (): Promise<FederationListResponse<TestRow>> => ({
+        items: [{ id: 'remote-a', title: 'Remote A' }],
+        nextCursor: 'remote-next',
+      }),
+    );
+    const service = new QuerySourceService(makeDb([PEER_A]), makeFederationClient(list));
+
+    const result = await service.list<TestRow>({
+      source: 'all',
+      resource: 'tasks',
+      request: {},
+      local: () => makeLocalResponse([{ id: 'local-1', title: 'Local One' }]),
+    });
+
+    expect(result).toEqual({
+      items: [
+        { id: 'local-1', title: 'Local One', _source: 'local' },
+        { id: 'remote-a', title: 'Remote A', _source: 'peer-a' },
+      ],
+      _partial: true,
+      _truncated: true,
+    });
+  });
+
+  it('returns _partial=true for source="all" when one peer fails without dropping successful sources', async () => {
+    const list = vi.fn(async (peerId: string): Promise<FederationListResponse<TestRow>> => {
+      if (peerId === PEER_B.id) {
+        throw new FederationClientError({
+          code: 'NETWORK',
+          message: 'peer unavailable',
+          peerId,
+        });
+      }
+      return { items: [{ id: 'remote-a', title: 'Remote A' }] };
+    });
+    const service = new QuerySourceService(makeDb([PEER_A, PEER_B]), makeFederationClient(list));
+
+    const result = await service.list<TestRow>({
+      source: 'all',
+      resource: 'tasks',
+      request: {},
+      local: () => makeLocalResponse([{ id: 'local-1', title: 'Local One' }]),
+    });
+
+    expect(result).toEqual({
+      items: [
+        { id: 'local-1', title: 'Local One', _source: 'local' },
+        { id: 'remote-a', title: 'Remote A', _source: 'peer-a' },
+      ],
+      _partial: true,
+    });
+  });
+
+  it('throws QuerySourceError when a federated host does not match an active outbound peer', async () => {
+    const list = vi.fn(async (): Promise<FederationListResponse<TestRow>> => ({ items: [] }));
+    const service = new QuerySourceService(makeDb([PEER_A]), makeFederationClient(list));
+
+    await expect(
+      service.list<TestRow>({
+        source: 'federated:missing.example.com',
+        resource: 'tasks',
+        request: {},
+        local: () => makeLocalResponse(),
+      }),
+    ).rejects.toMatchObject({
+      name: 'QuerySourceError',
+      code: 'PEER_NOT_FOUND',
+    } satisfies Partial<QuerySourceError>);
+  });
+});

apps/gateway/src/federation/client/index.ts

@@ -11,3 +11,13 @@ export {
   type FederationClientErrorCode,
   type FederationClientErrorOptions,
 } from './federation-client.service.js';
+export {
+  QuerySourceService,
+  QuerySourceError,
+  type QuerySource,
+  type QuerySourceErrorCode,
+  type QuerySourceErrorOptions,
+  type QuerySourceListOptions,
+  type QuerySourceListResponse,
+  type LocalListExecutor,
+} from './query-source.service.js';

apps/gateway/src/federation/client/query-source.service.ts

@@ -0,0 +1,261 @@
+/**
+ * QuerySourceService — gateway query source router (FED-M3-09).
+ *
+ * Accepts the federation query-layer `source` selector and routes list-style
+ * reads to local storage, one federated peer, or all active outbound peers.
+ *
+ * `source: "all"` is intentionally tolerant of per-peer failures: local data
+ * and successful peer responses are returned, and the envelope is marked
+ * `_partial: true`.  Local failures still reject because there is no safe local
+ * fallback and the gateway's own storage is expected to be authoritative.
+ */
+
+import { Inject, Injectable, Logger } from '@nestjs/common';
+import { and, eq, federationPeers, isNotNull, type Db } from '@mosaicstack/db';
+import {
+  SOURCE_LOCAL,
+  tagWithSource,
+  type FederationListResponse,
+  type SourceTag,
+} from '@mosaicstack/types';
+import { DB } from '../../database/database.module.js';
+import { FederationClientService } from './federation-client.service.js';
+
+export type QuerySource = 'local' | 'all' | `federated:${string}`;
+
+export type QuerySourceErrorCode = 'INVALID_SOURCE' | 'PEER_NOT_FOUND';
+
+export interface QuerySourceErrorOptions {
+  code: QuerySourceErrorCode;
+  message: string;
+  source: string;
+}
+
+export class QuerySourceError extends Error {
+  readonly code: QuerySourceErrorCode;
+  readonly source: string;
+
+  constructor(opts: QuerySourceErrorOptions) {
+    super(opts.message);
+    this.name = 'QuerySourceError';
+    this.code = opts.code;
+    this.source = opts.source;
+  }
+}
+
+export type LocalListExecutor<T extends object> = () => Promise<FederationListResponse<T> | T[]>;
+
+export interface QuerySourceListOptions<T extends object> {
+  source: QuerySource;
+  resource: string;
+  request?: Record<string, unknown>;
+  local: LocalListExecutor<T>;
+}
+
+export type QuerySourceListResponse<T extends object> = FederationListResponse<T & SourceTag>;
+
+interface OutboundPeer {
+  id: string;
+  commonName: string;
+  endpointUrl: string;
+}
+
+interface TaggedList<T extends object> {
+  items: Array<T & SourceTag>;
+  partial: boolean;
+  truncated: boolean;
+  nextCursor?: string;
+}
+
+@Injectable()
+export class QuerySourceService {
+  private readonly logger = new Logger(QuerySourceService.name);
+
+  constructor(
+    @Inject(DB) private readonly db: Db,
+    @Inject(FederationClientService) private readonly federationClient: FederationClientService,
+  ) {}
+
+  async list<T extends object>(
+    options: QuerySourceListOptions<T>,
+  ): Promise<QuerySourceListResponse<T>> {
+    const request = options.request ?? {};
+
+    if (options.source === 'local') {
+      const local = await this.runLocal(options.local);
+      return this.toResponse(this.tagList(local, SOURCE_LOCAL));
+    }
+
+    if (options.source === 'all') {
+      return this.listAll(options.resource, request, options.local);
+    }
+
+    if (options.source.startsWith('federated:')) {
+      const host = options.source.slice('federated:'.length).trim();
+      if (!host) {
+        throw new QuerySourceError({
+          code: 'INVALID_SOURCE',
+          message: 'Federated source must include a host after federated:',
+          source: options.source,
+        });
+      }
+
+      const peer = await this.findPeerByHost(host, options.source);
+      const remote = await this.federationClient.list<T>(peer.id, options.resource, request);
+      return this.toResponse(this.tagList(remote, peer.commonName));
+    }
+
+    throw new QuerySourceError({
+      code: 'INVALID_SOURCE',
+      message: `Unsupported query source: ${options.source}`,
+      source: options.source,
+    });
+  }
+
+  private async listAll<T extends object>(
+    resource: string,
+    request: Record<string, unknown>,
+    local: LocalListExecutor<T>,
+  ): Promise<QuerySourceListResponse<T>> {
+    const peers = await this.listActiveOutboundPeers();
+
+    const localPromise = this.runLocal(local).then((response) =>
+      this.tagList(response, SOURCE_LOCAL),
+    );
+    const remotePromises = peers.map(async (peer: OutboundPeer): Promise<TaggedList<T> | null> => {
+      try {
+        const response = await this.federationClient.list<T>(peer.id, resource, request);
+        return this.tagList(response, peer.commonName);
+      } catch (error: unknown) {
+        this.logger.warn(
+          `Federated query to peer ${peer.commonName} (${peer.id}) failed; returning partial all-source response: ${
+            error instanceof Error ? error.message : String(error)
+          }`,
+        );
+        return null;
+      }
+    });
+
+    const [localResult, ...remoteResults] = await Promise.all([localPromise, ...remotePromises]);
+    const successfulRemoteResults = remoteResults.filter(
+      (result: TaggedList<T> | null): result is TaggedList<T> => result !== null,
+    );
+    const allResults = [localResult, ...successfulRemoteResults];
+    const peerFailure = successfulRemoteResults.length !== peers.length;
+
+    return this.mergeTaggedLists(allResults, peerFailure);
+  }
+
+  private async runLocal<T extends object>(
+    local: LocalListExecutor<T>,
+  ): Promise<FederationListResponse<T>> {
+    const response = await local();
+    if (Array.isArray(response)) {
+      return { items: response };
+    }
+    return response;
+  }
+
+  private tagList<T extends object>(
+    response: FederationListResponse<T>,
+    source: string,
+  ): TaggedList<T> {
+    return {
+      items: tagWithSource(response.items, source),
+      partial: response._partial === true,
+      truncated: response._truncated === true || response.nextCursor !== undefined,
+      nextCursor: response.nextCursor,
+    };
+  }
+
+  private mergeTaggedLists<T extends object>(
+    lists: Array<TaggedList<T>>,
+    peerFailure: boolean,
+  ): QuerySourceListResponse<T> {
+    const items = lists.flatMap((list: TaggedList<T>) => list.items);
+    const partial =
+      peerFailure ||
+      lists.some((list: TaggedList<T>) => list.partial || list.nextCursor !== undefined);
+    const truncated = lists.some((list: TaggedList<T>) => list.truncated);
+
+    const response: QuerySourceListResponse<T> = { items };
+    if (partial) {
+      response._partial = true;
+    }
+    if (truncated) {
+      response._truncated = true;
+    }
+    return response;
+  }
+
+  private toResponse<T extends object>(tagged: TaggedList<T>): QuerySourceListResponse<T> {
+    const response: QuerySourceListResponse<T> = {
+      items: tagged.items,
+    };
+    if (tagged.nextCursor !== undefined) {
+      response.nextCursor = tagged.nextCursor;
+    }
+    if (tagged.partial) {
+      response._partial = true;
+    }
+    if (tagged.truncated) {
+      response._truncated = true;
+    }
+    return response;
+  }
+
+  private async findPeerByHost(sourceHost: string, source: string): Promise<OutboundPeer> {
+    const host = normalizeHost(sourceHost);
+    const peers = await this.listActiveOutboundPeers();
+    const peer = peers.find((candidate: OutboundPeer) => {
+      const commonName = normalizeHost(candidate.commonName);
+      const endpointHosts = endpointHostKeys(candidate.endpointUrl).map((endpointHost: string) =>
+        normalizeHost(endpointHost),
+      );
+      return commonName === host || endpointHosts.includes(host);
+    });
+
+    if (!peer) {
+      throw new QuerySourceError({
+        code: 'PEER_NOT_FOUND',
+        message: `No active outbound federation peer matches source ${source}`,
+        source,
+      });
+    }
+
+    return peer;
+  }
+
+  private async listActiveOutboundPeers(): Promise<OutboundPeer[]> {
+    const rows = await this.db
+      .select({
+        id: federationPeers.id,
+        commonName: federationPeers.commonName,
+        endpointUrl: federationPeers.endpointUrl,
+      })
+      .from(federationPeers)
+      .where(
+        and(
+          eq(federationPeers.state, 'active'),
+          isNotNull(federationPeers.endpointUrl),
+          isNotNull(federationPeers.clientKeyPem),
+        ),
+      )
+      .orderBy(federationPeers.commonName);
+
+    return rows.filter((row): row is OutboundPeer => typeof row.endpointUrl === 'string');
+  }
+}
+
+function normalizeHost(host: string): string {
+  return host.trim().toLowerCase();
+}
+
+function endpointHostKeys(endpointUrl: string): string[] {
+  try {
+    const url = new URL(endpointUrl);
+    return Array.from(new Set([url.host, url.hostname].filter((host: string) => host.length > 0)));
+  } catch {
+    return [];
+  }
+}

apps/gateway/src/federation/federation.module.ts

@@ -4,26 +4,35 @@ import { CaService } from './ca.service.js';
 import { EnrollmentController } from './enrollment.controller.js';
 import { EnrollmentService } from './enrollment.service.js';
 import { FederationController } from './federation.controller.js';
+import { CapabilitiesController } from './server/verbs/capabilities.controller.js';
 import { GrantsService } from './grants.service.js';
-import { FederationClientService } from './client/index.js';
-import { FederationAuthGuard } from './server/index.js';
+import { FederationClientService, QuerySourceService } from './client/index.js';
+import { FederationAuthGuard, FederationScopeService } from './server/index.js';
+import { ListController } from './server/verbs/list.controller.js';
+import { FederationListQueryService } from './server/verbs/list-query.service.js';
 
 @Module({
-  controllers: [EnrollmentController, FederationController],
+  controllers: [EnrollmentController, FederationController, CapabilitiesController, ListController],
   providers: [
     AdminGuard,
     CaService,
     EnrollmentService,
     GrantsService,
     FederationClientService,
+    QuerySourceService,
     FederationAuthGuard,
+    FederationScopeService,
+    FederationListQueryService,
   ],
   exports: [
     CaService,
     EnrollmentService,
     GrantsService,
     FederationClientService,
+    QuerySourceService,
     FederationAuthGuard,
+    FederationScopeService,
+    FederationListQueryService,
   ],
 })
 export class FederationModule {}

apps/gateway/src/federation/server/__tests__/scope.service.spec.ts

@@ -0,0 +1,324 @@
+/**
+ * Unit tests for FederationScopeService (FED-M3-04).
+ *
+ * Coverage:
+ *  - resource allowlist deny
+ *  - excluded resource deny
+ *  - invalid scope deny
+ *  - invalid requested limit deny
+ *  - native RBAC deny as subjectUserId
+ *  - scope/native filter intersection for personal and team rows
+ *  - native RBAC personal deny wins over scope include_personal allow/default
+ *  - max_rows_per_query cap
+ */
+
+import { beforeEach, describe, expect, it, vi } from 'vitest';
+import { FederationScopeService, type FederationNativeRbacEvaluator } from '../scope.service.js';
+import type { FederationContext } from '../federation-context.js';
+
+const GRANT_ID = 'grant-1';
+const PEER_ID = 'peer-1';
+const SUBJECT_USER_ID = 'user-1';
+
+function makeContext(scope: Record<string, unknown>): FederationContext {
+  return {
+    grantId: GRANT_ID,
+    peerId: PEER_ID,
+    subjectUserId: SUBJECT_USER_ID,
+    scope,
+  };
+}
+
+function makeNativeRbac(
+  result: Awaited<ReturnType<FederationNativeRbacEvaluator['evaluateReadAccess']>>,
+): FederationNativeRbacEvaluator {
+  return {
+    evaluateReadAccess: vi.fn().mockResolvedValue(result),
+  };
+}
+
+describe('FederationScopeService', () => {
+  let service: FederationScopeService;
+
+  beforeEach(() => {
+    service = new FederationScopeService();
+  });
+
+  it('allows a granted resource and returns a capped query filter', async () => {
+    const nativeRbac = makeNativeRbac({
+      allowed: true,
+      access: { includePersonal: true, teamIds: ['team-1', 'team-2'] },
+    });
+
+    const result = await service.evaluateAccess({
+      context: makeContext({
+        resources: ['tasks'],
+        filters: { tasks: { include_teams: ['team-1', 'team-3'], include_personal: true } },
+        max_rows_per_query: 50,
+      }),
+      resource: 'tasks',
+      requestedLimit: 500,
+      nativeRbac,
+    });
+
+    expect(result).toEqual({
+      allowed: true,
+      filter: {
+        resource: 'tasks',
+        subjectUserId: SUBJECT_USER_ID,
+        includePersonal: true,
+        teamIds: ['team-1'],
+        limit: 50,
+        maxRowsPerQuery: 50,
+      },
+    });
+    expect(nativeRbac.evaluateReadAccess).toHaveBeenCalledWith({
+      grantId: GRANT_ID,
+      peerId: PEER_ID,
+      subjectUserId: SUBJECT_USER_ID,
+      resource: 'tasks',
+    });
+  });
+
+  it('defaults absent resource filters to native RBAC personal and team visibility', async () => {
+    const result = await service.evaluateAccess({
+      context: makeContext({ resources: ['notes'], max_rows_per_query: 100 }),
+      resource: 'notes',
+      nativeRbac: makeNativeRbac({
+        allowed: true,
+        access: { includePersonal: true, teamIds: ['team-1', 'team-2'] },
+      }),
+    });
+
+    expect(result).toMatchObject({
+      allowed: true,
+      filter: {
+        includePersonal: true,
+        teamIds: ['team-1', 'team-2'],
+        limit: 100,
+      },
+    });
+  });
+
+  it('honors include_personal false even when native RBAC allows personal rows', async () => {
+    const result = await service.evaluateAccess({
+      context: makeContext({
+        resources: ['memory'],
+        filters: { memory: { include_personal: false } },
+        max_rows_per_query: 25,
+      }),
+      resource: 'memory',
+      nativeRbac: makeNativeRbac({
+        allowed: true,
+        access: { includePersonal: true, teamIds: [] },
+      }),
+    });
+
+    expect(result).toMatchObject({
+      allowed: true,
+      filter: {
+        includePersonal: false,
+        teamIds: [],
+      },
+    });
+  });
+
+  it('does not leak personal rows when scope allows personal but native RBAC denies personal', async () => {
+    const result = await service.evaluateAccess({
+      context: makeContext({
+        resources: ['tasks'],
+        filters: { tasks: { include_personal: true } },
+        max_rows_per_query: 25,
+      }),
+      resource: 'tasks',
+      nativeRbac: makeNativeRbac({
+        allowed: true,
+        access: { includePersonal: false, teamIds: ['team-1'] },
+      }),
+    });
+
+    expect(result).toMatchObject({
+      allowed: true,
+      filter: {
+        includePersonal: false,
+        teamIds: ['team-1'],
+      },
+    });
+  });
+
+  it('does not widen native RBAC when scope includes teams the user cannot access', async () => {
+    const result = await service.evaluateAccess({
+      context: makeContext({
+        resources: ['tasks'],
+        filters: { tasks: { include_teams: ['team-2'], include_personal: false } },
+        max_rows_per_query: 25,
+      }),
+      resource: 'tasks',
+      nativeRbac: makeNativeRbac({
+        allowed: true,
+        access: { includePersonal: true, teamIds: ['team-1'] },
+      }),
+    });
+
+    expect(result).toMatchObject({
+      allowed: true,
+      filter: {
+        includePersonal: false,
+        teamIds: [],
+      },
+    });
+  });
+
+  it('denies invalid grant scope before RBAC evaluation', async () => {
+    const nativeRbac = makeNativeRbac({
+      allowed: true,
+      access: { includePersonal: true, teamIds: [] },
+    });
+
+    const result = await service.evaluateAccess({
+      context: makeContext({ resources: [], max_rows_per_query: 100 }),
+      resource: 'tasks',
+      nativeRbac,
+    });
+
+    expect(result).toMatchObject({
+      allowed: false,
+      deny: {
+        code: 'invalid_scope',
+        stage: 'scope_parse',
+        statusCode: 400,
+        grantId: GRANT_ID,
+        subjectUserId: SUBJECT_USER_ID,
+        resource: 'tasks',
+      },
+    });
+    expect(nativeRbac.evaluateReadAccess).not.toHaveBeenCalled();
+  });
+
+  it('denies unsupported resource names before RBAC evaluation', async () => {
+    const nativeRbac = makeNativeRbac({
+      allowed: true,
+      access: { includePersonal: true, teamIds: [] },
+    });
+
+    const result = await service.evaluateAccess({
+      context: makeContext({ resources: ['tasks'], max_rows_per_query: 100 }),
+      resource: 'unknown_resource',
+      nativeRbac,
+    });
+
+    expect(result).toMatchObject({
+      allowed: false,
+      deny: {
+        code: 'invalid_resource',
+        stage: 'resource_allowlist',
+        statusCode: 403,
+      },
+    });
+    expect(nativeRbac.evaluateReadAccess).not.toHaveBeenCalled();
+  });
+
+  it('denies resources explicitly present in excluded_resources before allowlist miss', async () => {
+    const nativeRbac = makeNativeRbac({
+      allowed: true,
+      access: { includePersonal: true, teamIds: [] },
+    });
+
+    const result = await service.evaluateAccess({
+      context: makeContext({
+        resources: ['tasks'],
+        excluded_resources: ['credentials'],
+        max_rows_per_query: 100,
+      }),
+      resource: 'credentials',
+      nativeRbac,
+    });
+
+    expect(result).toMatchObject({
+      allowed: false,
+      deny: {
+        code: 'resource_excluded',
+        stage: 'resource_exclusion',
+        statusCode: 403,
+        resource: 'credentials',
+      },
+    });
+    expect(nativeRbac.evaluateReadAccess).not.toHaveBeenCalled();
+  });
+
+  it('denies supported resources that are not granted by scope', async () => {
+    const nativeRbac = makeNativeRbac({
+      allowed: true,
+      access: { includePersonal: true, teamIds: [] },
+    });
+
+    const result = await service.evaluateAccess({
+      context: makeContext({ resources: ['tasks'], max_rows_per_query: 100 }),
+      resource: 'notes',
+      nativeRbac,
+    });
+
+    expect(result).toMatchObject({
+      allowed: false,
+      deny: {
+        code: 'resource_not_granted',
+        stage: 'resource_allowlist',
+        statusCode: 403,
+        resource: 'notes',
+      },
+    });
+    expect(nativeRbac.evaluateReadAccess).not.toHaveBeenCalled();
+  });
+
+  it('denies invalid requested row limits before RBAC evaluation', async () => {
+    const nativeRbac = makeNativeRbac({
+      allowed: true,
+      access: { includePersonal: true, teamIds: [] },
+    });
+
+    const result = await service.evaluateAccess({
+      context: makeContext({ resources: ['tasks'], max_rows_per_query: 100 }),
+      resource: 'tasks',
+      requestedLimit: 0,
+      nativeRbac,
+    });
+
+    expect(result).toMatchObject({
+      allowed: false,
+      deny: {
+        code: 'invalid_limit',
+        stage: 'row_cap',
+        statusCode: 400,
+        details: { requestedLimit: 0 },
+      },
+    });
+    expect(nativeRbac.evaluateReadAccess).not.toHaveBeenCalled();
+  });
+
+  it('denies when native RBAC rejects subjectUserId access to the resource', async () => {
+    const result = await service.evaluateAccess({
+      context: makeContext({ resources: ['tasks'], max_rows_per_query: 100 }),
+      resource: 'tasks',
+      nativeRbac: makeNativeRbac({
+        allowed: false,
+        reason: 'read:tasks denied',
+        details: { permission: 'tasks:read' },
+      }),
+    });
+
+    expect(result).toEqual({
+      allowed: false,
+      deny: {
+        code: 'native_rbac_denied',
+        stage: 'native_rbac',
+        statusCode: 403,
+        message: 'read:tasks denied',
+        grantId: GRANT_ID,
+        peerId: PEER_ID,
+        subjectUserId: SUBJECT_USER_ID,
+        resource: 'tasks',
+        details: { permission: 'tasks:read' },
+      },
+    });
+  });
+});

apps/gateway/src/federation/server/index.ts

@@ -10,4 +10,22 @@
  */
 
 export { FederationAuthGuard } from './federation-auth.guard.js';
+export { FederationScopeService } from './scope.service.js';
 export type { FederationContext } from './federation-context.js';
+export type {
+  FederationNativeRbacAccess,
+  FederationNativeRbacAllowedResult,
+  FederationNativeRbacDeniedResult,
+  FederationNativeRbacEvaluator,
+  FederationNativeRbacRequest,
+  FederationNativeRbacResult,
+  FederationScopeAllowedResult,
+  FederationScopeDeniedResult,
+  FederationScopeDenyCode,
+  FederationScopeDenyDetails,
+  FederationScopeDenyReason,
+  FederationScopeDenyStage,
+  FederationScopeEvaluationInput,
+  FederationScopeEvaluationResult,
+  FederationScopeQueryFilter,
+} from './scope.service.js';

apps/gateway/src/federation/server/scope.service.ts

@@ -0,0 +1,272 @@
+/**
+ * FederationScopeService — M3 server-side scope enforcement pipeline.
+ *
+ * Pure trust-boundary service: it validates the grant scope, asks an injected
+ * native RBAC evaluator what the subject user can read locally, intersects that
+ * answer with the federation scope filters, and returns a query filter for the
+ * verb controllers. The service performs no DB calls directly.
+ */
+
+import { Injectable } from '@nestjs/common';
+import {
+  FEDERATION_RESOURCE_VALUES,
+  type FederationResource,
+  FederationScopeError,
+  parseFederationScope,
+} from '../scope-schema.js';
+import type { FederationContext } from './federation-context.js';
+
+const federationResourceSet: ReadonlySet<string> = new Set<string>(FEDERATION_RESOURCE_VALUES);
+
+export type FederationScopeDenyStage =
+  | 'scope_parse'
+  | 'resource_allowlist'
+  | 'resource_exclusion'
+  | 'native_rbac'
+  | 'row_cap';
+
+export type FederationScopeDenyCode =
+  | 'invalid_scope'
+  | 'invalid_resource'
+  | 'resource_not_granted'
+  | 'resource_excluded'
+  | 'native_rbac_denied'
+  | 'invalid_limit';
+
+export type FederationScopeDenyStatus = 400 | 403;
+
+export interface FederationScopeDenyDetails {
+  readonly [key: string]: string | number | boolean | readonly string[];
+}
+
+export interface FederationScopeDenyReason {
+  readonly code: FederationScopeDenyCode;
+  readonly stage: FederationScopeDenyStage;
+  readonly statusCode: FederationScopeDenyStatus;
+  readonly message: string;
+  readonly grantId: string;
+  readonly peerId: string;
+  readonly subjectUserId: string;
+  readonly resource: string;
+  readonly details?: FederationScopeDenyDetails;
+}
+
+export interface FederationNativeRbacRequest {
+  readonly grantId: string;
+  readonly peerId: string;
+  readonly subjectUserId: string;
+  readonly resource: FederationResource;
+}
+
+export interface FederationNativeRbacAccess {
+  /** Whether this user may read personal rows for this resource. */
+  readonly includePersonal: boolean;
+
+  /** Team IDs this user may read for this resource under native RBAC. */
+  readonly teamIds: readonly string[];
+}
+
+export interface FederationNativeRbacAllowedResult {
+  readonly allowed: true;
+  readonly access: FederationNativeRbacAccess;
+}
+
+export interface FederationNativeRbacDeniedResult {
+  readonly allowed: false;
+  readonly reason?: string;
+  readonly details?: FederationScopeDenyDetails;
+}
+
+export type FederationNativeRbacResult =
+  | FederationNativeRbacAllowedResult
+  | FederationNativeRbacDeniedResult;
+
+export interface FederationNativeRbacEvaluator {
+  evaluateReadAccess(request: FederationNativeRbacRequest): Promise<FederationNativeRbacResult>;
+}
+
+export interface FederationScopeEvaluationInput {
+  readonly context: FederationContext;
+  readonly resource: string;
+  readonly requestedLimit?: number;
+  readonly nativeRbac: FederationNativeRbacEvaluator;
+}
+
+export interface FederationScopeQueryFilter {
+  readonly resource: FederationResource;
+  readonly subjectUserId: string;
+  readonly includePersonal: boolean;
+  readonly teamIds: readonly string[];
+  readonly limit: number;
+  readonly maxRowsPerQuery: number;
+}
+
+export interface FederationScopeAllowedResult {
+  readonly allowed: true;
+  readonly filter: FederationScopeQueryFilter;
+}
+
+export interface FederationScopeDeniedResult {
+  readonly allowed: false;
+  readonly deny: FederationScopeDenyReason;
+}
+
+export type FederationScopeEvaluationResult =
+  | FederationScopeAllowedResult
+  | FederationScopeDeniedResult;
+
+function isFederationResource(resource: string): resource is FederationResource {
+  return federationResourceSet.has(resource);
+}
+
+function uniqueStrings(values: readonly string[]): readonly string[] {
+  return Array.from(new Set<string>(values));
+}
+
+function intersectTeamIds(
+  nativeTeamIds: readonly string[],
+  scopedTeamIds: readonly string[] | undefined,
+): readonly string[] {
+  const uniqueNativeTeamIds = uniqueStrings(nativeTeamIds);
+
+  if (scopedTeamIds === undefined) {
+    return uniqueNativeTeamIds;
+  }
+
+  const nativeSet = new Set<string>(uniqueNativeTeamIds);
+  return uniqueStrings(scopedTeamIds).filter((teamId: string): boolean => nativeSet.has(teamId));
+}
+
+function makeDenyReason(params: {
+  readonly code: FederationScopeDenyCode;
+  readonly stage: FederationScopeDenyStage;
+  readonly statusCode?: FederationScopeDenyStatus;
+  readonly message: string;
+  readonly context: FederationContext;
+  readonly resource: string;
+  readonly details?: FederationScopeDenyDetails;
+}): FederationScopeDeniedResult {
+  return {
+    allowed: false,
+    deny: {
+      code: params.code,
+      stage: params.stage,
+      statusCode: params.statusCode ?? 403,
+      message: params.message,
+      grantId: params.context.grantId,
+      peerId: params.context.peerId,
+      subjectUserId: params.context.subjectUserId,
+      resource: params.resource,
+      ...(params.details !== undefined ? { details: params.details } : {}),
+    },
+  };
+}
+
+@Injectable()
+export class FederationScopeService {
+  async evaluateAccess(
+    input: FederationScopeEvaluationInput,
+  ): Promise<FederationScopeEvaluationResult> {
+    const { context, resource, requestedLimit, nativeRbac } = input;
+
+    let scope: ReturnType<typeof parseFederationScope>;
+    try {
+      scope = parseFederationScope(context.scope);
+    } catch (error: unknown) {
+      const message =
+        error instanceof FederationScopeError
+          ? 'Federation grant scope is invalid'
+          : 'Federation grant scope could not be parsed';
+      const details = error instanceof Error ? { reason: error.message } : undefined;
+      return makeDenyReason({
+        code: 'invalid_scope',
+        stage: 'scope_parse',
+        statusCode: 400,
+        message,
+        context,
+        resource,
+        ...(details !== undefined ? { details } : {}),
+      });
+    }
+
+    if (!isFederationResource(resource)) {
+      return makeDenyReason({
+        code: 'invalid_resource',
+        stage: 'resource_allowlist',
+        message: 'Requested federation resource is not supported',
+        context,
+        resource,
+        details: { supportedResources: FEDERATION_RESOURCE_VALUES },
+      });
+    }
+
+    if (scope.excluded_resources.includes(resource)) {
+      return makeDenyReason({
+        code: 'resource_excluded',
+        stage: 'resource_exclusion',
+        message: 'Requested federation resource is explicitly excluded by grant scope',
+        context,
+        resource,
+      });
+    }
+
+    if (!scope.resources.includes(resource)) {
+      return makeDenyReason({
+        code: 'resource_not_granted',
+        stage: 'resource_allowlist',
+        message: 'Requested federation resource is not granted by scope',
+        context,
+        resource,
+        details: { grantedResources: scope.resources },
+      });
+    }
+
+    if (requestedLimit !== undefined && (!Number.isInteger(requestedLimit) || requestedLimit < 1)) {
+      return makeDenyReason({
+        code: 'invalid_limit',
+        stage: 'row_cap',
+        statusCode: 400,
+        message: 'Requested row limit must be a positive integer',
+        context,
+        resource,
+        details: { requestedLimit },
+      });
+    }
+
+    const nativeResult = await nativeRbac.evaluateReadAccess({
+      grantId: context.grantId,
+      peerId: context.peerId,
+      subjectUserId: context.subjectUserId,
+      resource,
+    });
+
+    if (!nativeResult.allowed) {
+      return makeDenyReason({
+        code: 'native_rbac_denied',
+        stage: 'native_rbac',
+        message: nativeResult.reason ?? 'Subject user is not allowed to read this resource',
+        context,
+        resource,
+        ...(nativeResult.details !== undefined ? { details: nativeResult.details } : {}),
+      });
+    }
+
+    const scopeFilter = scope.filters?.[resource];
+    const includePersonal =
+      Boolean(scopeFilter?.include_personal ?? true) && nativeResult.access.includePersonal;
+    const teamIds = intersectTeamIds(nativeResult.access.teamIds, scopeFilter?.include_teams);
+    const limit = Math.min(requestedLimit ?? scope.max_rows_per_query, scope.max_rows_per_query);
+
+    return {
+      allowed: true,
+      filter: {
+        resource,
+        subjectUserId: context.subjectUserId,
+        includePersonal,
+        teamIds,
+        limit,
+        maxRowsPerQuery: scope.max_rows_per_query,
+      },
+    };
+  }
+}

apps/gateway/src/federation/server/verbs/__tests__/capabilities.controller.spec.ts

@@ -0,0 +1,88 @@
+import 'reflect-metadata';
+import { RequestMethod } from '@nestjs/common';
+import { describe, expect, it } from 'vitest';
+import type { FastifyRequest } from 'fastify';
+import { FederationCapabilitiesResponseSchema, FEDERATION_VERBS } from '@mosaicstack/types';
+import { FederationScopeError } from '../../../scope-schema.js';
+import { FederationAuthGuard } from '../../federation-auth.guard.js';
+import { CapabilitiesController } from '../capabilities.controller.js';
+
+const VALID_SCOPE = {
+  resources: ['tasks', 'notes'],
+  excluded_resources: ['credentials'],
+  max_rows_per_query: 250,
+} as const;
+
+const DEFAULTED_SCOPE = {
+  resources: ['memory'],
+  max_rows_per_query: 10,
+} as const;
+
+function makeRequest(scope: Record<string, unknown>): FastifyRequest {
+  return {
+    federationContext: {
+      grantId: 'grant-1',
+      peerId: 'peer-1',
+      subjectUserId: 'user-1',
+      scope,
+    },
+  } as FastifyRequest;
+}
+
+describe('CapabilitiesController', () => {
+  it('declares GET /api/federation/v1/capabilities', () => {
+    expect(Reflect.getMetadata('path', CapabilitiesController)).toBe(
+      'api/federation/v1/capabilities',
+    );
+    expect(Reflect.getMetadata('path', CapabilitiesController.prototype.getCapabilities)).toBe('/');
+    expect(Reflect.getMetadata('method', CapabilitiesController.prototype.getCapabilities)).toBe(
+      RequestMethod.GET,
+    );
+  });
+
+  it('is protected only by FederationAuthGuard', () => {
+    const guards = Reflect.getMetadata('__guards__', CapabilitiesController) as unknown[];
+
+    expect(guards).toEqual([FederationAuthGuard]);
+  });
+
+  it('returns resources, excluded resources, max rows, and M3 supported verbs from the active grant scope', () => {
+    const controller = new CapabilitiesController();
+
+    const response = controller.getCapabilities(makeRequest(VALID_SCOPE));
+
+    expect(response).toEqual({
+      resources: ['tasks', 'notes'],
+      excluded_resources: ['credentials'],
+      max_rows_per_query: 250,
+      supported_verbs: [...FEDERATION_VERBS],
+    });
+    expect(FederationCapabilitiesResponseSchema.safeParse(response).success).toBe(true);
+  });
+
+  it('applies scope defaults without RBAC or resource filtering', () => {
+    const controller = new CapabilitiesController();
+
+    const response = controller.getCapabilities(makeRequest(DEFAULTED_SCOPE));
+
+    expect(response).toEqual({
+      resources: ['memory'],
+      excluded_resources: [],
+      max_rows_per_query: 10,
+      supported_verbs: ['list', 'get', 'capabilities'],
+    });
+  });
+
+  it('rejects invalid scope state instead of returning an invalid capabilities contract', () => {
+    const controller = new CapabilitiesController();
+
+    expect(() =>
+      controller.getCapabilities(
+        makeRequest({
+          resources: [],
+          max_rows_per_query: 0,
+        }),
+      ),
+    ).toThrow(FederationScopeError);
+  });
+});

apps/gateway/src/federation/server/verbs/__tests__/list-query.service.spec.ts

@@ -0,0 +1,428 @@
+import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest';
+import {
+  createPgliteDb,
+  insights,
+  missionTasks,
+  missions,
+  preferences,
+  projects,
+  runPgliteMigrations,
+  teams,
+  users,
+  type Db,
+  type DbHandle,
+} from '@mosaicstack/db';
+import type { FederationScopeQueryFilter } from '../../scope.service.js';
+import { FederationListQueryService } from '../list-query.service.js';
+
+const TASK_FILTER: FederationScopeQueryFilter = {
+  resource: 'tasks',
+  subjectUserId: 'user-1',
+  includePersonal: true,
+  teamIds: [],
+  limit: 2,
+  maxRowsPerQuery: 2,
+};
+
+const SUBJECT_USER_ID = 'fed-m3-05-subject';
+const OTHER_USER_ID = 'fed-m3-05-other';
+const TEAM_ID = '05000000-0000-4000-8000-000000000001';
+const UNAUTHORIZED_TEAM_ID = '05000000-0000-4000-8000-000000000002';
+const PERSONAL_PROJECT_ID = '05000000-0000-4000-8000-000000000101';
+const TEAM_PROJECT_ID = '05000000-0000-4000-8000-000000000102';
+const UNAUTHORIZED_PROJECT_ID = '05000000-0000-4000-8000-000000000103';
+const PERSONAL_MISSION_ID = '05000000-0000-4000-8000-000000000201';
+const TEAM_MISSION_ID = '05000000-0000-4000-8000-000000000202';
+const UNAUTHORIZED_MISSION_ID = '05000000-0000-4000-8000-000000000203';
+const SUBJECT_TEAM_NOTE_ID = '05000000-0000-4000-8000-000000000301';
+const OTHER_TEAM_NOTE_ID = '05000000-0000-4000-8000-000000000302';
+const SUBJECT_PERSONAL_NOTE_ID = '05000000-0000-4000-8000-000000000303';
+const SUBJECT_UNAUTHORIZED_NOTE_ID = '05000000-0000-4000-8000-000000000304';
+const INSIGHT_ONE_ID = '05000000-0000-4000-8000-000000000401';
+const INSIGHT_TWO_ID = '05000000-0000-4000-8000-000000000402';
+const PREFERENCE_ONE_ID = '05000000-0000-4000-8000-000000000501';
+const PREFERENCE_TWO_ID = '05000000-0000-4000-8000-000000000502';
+
+let dbHandle: DbHandle | undefined;
+
+function makeService() {
+  return new FederationListQueryService({} as Db);
+}
+
+function makeDbService() {
+  if (!dbHandle) {
+    throw new Error('test DB not initialized');
+  }
+  return new FederationListQueryService(dbHandle.db);
+}
+
+async function seedNotesFixture() {
+  if (!dbHandle) {
+    throw new Error('test DB not initialized');
+  }
+
+  await dbHandle.db.insert(users).values([
+    {
+      id: SUBJECT_USER_ID,
+      name: 'Federation Subject',
+      email: `${SUBJECT_USER_ID}@example.test`,
+      emailVerified: false,
+    },
+    {
+      id: OTHER_USER_ID,
+      name: 'Federation Other',
+      email: `${OTHER_USER_ID}@example.test`,
+      emailVerified: false,
+    },
+  ]);
+
+  await dbHandle.db.insert(teams).values([
+    {
+      id: TEAM_ID,
+      name: 'FED-M3-05 Team',
+      slug: 'fed-m3-05-team',
+      ownerId: SUBJECT_USER_ID,
+      managerId: SUBJECT_USER_ID,
+    },
+    {
+      id: UNAUTHORIZED_TEAM_ID,
+      name: 'FED-M3-05 Unauthorized Team',
+      slug: 'fed-m3-05-unauthorized-team',
+      ownerId: OTHER_USER_ID,
+      managerId: OTHER_USER_ID,
+    },
+  ]);
+
+  await dbHandle.db.insert(projects).values([
+    {
+      id: PERSONAL_PROJECT_ID,
+      name: 'FED-M3-05 Personal Project',
+      ownerId: SUBJECT_USER_ID,
+      ownerType: 'user',
+    },
+    {
+      id: TEAM_PROJECT_ID,
+      name: 'FED-M3-05 Team Project',
+      teamId: TEAM_ID,
+      ownerType: 'team',
+    },
+    {
+      id: UNAUTHORIZED_PROJECT_ID,
+      name: 'FED-M3-05 Unauthorized Project',
+      teamId: UNAUTHORIZED_TEAM_ID,
+      ownerType: 'team',
+    },
+  ]);
+
+  await dbHandle.db.insert(missions).values([
+    {
+      id: PERSONAL_MISSION_ID,
+      name: 'FED-M3-05 Personal Mission',
+      projectId: PERSONAL_PROJECT_ID,
+      userId: SUBJECT_USER_ID,
+    },
+    {
+      id: TEAM_MISSION_ID,
+      name: 'FED-M3-05 Team Mission',
+      projectId: TEAM_PROJECT_ID,
+      userId: SUBJECT_USER_ID,
+    },
+    {
+      id: UNAUTHORIZED_MISSION_ID,
+      name: 'FED-M3-05 Unauthorized Mission',
+      projectId: UNAUTHORIZED_PROJECT_ID,
+      userId: SUBJECT_USER_ID,
+    },
+  ]);
+
+  await dbHandle.db.insert(missionTasks).values([
+    {
+      id: SUBJECT_TEAM_NOTE_ID,
+      missionId: TEAM_MISSION_ID,
+      userId: SUBJECT_USER_ID,
+      notes: 'subject note on team mission',
+      createdAt: new Date('2026-06-24T03:00:00.000Z'),
+      updatedAt: new Date('2026-06-24T03:00:00.000Z'),
+    },
+    {
+      id: OTHER_TEAM_NOTE_ID,
+      missionId: TEAM_MISSION_ID,
+      userId: OTHER_USER_ID,
+      notes: 'other user note on team mission',
+      createdAt: new Date('2026-06-24T02:00:00.000Z'),
+      updatedAt: new Date('2026-06-24T02:00:00.000Z'),
+    },
+    {
+      id: SUBJECT_PERSONAL_NOTE_ID,
+      missionId: PERSONAL_MISSION_ID,
+      userId: SUBJECT_USER_ID,
+      notes: 'subject note on personal mission',
+      createdAt: new Date('2026-06-24T01:00:00.000Z'),
+      updatedAt: new Date('2026-06-24T01:00:00.000Z'),
+    },
+    {
+      id: SUBJECT_UNAUTHORIZED_NOTE_ID,
+      missionId: UNAUTHORIZED_MISSION_ID,
+      userId: SUBJECT_USER_ID,
+      notes: 'subject note outside grant-visible missions',
+      createdAt: new Date('2026-06-24T04:00:00.000Z'),
+      updatedAt: new Date('2026-06-24T04:00:00.000Z'),
+    },
+  ]);
+
+  const memoryCreatedAt = new Date('2026-06-24T05:00:00.000Z');
+  await dbHandle.db.insert(insights).values([
+    {
+      id: INSIGHT_ONE_ID,
+      userId: SUBJECT_USER_ID,
+      content: 'first insight',
+      source: 'agent',
+      createdAt: memoryCreatedAt,
+      updatedAt: memoryCreatedAt,
+    },
+    {
+      id: INSIGHT_TWO_ID,
+      userId: SUBJECT_USER_ID,
+      content: 'second insight',
+      source: 'agent',
+      createdAt: memoryCreatedAt,
+      updatedAt: memoryCreatedAt,
+    },
+  ]);
+
+  await dbHandle.db.insert(preferences).values([
+    {
+      id: PREFERENCE_ONE_ID,
+      userId: SUBJECT_USER_ID,
+      key: 'fed-m3-05-pref-1',
+      value: { enabled: true },
+      createdAt: memoryCreatedAt,
+      updatedAt: memoryCreatedAt,
+    },
+    {
+      id: PREFERENCE_TWO_ID,
+      userId: SUBJECT_USER_ID,
+      key: 'fed-m3-05-pref-2',
+      value: { enabled: false },
+      createdAt: memoryCreatedAt,
+      updatedAt: memoryCreatedAt,
+    },
+  ]);
+}
+
+function stubRows(
+  service: FederationListQueryService,
+  ...pages: Array<Array<Record<string, unknown>>>
+) {
+  const mock = vi.fn();
+  for (const page of pages) {
+    mock.mockResolvedValueOnce(page);
+  }
+  (
+    service as unknown as {
+      listAllRows: (
+        _filter: FederationScopeQueryFilter,
+        _rowLimit: number,
+        _cursor: unknown,
+      ) => Promise<Array<Record<string, unknown>>>;
+    }
+  ).listAllRows = mock;
+  return mock;
+}
+
+describe('FederationListQueryService', () => {
+  beforeAll(async () => {
+    dbHandle = createPgliteDb(`memory://fed-m3-05-list-${Date.now()}`);
+    await runPgliteMigrations(dbHandle);
+    await seedNotesFixture();
+  });
+
+  afterAll(async () => {
+    await dbHandle?.close();
+    dbHandle = undefined;
+  });
+
+  it('denies sensitive resources in native RBAC for M3 list reads', async () => {
+    const service = makeService();
+
+    await expect(
+      service.evaluateReadAccess({
+        grantId: 'grant-1',
+        peerId: 'peer-1',
+        subjectUserId: 'user-1',
+        resource: 'credentials',
+      }),
+    ).resolves.toMatchObject({
+      allowed: false,
+      reason: 'credentials federation list access is not implemented in M3',
+    });
+  });
+
+  it('allows personal memory reads without requiring team lookup', async () => {
+    const service = makeService();
+
+    await expect(
+      service.evaluateReadAccess({
+        grantId: 'grant-1',
+        peerId: 'peer-1',
+        subjectUserId: 'user-1',
+        resource: 'memory',
+      }),
+    ).resolves.toEqual({
+      allowed: true,
+      access: { includePersonal: true, teamIds: [] },
+    });
+  });
+
+  it('applies the scope row cap and returns an opaque next cursor when truncated', async () => {
+    const service = makeService();
+    const listAllRows = stubRows(
+      service,
+      [
+        { id: '3', createdAt: new Date('2026-06-24T03:00:00.000Z') },
+        { id: '2', createdAt: new Date('2026-06-24T02:00:00.000Z') },
+        { id: '1', createdAt: new Date('2026-06-24T01:00:00.000Z') },
+      ],
+      [{ id: '1', createdAt: new Date('2026-06-24T01:00:00.000Z') }],
+    );
+
+    const firstPage = await service.list({ filter: TASK_FILTER });
+
+    expect(firstPage).toEqual({
+      items: [
+        { id: '3', createdAt: new Date('2026-06-24T03:00:00.000Z') },
+        { id: '2', createdAt: new Date('2026-06-24T02:00:00.000Z') },
+      ],
+      truncated: true,
+      nextCursor: expect.any(String),
+    });
+
+    expect(listAllRows).toHaveBeenNthCalledWith(1, TASK_FILTER, 3, undefined);
+
+    const secondPage = await service.list({ filter: TASK_FILTER, cursor: firstPage.nextCursor });
+    expect(secondPage).toEqual({
+      items: [{ id: '1', createdAt: new Date('2026-06-24T01:00:00.000Z') }],
+      truncated: false,
+    });
+    expect(listAllRows).toHaveBeenNthCalledWith(
+      2,
+      TASK_FILTER,
+      3,
+      expect.objectContaining({ id: '2' }),
+    );
+  });
+
+  it('rejects invalid cursors instead of falling back to the first page', async () => {
+    const service = makeService();
+    stubRows(service, [{ id: '1' }]);
+
+    await expect(service.list({ filter: TASK_FILTER, cursor: 'not-base64-json' })).rejects.toThrow(
+      'Invalid federation list cursor',
+    );
+  });
+
+  it('throws when a truncated page cannot encode a resumable cursor', async () => {
+    const service = makeService();
+    stubRows(service, [
+      { id: '2', createdAt: 'not-a-date' },
+      { id: '1', createdAt: 'not-a-date' },
+    ]);
+
+    await expect(service.list({ filter: { ...TASK_FILTER, limit: 1 } })).rejects.toThrow(
+      'Federation list cursor cannot be encoded',
+    );
+  });
+
+  it('throws on unsupported resources instead of crashing pagination', async () => {
+    const service = makeService();
+
+    await expect(
+      service.list({
+        filter: {
+          ...TASK_FILTER,
+          resource: 'unknown-resource' as FederationScopeQueryFilter['resource'],
+        },
+      }),
+    ).rejects.toThrow('Unsupported federation list resource');
+  });
+
+  it('does not leak another user mission task notes through team-scoped note reads', async () => {
+    const service = makeDbService();
+
+    const result = await service.list({
+      filter: {
+        resource: 'notes',
+        subjectUserId: SUBJECT_USER_ID,
+        includePersonal: false,
+        teamIds: [TEAM_ID],
+        limit: 10,
+        maxRowsPerQuery: 10,
+      },
+    });
+
+    const ids = result.items.map((item) => item['id']);
+    expect(ids).toEqual([SUBJECT_TEAM_NOTE_ID]);
+    expect(ids).not.toContain(OTHER_TEAM_NOTE_ID);
+  });
+
+  it('does not return subject personal mission task notes when includePersonal is false', async () => {
+    const service = makeDbService();
+
+    const result = await service.list({
+      filter: {
+        resource: 'notes',
+        subjectUserId: SUBJECT_USER_ID,
+        includePersonal: false,
+        teamIds: [TEAM_ID],
+        limit: 10,
+        maxRowsPerQuery: 10,
+      },
+    });
+
+    expect(result.items.map((item) => item['id'])).not.toContain(SUBJECT_PERSONAL_NOTE_ID);
+  });
+
+  it('does not return subject notes from missions outside the grant-visible project set', async () => {
+    const service = makeDbService();
+
+    const result = await service.list({
+      filter: {
+        resource: 'notes',
+        subjectUserId: SUBJECT_USER_ID,
+        includePersonal: true,
+        teamIds: [TEAM_ID],
+        limit: 10,
+        maxRowsPerQuery: 10,
+      },
+    });
+
+    const ids = result.items.map((item) => item['id']);
+    expect(ids).toContain(SUBJECT_PERSONAL_NOTE_ID);
+    expect(ids).toContain(SUBJECT_TEAM_NOTE_ID);
+    expect(ids).not.toContain(SUBJECT_UNAUTHORIZED_NOTE_ID);
+    expect(ids).not.toContain(OTHER_TEAM_NOTE_ID);
+  });
+
+  it('paginates memory deterministically across insights and preferences', async () => {
+    const service = makeDbService();
+    const filter: FederationScopeQueryFilter = {
+      resource: 'memory',
+      subjectUserId: SUBJECT_USER_ID,
+      includePersonal: true,
+      teamIds: [],
+      limit: 2,
+      maxRowsPerQuery: 2,
+    };
+
+    const firstPage = await service.list({ filter });
+    const secondPage = await service.list({ filter, cursor: firstPage.nextCursor });
+    const firstPageIds = firstPage.items.map((item) => item['id']);
+    const secondPageIds = secondPage.items.map((item) => item['id']);
+    const allIds = [...firstPageIds, ...secondPageIds];
+
+    expect(firstPage).toMatchObject({ truncated: true, nextCursor: expect.any(String) });
+    expect(firstPageIds).toEqual([INSIGHT_TWO_ID, INSIGHT_ONE_ID]);
+    expect(secondPageIds).toEqual([PREFERENCE_TWO_ID, PREFERENCE_ONE_ID]);
+    expect(new Set(allIds).size).toBe(allIds.length);
+  });
+});

apps/gateway/src/federation/server/verbs/__tests__/list.controller.spec.ts

@@ -0,0 +1,188 @@
+import 'reflect-metadata';
+import { RequestMethod } from '@nestjs/common';
+import type { FastifyRequest } from 'fastify';
+import { beforeEach, describe, expect, it, vi } from 'vitest';
+import { FederationAuthGuard } from '../../federation-auth.guard.js';
+import type {
+  FederationScopeEvaluationResult,
+  FederationScopeQueryFilter,
+} from '../../scope.service.js';
+import { ListController } from '../list.controller.js';
+import type { FederationListQueryResult } from '../list-query.service.js';
+
+const FEDERATION_CONTEXT = {
+  grantId: 'grant-1',
+  peerId: 'peer-1',
+  subjectUserId: 'user-1',
+  scope: { resources: ['tasks'], max_rows_per_query: 25 },
+};
+
+const TASK_FILTER: FederationScopeQueryFilter = {
+  resource: 'tasks',
+  subjectUserId: 'user-1',
+  includePersonal: true,
+  teamIds: ['team-1'],
+  limit: 10,
+  maxRowsPerQuery: 25,
+};
+
+function makeRequest(): FastifyRequest {
+  return { federationContext: FEDERATION_CONTEXT } as unknown as FastifyRequest;
+}
+
+function allowedScope(
+  filter: FederationScopeQueryFilter = TASK_FILTER,
+): FederationScopeEvaluationResult {
+  return { allowed: true, filter };
+}
+
+function makeController(opts?: {
+  scopeResult?: FederationScopeEvaluationResult;
+  queryResult?: FederationListQueryResult;
+}) {
+  const scope = {
+    evaluateAccess: vi.fn().mockResolvedValue(opts?.scopeResult ?? allowedScope()),
+  };
+  const query = {
+    evaluateReadAccess: vi.fn(),
+    list: vi.fn().mockResolvedValue(
+      opts?.queryResult ?? {
+        items: [
+          {
+            id: 'task-1',
+            title: 'Federated task',
+            createdAt: new Date('2026-06-24T00:00:00.000Z'),
+          },
+        ],
+        truncated: false,
+      },
+    ),
+  };
+
+  return {
+    controller: new ListController(scope as never, query as never),
+    scope,
+    query,
+  };
+}
+
+describe('ListController', () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+  });
+
+  it('declares POST /api/federation/v1/list/:resource protected only by FederationAuthGuard', () => {
+    expect(Reflect.getMetadata('path', ListController)).toBe('api/federation/v1/list');
+    expect(Reflect.getMetadata('path', ListController.prototype.list)).toBe(':resource');
+    expect(Reflect.getMetadata('method', ListController.prototype.list)).toBe(RequestMethod.POST);
+    expect(Reflect.getMetadata('__guards__', ListController)).toEqual([FederationAuthGuard]);
+  });
+
+  it('runs AuthGuard context through ScopeService and returns local-source tagged rows', async () => {
+    const { controller, scope, query } = makeController();
+
+    const response = await controller.list('tasks', makeRequest(), { limit: 10 });
+
+    expect(scope.evaluateAccess).toHaveBeenCalledWith({
+      context: FEDERATION_CONTEXT,
+      resource: 'tasks',
+      requestedLimit: 10,
+      nativeRbac: query,
+    });
+    expect(query.list).toHaveBeenCalledWith({ filter: TASK_FILTER, cursor: undefined });
+    expect(response).toEqual({
+      items: [
+        {
+          id: 'task-1',
+          title: 'Federated task',
+          createdAt: new Date('2026-06-24T00:00:00.000Z'),
+          _source: 'local',
+        },
+      ],
+    });
+  });
+
+  it('preserves pagination metadata when row cap truncates the query layer result', async () => {
+    const { controller } = makeController({
+      queryResult: {
+        items: [{ id: 'task-1' }],
+        nextCursor: 'cursor-2',
+        truncated: true,
+      },
+    });
+
+    const response = await controller.list('tasks', makeRequest(), { cursor: 'cursor-1' });
+
+    expect(response).toEqual({
+      items: [{ id: 'task-1', _source: 'local' }],
+      nextCursor: 'cursor-2',
+      _truncated: true,
+    });
+  });
+
+  it('returns a federation error envelope when auth guard context is missing', async () => {
+    const { controller, scope, query } = makeController();
+
+    await expect(
+      controller.list('tasks', {} as unknown as FastifyRequest, {}),
+    ).rejects.toMatchObject({
+      response: {
+        error: {
+          code: 'unauthorized',
+          message: 'Federation context missing',
+        },
+      },
+      status: 401,
+    });
+    expect(scope.evaluateAccess).not.toHaveBeenCalled();
+    expect(query.list).not.toHaveBeenCalled();
+  });
+
+  it('returns a federation error envelope when scope evaluation denies access', async () => {
+    const { controller, query } = makeController({
+      scopeResult: {
+        allowed: false,
+        deny: {
+          code: 'resource_excluded',
+          stage: 'resource_exclusion',
+          statusCode: 403,
+          message: 'Requested federation resource is explicitly excluded by grant scope',
+          grantId: 'grant-1',
+          peerId: 'peer-1',
+          subjectUserId: 'user-1',
+          resource: 'credentials',
+        },
+      },
+    });
+
+    await expect(controller.list('credentials', makeRequest(), {})).rejects.toMatchObject({
+      response: {
+        error: {
+          code: 'scope_violation',
+          message: 'Requested federation resource is explicitly excluded by grant scope',
+        },
+      },
+      status: 403,
+    });
+    expect(query.list).not.toHaveBeenCalled();
+  });
+
+  it('rejects malformed request body fields before querying storage', async () => {
+    const { controller, scope, query } = makeController();
+
+    await expect(controller.list('tasks', makeRequest(), { cursor: 123 })).rejects.toMatchObject({
+      response: { error: { code: 'invalid_request' } },
+      status: 400,
+    });
+    await expect(controller.list('tasks', makeRequest(), { limit: false })).rejects.toMatchObject({
+      response: { error: { code: 'invalid_request' } },
+      status: 400,
+    });
+    await expect(controller.list('tasks', makeRequest(), { limit: 'abc' })).rejects.toMatchObject({
+      response: { error: { code: 'invalid_request' } },
+      status: 400,
+    });
+    expect(scope.evaluateAccess).not.toHaveBeenCalled();
+    expect(query.list).not.toHaveBeenCalled();
+  });
+});

apps/gateway/src/federation/server/verbs/capabilities.controller.ts

@@ -0,0 +1,38 @@
+/**
+ * Federation capabilities verb (FED-M3-07).
+ *
+ * Returns the read-only capability envelope for the active grant attached by
+ * FederationAuthGuard. This endpoint intentionally does not invoke native RBAC
+ * or ScopeService: an active grant is enough to ask what the grant allows.
+ */
+
+import { Controller, Get, Req, UseGuards } from '@nestjs/common';
+import type { FastifyRequest } from 'fastify';
+import {
+  FEDERATION_VERBS,
+  type FederationCapabilitiesResponse,
+  type FederationVerb,
+} from '@mosaicstack/types';
+import { parseFederationScope } from '../../scope-schema.js';
+import { FederationAuthGuard } from '../federation-auth.guard.js';
+import '../federation-context.js';
+
+@Controller('api/federation/v1/capabilities')
+@UseGuards(FederationAuthGuard)
+export class CapabilitiesController {
+  @Get()
+  getCapabilities(@Req() request: FastifyRequest): FederationCapabilitiesResponse {
+    if (!request.federationContext) {
+      throw new Error('Federation context missing after auth guard');
+    }
+
+    const scope = parseFederationScope(request.federationContext.scope);
+
+    return {
+      resources: [...scope.resources],
+      excluded_resources: [...scope.excluded_resources],
+      max_rows_per_query: scope.max_rows_per_query,
+      supported_verbs: [...FEDERATION_VERBS] satisfies FederationVerb[],
+    };
+  }
+}

apps/gateway/src/federation/server/verbs/list-query.service.ts

@@ -0,0 +1,408 @@
+/**
+ * Federation list query layer (FED-M3-05).
+ *
+ * Read-only DB adapter used by ListController after FederationAuthGuard and
+ * FederationScopeService have established the subject user, allowed resource,
+ * native-RBAC intersection, and row cap. Audit writes are intentionally
+ * deferred to M4.
+ */
+
+import { Inject, Injectable } from '@nestjs/common';
+import {
+  and,
+  desc,
+  eq,
+  inArray,
+  insights,
+  isNotNull,
+  lt,
+  missionTasks,
+  missions,
+  or,
+  preferences,
+  projects,
+  tasks,
+  teamMembers,
+  type Db,
+} from '@mosaicstack/db';
+import type {
+  FederationNativeRbacEvaluator,
+  FederationNativeRbacRequest,
+  FederationNativeRbacResult,
+  FederationScopeQueryFilter,
+} from '../scope.service.js';
+import { DB } from '../../../database/database.module.js';
+
+export interface FederationListQueryRequest {
+  readonly filter: FederationScopeQueryFilter;
+  readonly cursor?: string;
+}
+
+export interface FederationListQueryResult<T extends object = Record<string, unknown>> {
+  readonly items: T[];
+  readonly nextCursor?: string;
+  readonly truncated: boolean;
+}
+
+type CursorSource = 'insights' | 'preferences';
+const CURSOR_SOURCE = Symbol('federationCursorSource');
+
+type RowObject = Record<string, unknown> & { readonly [CURSOR_SOURCE]?: CursorSource };
+
+interface KeysetCursor {
+  readonly createdAt: Date;
+  readonly id: string;
+  readonly source?: CursorSource;
+}
+
+function encodeCursor(row: RowObject): string {
+  const createdAt = row['createdAt'];
+  const id = row['id'];
+  if (!(createdAt instanceof Date) || typeof id !== 'string') {
+    throw new Error('Federation list cursor cannot be encoded');
+  }
+
+  const source = row[CURSOR_SOURCE];
+  return Buffer.from(
+    JSON.stringify({ createdAt: createdAt.toISOString(), id, ...(source ? { source } : {}) }),
+    'utf8',
+  ).toString('base64url');
+}
+
+function decodeCursor(cursor: string | undefined): KeysetCursor | undefined {
+  if (cursor === undefined) {
+    return undefined;
+  }
+
+  try {
+    const parsed = JSON.parse(Buffer.from(cursor, 'base64url').toString('utf8')) as unknown;
+    if (typeof parsed !== 'object' || parsed === null) {
+      throw new Error('cursor must be an object');
+    }
+
+    const { createdAt, id, source } = parsed as {
+      createdAt?: unknown;
+      id?: unknown;
+      source?: unknown;
+    };
+    if (typeof createdAt !== 'string' || typeof id !== 'string' || id.length === 0) {
+      throw new Error('cursor is missing createdAt or id');
+    }
+    if (source !== undefined && source !== 'insights' && source !== 'preferences') {
+      throw new Error('cursor source is invalid');
+    }
+
+    const date = new Date(createdAt);
+    if (Number.isNaN(date.getTime())) {
+      throw new Error('cursor createdAt is invalid');
+    }
+
+    return { createdAt: date, id, ...(source ? { source } : {}) };
+  } catch {
+    throw new Error('Invalid federation list cursor');
+  }
+}
+
+function paginate<T extends RowObject>(rows: T[], limit: number): FederationListQueryResult<T> {
+  const page = rows.slice(0, limit);
+  const hasMore = rows.length > limit;
+  const nextCursor = hasMore ? encodeCursor(page[page.length - 1] ?? {}) : undefined;
+
+  return {
+    items: page,
+    truncated: hasMore,
+    ...(nextCursor !== undefined ? { nextCursor } : {}),
+  };
+}
+
+function markCursorSource<T extends RowObject>(row: T, source: CursorSource): T {
+  Object.defineProperty(row, CURSOR_SOURCE, {
+    value: source,
+    enumerable: false,
+    configurable: false,
+  });
+  return row;
+}
+
+function sortRows(rows: RowObject[]): RowObject[] {
+  return [...rows].sort((a, b) => {
+    const aTime = a['createdAt'] instanceof Date ? a['createdAt'].getTime() : 0;
+    const bTime = b['createdAt'] instanceof Date ? b['createdAt'].getTime() : 0;
+    if (aTime !== bTime) {
+      return bTime - aTime;
+    }
+    return String(b['id'] ?? '').localeCompare(String(a['id'] ?? ''));
+  });
+}
+
+@Injectable()
+export class FederationListQueryService implements FederationNativeRbacEvaluator {
+  constructor(@Inject(DB) private readonly db: Db) {}
+
+  async evaluateReadAccess(
+    request: FederationNativeRbacRequest,
+  ): Promise<FederationNativeRbacResult> {
+    if (request.resource === 'credentials' || request.resource === 'api_keys') {
+      return {
+        allowed: false,
+        reason: `${request.resource} federation list access is not implemented in M3`,
+        details: { resource: request.resource },
+      };
+    }
+
+    if (request.resource === 'memory') {
+      return { allowed: true, access: { includePersonal: true, teamIds: [] } };
+    }
+
+    const teamIds = await this.listSubjectTeamIds(request.subjectUserId);
+    return { allowed: true, access: { includePersonal: true, teamIds } };
+  }
+
+  async list<T extends RowObject = RowObject>(
+    request: FederationListQueryRequest,
+  ): Promise<FederationListQueryResult<T>> {
+    const cursor = decodeCursor(request.cursor);
+    const rows = await this.listAllRows(request.filter, request.filter.limit + 1, cursor);
+    return paginate(rows as T[], request.filter.limit);
+  }
+
+  private async listAllRows(
+    filter: FederationScopeQueryFilter,
+    rowLimit: number,
+    cursor: KeysetCursor | undefined,
+  ): Promise<RowObject[]> {
+    switch (filter.resource) {
+      case 'tasks':
+        return this.listTasks(filter, rowLimit, cursor);
+      case 'notes':
+        return this.listNotes(filter, rowLimit, cursor);
+      case 'memory':
+        return this.listMemory(filter, rowLimit, cursor);
+      case 'credentials':
+      case 'api_keys':
+        return [];
+      default:
+        throw new Error(`Unsupported federation list resource: ${String(filter.resource)}`);
+    }
+  }
+
+  private async listSubjectTeamIds(subjectUserId: string): Promise<string[]> {
+    const rows = await this.db
+      .select({ teamId: teamMembers.teamId })
+      .from(teamMembers)
+      .where(eq(teamMembers.userId, subjectUserId));
+
+    return rows.map((row) => row.teamId);
+  }
+
+  private async listAccessibleProjectIds(filter: FederationScopeQueryFilter): Promise<string[]> {
+    const clauses = [];
+    if (filter.includePersonal) {
+      clauses.push(and(eq(projects.ownerType, 'user'), eq(projects.ownerId, filter.subjectUserId)));
+    }
+    if (filter.teamIds.length > 0) {
+      clauses.push(
+        and(eq(projects.ownerType, 'team'), inArray(projects.teamId, [...filter.teamIds])),
+      );
+    }
+
+    if (clauses.length === 0) {
+      return [];
+    }
+
+    const rows = await this.db
+      .select({ id: projects.id })
+      .from(projects)
+      .where(clauses.length === 1 ? clauses[0] : or(...clauses));
+
+    return rows.map((row) => row.id);
+  }
+
+  private async listMissionIds(projectIds: readonly string[]): Promise<string[]> {
+    if (projectIds.length === 0) {
+      return [];
+    }
+
+    const rows = await this.db
+      .select({ id: missions.id })
+      .from(missions)
+      .where(inArray(missions.projectId, [...projectIds]));
+
+    return rows.map((row) => row.id);
+  }
+
+  private async listTasks(
+    filter: FederationScopeQueryFilter,
+    rowLimit: number,
+    cursor: KeysetCursor | undefined,
+  ): Promise<RowObject[]> {
+    const projectIds = await this.listAccessibleProjectIds(filter);
+    const missionIds = await this.listMissionIds(projectIds);
+    const clauses = [];
+
+    if (projectIds.length > 0) {
+      clauses.push(inArray(tasks.projectId, projectIds));
+    }
+    if (missionIds.length > 0) {
+      clauses.push(inArray(tasks.missionId, missionIds));
+    }
+
+    if (clauses.length === 0) {
+      return [];
+    }
+
+    const scopeClause = clauses.length === 1 ? clauses[0] : or(...clauses);
+    const cursorClause = cursor
+      ? or(
+          lt(tasks.createdAt, cursor.createdAt),
+          and(eq(tasks.createdAt, cursor.createdAt), lt(tasks.id, cursor.id)),
+        )
+      : undefined;
+
+    const rows = await this.db
+      .select({
+        id: tasks.id,
+        title: tasks.title,
+        description: tasks.description,
+        status: tasks.status,
+        priority: tasks.priority,
+        projectId: tasks.projectId,
+        missionId: tasks.missionId,
+        assignee: tasks.assignee,
+        tags: tasks.tags,
+        dueDate: tasks.dueDate,
+        metadata: tasks.metadata,
+        createdAt: tasks.createdAt,
+        updatedAt: tasks.updatedAt,
+      })
+      .from(tasks)
+      .where(and(scopeClause, cursorClause))
+      .orderBy(desc(tasks.createdAt), desc(tasks.id))
+      .limit(rowLimit);
+
+    return sortRows(rows as RowObject[]);
+  }
+
+  private async listNotes(
+    filter: FederationScopeQueryFilter,
+    rowLimit: number,
+    cursor: KeysetCursor | undefined,
+  ): Promise<RowObject[]> {
+    const projectIds = await this.listAccessibleProjectIds(filter);
+    const missionIds = await this.listMissionIds(projectIds);
+
+    if (missionIds.length === 0) {
+      return [];
+    }
+
+    // mission_tasks rows are user-scoped even when the mission belongs to a team.
+    // Team visibility can narrow the mission set, but it must never widen the
+    // query to other users' mission task notes.
+    const scopeClause = and(
+      eq(missionTasks.userId, filter.subjectUserId),
+      inArray(missionTasks.missionId, missionIds),
+    );
+    const cursorClause = cursor
+      ? or(
+          lt(missionTasks.createdAt, cursor.createdAt),
+          and(eq(missionTasks.createdAt, cursor.createdAt), lt(missionTasks.id, cursor.id)),
+        )
+      : undefined;
+
+    const rows = await this.db
+      .select({
+        id: missionTasks.id,
+        missionId: missionTasks.missionId,
+        taskId: missionTasks.taskId,
+        status: missionTasks.status,
+        content: missionTasks.notes,
+        createdAt: missionTasks.createdAt,
+        updatedAt: missionTasks.updatedAt,
+      })
+      .from(missionTasks)
+      .where(and(scopeClause, cursorClause, isNotNull(missionTasks.notes)))
+      .orderBy(desc(missionTasks.createdAt), desc(missionTasks.id))
+      .limit(rowLimit);
+
+    return sortRows(rows.filter((row) => row.content !== '') as RowObject[]);
+  }
+
+  private async listMemory(
+    filter: FederationScopeQueryFilter,
+    rowLimit: number,
+    cursor: KeysetCursor | undefined,
+  ): Promise<RowObject[]> {
+    if (!filter.includePersonal) {
+      return [];
+    }
+    if (cursor && cursor.source === undefined) {
+      throw new Error('Invalid federation list cursor');
+    }
+
+    const rows: RowObject[] = [];
+
+    // Memory spans two physical tables. To keep pagination deterministic and
+    // resumable without a SQL UNION, M3 emits a fixed block order: all insights
+    // first, then preferences. The opaque cursor records which table produced
+    // the boundary row, so the next page never re-applies one table's keyset to
+    // the other table (which could duplicate/skip rows at equal timestamps).
+    if (cursor?.source !== 'preferences') {
+      const insightCursorClause = cursor
+        ? or(
+            lt(insights.createdAt, cursor.createdAt),
+            and(eq(insights.createdAt, cursor.createdAt), lt(insights.id, cursor.id)),
+          )
+        : undefined;
+      const insightRows = await this.db
+        .select({
+          id: insights.id,
+          kind: insights.source,
+          content: insights.content,
+          category: insights.category,
+          relevanceScore: insights.relevanceScore,
+          metadata: insights.metadata,
+          createdAt: insights.createdAt,
+          updatedAt: insights.updatedAt,
+        })
+        .from(insights)
+        .where(and(eq(insights.userId, filter.subjectUserId), insightCursorClause))
+        .orderBy(desc(insights.createdAt), desc(insights.id))
+        .limit(rowLimit);
+
+      rows.push(...(insightRows as RowObject[]).map((row) => markCursorSource(row, 'insights')));
+    }
+
+    const remaining = rowLimit - rows.length;
+    if (remaining <= 0) {
+      return rows;
+    }
+
+    const preferenceCursorClause =
+      cursor?.source === 'preferences'
+        ? or(
+            lt(preferences.createdAt, cursor.createdAt),
+            and(eq(preferences.createdAt, cursor.createdAt), lt(preferences.id, cursor.id)),
+          )
+        : undefined;
+    const preferenceRows = await this.db
+      .select({
+        id: preferences.id,
+        kind: preferences.category,
+        key: preferences.key,
+        value: preferences.value,
+        source: preferences.source,
+        mutable: preferences.mutable,
+        createdAt: preferences.createdAt,
+        updatedAt: preferences.updatedAt,
+      })
+      .from(preferences)
+      .where(and(eq(preferences.userId, filter.subjectUserId), preferenceCursorClause))
+      .orderBy(desc(preferences.createdAt), desc(preferences.id))
+      .limit(remaining);
+
+    rows.push(
+      ...(preferenceRows as RowObject[]).map((row) => markCursorSource(row, 'preferences')),
+    );
+    return rows;
+  }
+}

apps/gateway/src/federation/server/verbs/list.controller.ts

@@ -0,0 +1,147 @@
+/**
+ * Federation list verb (FED-M3-05).
+ *
+ * POST /api/federation/v1/list/:resource
+ *
+ * Pipeline: FederationAuthGuard attaches the active grant context, then
+ * FederationScopeService enforces grant scope + native RBAC intersection, then
+ * the read-only query layer returns capped rows tagged with `_source`. Read
+ * audit-log writes are deferred to M4; this controller does not persist request
+ * or response bodies.
+ */
+
+import {
+  Body,
+  Controller,
+  HttpException,
+  Inject,
+  Param,
+  Post,
+  Req,
+  UseGuards,
+} from '@nestjs/common';
+import type { FastifyRequest } from 'fastify';
+import {
+  FederationInvalidRequestError,
+  FederationScopeViolationError,
+  FederationUnauthorizedError,
+  SOURCE_LOCAL,
+  tagWithSource,
+  type FederationListResponse,
+  type SourceTag,
+} from '@mosaicstack/types';
+import { FederationAuthGuard } from '../federation-auth.guard.js';
+import '../federation-context.js';
+import { FederationScopeService } from '../scope.service.js';
+import { FederationListQueryService } from './list-query.service.js';
+
+interface FederationListRequestBody {
+  readonly limit?: unknown;
+  readonly cursor?: unknown;
+}
+
+type FederatedRow = Record<string, unknown> & SourceTag;
+
+function parseLimit(body: FederationListRequestBody | undefined): number | undefined {
+  if (body?.limit === undefined) {
+    return undefined;
+  }
+
+  const parsed =
+    typeof body.limit === 'number'
+      ? body.limit
+      : typeof body.limit === 'string' && body.limit.trim().length > 0
+        ? Number(body.limit)
+        : Number.NaN;
+
+  if (!Number.isSafeInteger(parsed) || parsed < 1) {
+    throw new HttpException(
+      new FederationInvalidRequestError(
+        'Federation list limit must be a positive integer',
+      ).toEnvelope(),
+      400,
+    );
+  }
+
+  return parsed;
+}
+
+function parseCursor(body: FederationListRequestBody | undefined): string | undefined {
+  if (body?.cursor === undefined) {
+    return undefined;
+  }
+  if (typeof body.cursor === 'string') {
+    return body.cursor;
+  }
+  throw new HttpException(
+    new FederationInvalidRequestError('Federation list cursor must be a string').toEnvelope(),
+    400,
+  );
+}
+
+@Controller('api/federation/v1/list')
+@UseGuards(FederationAuthGuard)
+export class ListController {
+  constructor(
+    @Inject(FederationScopeService) private readonly scope: FederationScopeService,
+    @Inject(FederationListQueryService) private readonly query: FederationListQueryService,
+  ) {}
+
+  @Post(':resource')
+  async list(
+    @Param('resource') resource: string,
+    @Req() request: FastifyRequest,
+    @Body() body?: FederationListRequestBody,
+  ): Promise<FederationListResponse<FederatedRow>> {
+    if (!request.federationContext) {
+      throw new HttpException(
+        new FederationUnauthorizedError('Federation context missing').toEnvelope(),
+        401,
+      );
+    }
+
+    const requestedLimit = parseLimit(body);
+    const cursor = parseCursor(body);
+    const scopeResult = await this.scope.evaluateAccess({
+      context: request.federationContext,
+      resource,
+      requestedLimit,
+      nativeRbac: this.query,
+    });
+
+    if (!scopeResult.allowed) {
+      const ErrorClass =
+        scopeResult.deny.statusCode === 400
+          ? FederationInvalidRequestError
+          : FederationScopeViolationError;
+      throw new HttpException(
+        new ErrorClass(scopeResult.deny.message, scopeResult.deny).toEnvelope(),
+        scopeResult.deny.statusCode,
+      );
+    }
+
+    let result: Awaited<ReturnType<FederationListQueryService['list']>>;
+    try {
+      result = await this.query.list({ filter: scopeResult.filter, cursor });
+    } catch (error: unknown) {
+      if (error instanceof Error && error.message === 'Invalid federation list cursor') {
+        throw new HttpException(
+          new FederationInvalidRequestError('Federation list cursor is invalid').toEnvelope(),
+          400,
+        );
+      }
+      throw error;
+    }
+
+    const response: FederationListResponse<FederatedRow> = {
+      items: tagWithSource(result.items, SOURCE_LOCAL),
+    };
+    if (result.nextCursor !== undefined) {
+      response.nextCursor = result.nextCursor;
+    }
+    if (result.truncated) {
+      response._truncated = true;
+    }
+    return response;
+  }
+}

apps/gateway/src/gc/session-gc.service.spec.ts

@@ -3,6 +3,7 @@ import { Logger } from '@nestjs/common';
 import type { QueueHandle } from '@mosaicstack/queue';
 import type { LogService } from '@mosaicstack/log';
 import { SessionGCService } from './session-gc.service.js';
+import { CommandAuthorizationService } from '../commands/command-authorization.service.js';
 
 type MockRedis = {
   scan: ReturnType<typeof vi.fn>;
@@ -12,7 +13,12 @@ type MockRedis = {
 describe('SessionGCService', () => {
   let service: SessionGCService;
   let mockRedis: MockRedis;
-  let mockLogService: { logs: { promoteToWarm: ReturnType<typeof vi.fn> } };
+  let mockLogService: {
+    logs: {
+      promoteSessionToWarm: ReturnType<typeof vi.fn>;
+      promoteToWarm: ReturnType<typeof vi.fn>;
+    };
+  };
 
   /**
    * Helper: build a scan mock that returns all provided keys in a single
@@ -30,6 +36,7 @@ describe('SessionGCService', () => {
 
     mockLogService = {
       logs: {
+        promoteSessionToWarm: vi.fn().mockResolvedValue(0),
         promoteToWarm: vi.fn().mockResolvedValue(0),
       },
     };
@@ -59,54 +66,76 @@ describe('SessionGCService', () => {
     expect(result.cleaned.valkeyKeys).toBeUndefined();
   });
 
+  it('escapes glob metacharacters in a session identifier', async () => {
+    await service.collect('abc*?[tenant]\\escape');
+
+    expect(mockRedis.scan).toHaveBeenCalledWith(
+      '0',
+      'MATCH',
+      'mosaic:session:abc\\*\\?\\[tenant\\]\\\\escape:*',
+      'COUNT',
+      100,
+    );
+  });
+
+  it('preserves a valid durable approval after session GC', async () => {
+    const entries = new Map<string, string>();
+    const redis = {
+      scan: vi.fn().mockResolvedValue(['0', ['mosaic:session:owned:state']]),
+      get: vi.fn(async (key: string) => entries.get(key) ?? null),
+      set: vi.fn(async (key: string, value: string) => entries.set(key, value)),
+      del: vi.fn(async (...keys: string[]) => {
+        let deleted = 0;
+        for (const key of keys) deleted += Number(entries.delete(key));
+        return deleted;
+      }),
+    };
+    const authorization = new CommandAuthorizationService(
+      {
+        select: () => ({
+          from: () => ({ where: () => ({ limit: async () => [{ role: 'admin' }] }) }),
+        }),
+      } as never,
+      redis,
+    );
+    const command = {
+      name: 'gc',
+      description: 'System-wide garbage collection',
+      aliases: [],
+      scope: 'admin',
+      execution: 'socket',
+      available: true,
+    } as never;
+    const payload = { command: 'gc', conversationId: 'owned' };
+    const approval = await authorization.createApproval(command, payload, 'admin-1');
+    const approvalKey = `interaction:command-approval:${approval!.approvalId}`;
+    const gc = new SessionGCService(redis as never, mockLogService as unknown as LogService);
+
+    await gc.collect('owned');
+
+    expect(entries.has(approvalKey)).toBe(true);
+    await expect(
+      authorization.authorize(command, payload, 'admin-1', approval!.approvalId),
+    ).resolves.toEqual({ allowed: true });
+  });
+
   it('collect() returns sessionId in result', async () => {
     const result = await service.collect('test-session-id');
     expect(result.sessionId).toBe('test-session-id');
   });
 
-  it('fullCollect() deletes all session keys', async () => {
-    mockRedis.scan = makeScanMock(['mosaic:session:abc:system', 'mosaic:session:xyz:foo']);
-    const result = await service.fullCollect();
-    expect(mockRedis.del).toHaveBeenCalled();
-    expect(result.valkeyKeys).toBe(2);
+  it('collect() demotes logs only for the requested session', async () => {
+    await service.collect('owned-session');
+
+    expect(mockLogService.logs.promoteSessionToWarm).toHaveBeenCalledWith(
+      'owned-session',
+      expect.any(Date),
+    );
+    expect(mockLogService.logs.promoteToWarm).not.toHaveBeenCalled();
   });
 
-  it('fullCollect() with no keys returns 0 valkeyKeys', async () => {
-    mockRedis.scan = makeScanMock([]);
-    const result = await service.fullCollect();
-    expect(result.valkeyKeys).toBe(0);
-    expect(mockRedis.del).not.toHaveBeenCalled();
-  });
-
-  it('fullCollect() returns duration', async () => {
-    const result = await service.fullCollect();
-    expect(result.duration).toBeGreaterThanOrEqual(0);
-  });
-
-  it('sweepOrphans() extracts unique session IDs and collects them', async () => {
-    // First scan call returns the global session list; subsequent calls return
-    // per-session keys during collect().
-    mockRedis.scan = vi
-      .fn()
-      .mockResolvedValueOnce([
-        '0',
-        ['mosaic:session:abc:system', 'mosaic:session:abc:messages', 'mosaic:session:xyz:system'],
-      ])
-      // collect('abc') scan
-      .mockResolvedValueOnce(['0', ['mosaic:session:abc:system', 'mosaic:session:abc:messages']])
-      // collect('xyz') scan
-      .mockResolvedValueOnce(['0', ['mosaic:session:xyz:system']]);
-    mockRedis.del.mockResolvedValue(1);
-
-    const result = await service.sweepOrphans();
-    expect(result.orphanedSessions).toBeGreaterThanOrEqual(0);
-    expect(result.duration).toBeGreaterThanOrEqual(0);
-  });
-
-  it('sweepOrphans() returns empty when no session keys', async () => {
-    mockRedis.scan = makeScanMock([]);
-    const result = await service.sweepOrphans();
-    expect(result.orphanedSessions).toBe(0);
-    expect(result.totalCleaned).toHaveLength(0);
+  it('does not expose automatic global GC entry points', () => {
+    expect('fullCollect' in service).toBe(false);
+    expect('sweepOrphans' in service).toBe(false);
   });
 });

apps/gateway/src/gc/session-gc.service.ts

@@ -1,4 +1,4 @@
-import { Inject, Injectable, Logger, type OnModuleInit } from '@nestjs/common';
+import { Inject, Injectable } from '@nestjs/common';
 import type { QueueHandle } from '@mosaicstack/queue';
 import type { LogService } from '@mosaicstack/log';
 import { LOG_SERVICE } from '../log/log.tokens.js';
@@ -13,49 +13,18 @@ export interface GCResult {
   };
 }
 
-export interface GCSweepResult {
-  orphanedSessions: number;
-  totalCleaned: GCResult[];
-  duration: number;
-}
-
-export interface FullGCResult {
-  valkeyKeys: number;
-  logsDemoted: number;
-  jobsPurged: number;
-  tempFilesRemoved: number;
-  duration: number;
+/** Escape Redis glob metacharacters so a session identifier is always literal. */
+function escapeRedisGlobLiteral(value: string): string {
+  return value.replace(/[\\*?\[\]]/g, '\\$&');
 }
 
 @Injectable()
-export class SessionGCService implements OnModuleInit {
-  private readonly logger = new Logger(SessionGCService.name);
-
+export class SessionGCService {
   constructor(
     @Inject(REDIS) private readonly redis: QueueHandle['redis'],
     @Inject(LOG_SERVICE) private readonly logService: LogService,
   ) {}
 
-  onModuleInit(): void {
-    // Fire-and-forget: run full GC asynchronously so it does not block the
-    // NestJS bootstrap chain.  Cold-start GC typically takes 100–500 ms
-    // depending on Valkey key count; deferring it removes that latency from
-    // the TTFB of the first HTTP request.
-    this.fullCollect()
-      .then((result) => {
-        this.logger.log(
-          `Full GC complete: ${result.valkeyKeys} Valkey keys, ` +
-            `${result.logsDemoted} logs demoted, ` +
-            `${result.jobsPurged} jobs purged, ` +
-            `${result.tempFilesRemoved} temp dirs removed ` +
-            `(${result.duration}ms)`,
-        );
-      })
-      .catch((err: unknown) => {
-        this.logger.error('Cold-start GC failed', err instanceof Error ? err.stack : String(err));
-      });
-  }
-
   /**
    * Scan Valkey for all keys matching a pattern using SCAN (non-blocking).
    * KEYS is avoided because it blocks the Valkey event loop for the full scan
@@ -79,86 +48,20 @@ export class SessionGCService implements OnModuleInit {
     const result: GCResult = { sessionId, cleaned: {} };
 
     // 1. Valkey: delete all session-scoped keys
-    const pattern = `mosaic:session:${sessionId}:*`;
+    const pattern = `mosaic:session:${escapeRedisGlobLiteral(sessionId)}:*`;
     const valkeyKeys = await this.scanKeys(pattern);
     if (valkeyKeys.length > 0) {
       await this.redis.del(...valkeyKeys);
       result.cleaned.valkeyKeys = valkeyKeys.length;
     }
 
-    // 2. PG: demote hot-tier agent_logs for this session to warm
-    const cutoff = new Date(); // demote all hot logs for this session
-    const logsDemoted = await this.logService.logs.promoteToWarm(cutoff);
+    // 2. PG: demote hot-tier agent logs for this session only.
+    const cutoff = new Date();
+    const logsDemoted = await this.logService.logs.promoteSessionToWarm(sessionId, cutoff);
     if (logsDemoted > 0) {
       result.cleaned.logsDemoted = logsDemoted;
     }
 
     return result;
   }
-
-  /**
-   * Sweep GC — find orphaned artifacts from dead sessions.
-   * System-wide operation: only call from admin-authorized paths or internal
-   * scheduled jobs.  Individual session cleanup is handled by collect().
-   */
-  async sweepOrphans(): Promise<GCSweepResult> {
-    const start = Date.now();
-    const cleaned: GCResult[] = [];
-
-    // 1. Find all session-scoped Valkey keys (non-blocking SCAN)
-    const allSessionKeys = await this.scanKeys('mosaic:session:*');
-
-    // Extract unique session IDs from keys
-    const sessionIds = new Set<string>();
-    for (const key of allSessionKeys) {
-      const match = key.match(/^mosaic:session:([^:]+):/);
-      if (match) sessionIds.add(match[1]!);
-    }
-
-    // 2. For each session ID, collect stale keys
-    for (const sessionId of sessionIds) {
-      const gcResult = await this.collect(sessionId);
-      if (Object.keys(gcResult.cleaned).length > 0) {
-        cleaned.push(gcResult);
-      }
-    }
-
-    return {
-      orphanedSessions: cleaned.length,
-      totalCleaned: cleaned,
-      duration: Date.now() - start,
-    };
-  }
-
-  /**
-   * Full GC — aggressive collection for cold start.
-   * Assumes no sessions survived the restart.
-   */
-  async fullCollect(): Promise<FullGCResult> {
-    const start = Date.now();
-
-    // 1. Valkey: delete ALL session-scoped keys (non-blocking SCAN)
-    const sessionKeys = await this.scanKeys('mosaic:session:*');
-    if (sessionKeys.length > 0) {
-      await this.redis.del(...sessionKeys);
-    }
-
-    // 2. NOTE: channel keys are NOT collected on cold start
-    //    (discord/telegram plugins may reconnect and resume)
-
-    // 3. PG: demote stale hot-tier logs older than 24h to warm
-    const hotCutoff = new Date(Date.now() - 24 * 60 * 60 * 1000);
-    const logsDemoted = await this.logService.logs.promoteToWarm(hotCutoff);
-
-    // 4. No summarization job purge API available yet
-    const jobsPurged = 0;
-
-    return {
-      valkeyKeys: sessionKeys.length,
-      logsDemoted,
-      jobsPurged,
-      tempFilesRemoved: 0,
-      duration: Date.now() - start,
-    };
-  }
 }

apps/gateway/src/health/health.controller.test.ts

@@ -0,0 +1,12 @@
+import { describe, expect, it } from 'vitest';
+import { HealthController } from './health.controller.js';
+
+describe('HealthController', (): void => {
+  it('exposes liveness and readiness without configuration details', (): void => {
+    const controller = new HealthController();
+
+    expect(controller.check()).toEqual({ status: 'ok' });
+    expect(controller.ready()).toEqual({ status: 'ready' });
+    expect(JSON.stringify(controller.ready())).not.toContain('credential');
+  });
+});

apps/gateway/src/health/health.controller.ts

@@ -6,4 +6,10 @@ export class HealthController {
   check(): { status: string } {
     return { status: 'ok' };
   }
+
+  /** Readiness intentionally exposes no configuration, provider, or credential details. */
+  @Get('ready')
+  ready(): { status: string } {
+    return { status: 'ready' };
+  }
 }

apps/gateway/src/log/cron.service.ts

@@ -6,11 +6,10 @@ import {
   type OnModuleDestroy,
 } from '@nestjs/common';
 import { SummarizationService } from './summarization.service.js';
-import { SessionGCService } from '../gc/session-gc.service.js';
 import {
   QueueService,
-  QUEUE_SUMMARIZATION,
   QUEUE_GC,
+  QUEUE_SUMMARIZATION,
   QUEUE_TIER_MANAGEMENT,
 } from '../queue/queue.service.js';
 import type { Worker } from 'bullmq';
@@ -23,14 +22,12 @@ export class CronService implements OnModuleInit, OnModuleDestroy {
 
   constructor(
     @Inject(SummarizationService) private readonly summarization: SummarizationService,
-    @Inject(SessionGCService) private readonly sessionGC: SessionGCService,
     @Inject(QueueService) private readonly queueService: QueueService,
   ) {}
 
   async onModuleInit(): Promise<void> {
     const summarizationSchedule = process.env['SUMMARIZATION_CRON'] ?? '0 */6 * * *'; // every 6 hours
     const tierManagementSchedule = process.env['TIER_MANAGEMENT_CRON'] ?? '0 3 * * *'; // daily at 3am
-    const gcSchedule = process.env['SESSION_GC_CRON'] ?? '0 4 * * *'; // daily at 4am
 
     // M6-003: Summarization repeatable job
     await this.queueService.addRepeatableJob(
@@ -56,15 +53,12 @@ export class CronService implements OnModuleInit, OnModuleDestroy {
     });
     this.registeredWorkers.push(tierWorker);
 
-    // M6-004: GC repeatable job
-    await this.queueService.addRepeatableJob(QUEUE_GC, 'session-gc', {}, gcSchedule);
-    const gcWorker = this.queueService.registerWorker(QUEUE_GC, async () => {
-      await this.sessionGC.sweepOrphans();
-    });
-    this.registeredWorkers.push(gcWorker);
+    // Retire any repeatable global GC schedule created by older deployments.
+    // Session cleanup is now triggered only by an authorized session lifecycle operation.
+    await this.queueService.removeRepeatableJobs(QUEUE_GC, 'session-gc');
 
     this.logger.log(
-      `BullMQ jobs scheduled: summarization="${summarizationSchedule}", tier="${tierManagementSchedule}", gc="${gcSchedule}"`,
+      `BullMQ jobs scheduled: summarization="${summarizationSchedule}", tier="${tierManagementSchedule}"`,
     );
   }
 

apps/gateway/src/mcp/mcp.controller.ts

@@ -3,7 +3,11 @@ import { Logger } from '@nestjs/common';
 import { fromNodeHeaders } from 'better-auth/node';
 import type { Auth } from '@mosaicstack/auth';
 import type { NestFastifyApplication } from '@nestjs/platform-fastify';
-import type { McpService } from './mcp.service.js';
+import {
+  createMcpActorContext,
+  deriveMcpToolScopesForUser,
+  type McpService,
+} from './mcp.service.js';
 import { AUTH } from '../auth/auth.tokens.js';
 
 /**
@@ -67,14 +71,25 @@ async function handleMcpRequest(
     return;
   }
 
-  const userId = result.user.id;
+  const authUser = result.user as {
+    id: string;
+    role?: string | null;
+    tenantId?: string | null;
+    organizationId?: string | null;
+  };
+  const actor = createMcpActorContext({
+    userId: authUser.id,
+    role: authUser.role,
+    tenantId: authUser.tenantId ?? authUser.organizationId ?? undefined,
+    scopes: deriveMcpToolScopesForUser({ role: authUser.role }),
+  });
 
   // ─── Session routing ─────────────────────────────────────────────────────
   const sessionId = req.raw.headers['mcp-session-id'];
 
   if (typeof sessionId === 'string' && sessionId.length > 0) {
     // Existing session request
-    const transport = mcpService.getSession(sessionId);
+    const transport = mcpService.getSession(sessionId, actor);
     if (!transport) {
       logger.warn(`MCP session not found: ${sessionId}`);
       reply.raw.writeHead(404, { 'Content-Type': 'application/json' });
@@ -112,8 +127,10 @@ async function handleMcpRequest(
   }
 
   // Create new session and handle this initializing request
-  const { transport } = mcpService.createSession(userId);
-  logger.log(`New MCP session created for user ${userId}`);
+  const { transport } = mcpService.createSession(actor);
+  logger.log(
+    `New MCP session created for actor=${actor.userId} tenant=${actor.tenantId} correlation=${actor.correlationId}`,
+  );
 
   await transport.handleRequest(req.raw, reply.raw, body);
 }

apps/gateway/src/mcp/mcp.service.spec.ts

@@ -0,0 +1,461 @@
+import { describe, expect, it, vi } from 'vitest';
+import type { z } from 'zod';
+import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+import type { Brain } from '@mosaicstack/brain';
+import type { Memory } from '@mosaicstack/memory';
+import type { EmbeddingService } from '../memory/embedding.service.js';
+import type { CoordService } from '../coord/coord.service.js';
+import {
+  assertMcpToolAuthorized,
+  createMcpActorContext,
+  deriveMcpToolScopesForUser,
+  MCP_TOOL_SCOPES,
+  McpService,
+  type McpToolName,
+} from './mcp.service.js';
+
+type ToolResult = { content: Array<{ type: 'text'; text: string }> };
+type ToolHandler = (params: Record<string, unknown>) => Promise<ToolResult>;
+
+interface CapturedTool {
+  inputSchema: z.ZodType;
+  handler: ToolHandler;
+}
+
+function makeCapturingServer(): { server: McpServer; tools: Map<string, CapturedTool> } {
+  const tools = new Map<string, CapturedTool>();
+  const server = {
+    registerTool(name: string, config: { inputSchema: z.ZodType }, handler: ToolHandler): void {
+      tools.set(name, { inputSchema: config.inputSchema, handler });
+    },
+  };
+  return { server: server as unknown as McpServer, tools };
+}
+
+function makeService(opts?: {
+  projects?: Array<Record<string, unknown> & { id: string; ownerId?: string | null }>;
+  missions?: Array<
+    Record<string, unknown> & { id: string; projectId?: string | null; userId?: string | null }
+  >;
+  tasks?: Array<
+    Record<string, unknown> & {
+      id: string;
+      projectId?: string | null;
+      missionId?: string | null;
+      status?: string;
+    }
+  >;
+}) {
+  const projects = opts?.projects ?? [];
+  const missions = opts?.missions ?? [];
+  const tasks = opts?.tasks ?? [];
+  const brain = {
+    projects: {
+      findAll: vi.fn(async () => projects),
+      findById: vi.fn(async (id: string) => projects.find((project) => project.id === id) ?? null),
+    },
+    tasks: {
+      findAll: vi.fn(async () => tasks),
+      findById: vi.fn(async (id: string) => tasks.find((task) => task.id === id) ?? null),
+      findByProject: vi.fn(async (projectId: string) =>
+        tasks.filter((task) => task.projectId === projectId),
+      ),
+      findByMission: vi.fn(async (missionId: string) =>
+        tasks.filter((task) => task.missionId === missionId),
+      ),
+      findByStatus: vi.fn(async (status: string) => tasks.filter((task) => task.status === status)),
+      create: vi.fn(async (task: Record<string, unknown>) => ({ id: 'task-1', ...task })),
+      update: vi.fn(async (id: string, updates: Record<string, unknown>) => ({ id, ...updates })),
+    },
+    missions: {
+      findAll: vi.fn(async () => missions),
+      findById: vi.fn(async (id: string) => missions.find((mission) => mission.id === id) ?? null),
+      findByProject: vi.fn(async (projectId: string) =>
+        missions.filter((mission) => mission.projectId === projectId),
+      ),
+    },
+    conversations: {
+      findAll: vi.fn(async (userId: string) => [{ id: 'conversation-1', userId }]),
+    },
+  } as unknown as Brain;
+
+  const memory = {
+    insights: {
+      searchByEmbedding: vi.fn(async (userId: string) => [{ id: 'insight-1', userId }]),
+      create: vi.fn(async (insight: Record<string, unknown>) => ({ id: 'insight-2', ...insight })),
+    },
+    preferences: {
+      findByUser: vi.fn(async (userId: string) => [{ key: 'theme', userId }]),
+      findByUserAndCategory: vi.fn(async (userId: string, category: string) => [
+        { key: 'theme', userId, category },
+      ]),
+      upsert: vi.fn(async (preference: Record<string, unknown>) => ({
+        id: 'pref-1',
+        ...preference,
+      })),
+    },
+  } as unknown as Memory;
+
+  const embeddings = {
+    available: true,
+    embed: vi.fn(async () => [0.1, 0.2, 0.3]),
+  } as unknown as EmbeddingService;
+
+  const coord = {
+    getMissionStatus: vi.fn(async () => null),
+    listTasks: vi.fn(async () => []),
+    getTaskStatus: vi.fn(async () => null),
+  } as unknown as CoordService;
+
+  return {
+    service: new McpService(brain, memory, embeddings, coord),
+    brain: brain as unknown as {
+      conversations: { findAll: ReturnType<typeof vi.fn> };
+      tasks: {
+        create: ReturnType<typeof vi.fn>;
+        update: ReturnType<typeof vi.fn>;
+      };
+    },
+    memory: memory as unknown as {
+      insights: { searchByEmbedding: ReturnType<typeof vi.fn> };
+    },
+    coord: coord as unknown as {
+      listTasks: ReturnType<typeof vi.fn>;
+    },
+  };
+}
+
+function getTool(tools: Map<string, CapturedTool>, name: McpToolName): CapturedTool {
+  const tool = tools.get(name);
+  if (!tool) throw new Error(`Missing captured tool ${name}`);
+  return tool;
+}
+
+function makeMemberActor(userId = 'authenticated-user') {
+  return createMcpActorContext({
+    userId,
+    role: 'member',
+    scopes: deriveMcpToolScopesForUser({ role: 'member' }),
+  });
+}
+
+function makeAdminActor(userId = 'admin-user', tenantId?: string) {
+  return createMcpActorContext({
+    userId,
+    tenantId,
+    role: 'admin',
+    scopes: deriveMcpToolScopesForUser({ role: 'admin' }),
+  });
+}
+
+function makePlatformAdminActor(userId = 'platform-admin-user') {
+  return createMcpActorContext({
+    userId,
+    role: 'platform-admin',
+    scopes: deriveMcpToolScopesForUser({ role: 'platform-admin' }),
+  });
+}
+
+describe('MCP actor identity and tool scope enforcement', () => {
+  it('derives immutable actor, tenant, channel, correlation, and explicit tool scopes server-side', () => {
+    const actor = createMcpActorContext({
+      userId: ' user-authenticated ',
+      role: 'member',
+      scopes: deriveMcpToolScopesForUser({ role: 'member' }),
+    });
+
+    expect(actor.userId).toBe('user-authenticated');
+    expect(actor.tenantId).toBe('user:user-authenticated');
+    expect(actor.role).toBe('member');
+    expect(actor.channel).toBe('mcp');
+    expect(actor.correlationId).toMatch(/[0-9a-f-]{36}/i);
+    expect(actor.scopes.has(MCP_TOOL_SCOPES.memory_search)).toBe(true);
+    expect(actor.scopes.has(MCP_TOOL_SCOPES.coord_list_tasks)).toBe(false);
+    expect(
+      deriveMcpToolScopesForUser({ role: 'admin' }).has(MCP_TOOL_SCOPES.coord_list_tasks),
+    ).toBe(false);
+    expect(
+      deriveMcpToolScopesForUser({ role: 'platform-admin' }).has(MCP_TOOL_SCOPES.coord_list_tasks),
+    ).toBe(true);
+  });
+
+  it('fails closed when scopes are not supplied by the authenticated context policy', () => {
+    const actor = createMcpActorContext({ userId: 'user-authenticated' });
+
+    expect(actor.scopes.size).toBe(0);
+    expect(() => assertMcpToolAuthorized(actor, 'memory_search', { query: 'notes' })).toThrow(
+      'MCP tool scope denied: memory:insight:read',
+    );
+  });
+
+  it('fails closed when a tool caller supplies actor or tenant identity fields', () => {
+    const actor = createMcpActorContext({ userId: 'user-authenticated' });
+
+    expect(() =>
+      assertMcpToolAuthorized(actor, 'memory_search', {
+        userId: 'victim-user',
+        query: 'private data',
+      }),
+    ).toThrow('MCP caller-controlled identity field is forbidden: userId');
+
+    expect(() =>
+      assertMcpToolAuthorized(actor, 'coord_list_tasks', {
+        tenantId: 'victim-tenant',
+        projectPath: '/tmp/project',
+      }),
+    ).toThrow('MCP caller-controlled identity field is forbidden: tenantId');
+
+    expect(() =>
+      assertMcpToolAuthorized(actor, 'brain_create_task', {
+        title: 'forged org',
+        organizationId: 'victim-org',
+      }),
+    ).toThrow('MCP caller-controlled identity field is forbidden: organizationId');
+
+    expect(() =>
+      assertMcpToolAuthorized(actor, 'brain_update_task', {
+        title: 'forged team',
+        teamId: 'victim-team',
+      }),
+    ).toThrow('MCP caller-controlled identity field is forbidden: teamId');
+  });
+
+  it('fails closed when the server-derived actor lacks the required per-tool scope', () => {
+    const actor = createMcpActorContext({ userId: 'user-authenticated', scopes: [] });
+
+    expect(() => assertMcpToolAuthorized(actor, 'memory_search', { query: 'notes' })).toThrow(
+      'MCP tool scope denied: memory:insight:read',
+    );
+  });
+
+  it('removes caller-controlled userId from memory schemas and never queries victim memory', async () => {
+    const { service, memory } = makeService();
+    const { server, tools } = makeCapturingServer();
+    const actor = makeMemberActor('authenticated-user');
+
+    service.registerTools(server, actor);
+    const tool = getTool(tools, 'memory_search');
+
+    expect(tool.inputSchema.safeParse({ userId: 'victim-user', query: 'anything' }).success).toBe(
+      false,
+    );
+    await expect(tool.handler({ userId: 'victim-user', query: 'anything' })).rejects.toThrow(
+      'MCP caller-controlled identity field is forbidden: userId',
+    );
+    expect(memory.insights.searchByEmbedding).not.toHaveBeenCalled();
+
+    await tool.handler({ query: 'only my notes' });
+    expect(memory.insights.searchByEmbedding).toHaveBeenCalledWith(
+      'authenticated-user',
+      [0.1, 0.2, 0.3],
+      5,
+    );
+  });
+
+  it('binds conversation listing to the authenticated actor instead of a caller-supplied userId', async () => {
+    const { service, brain } = makeService();
+    const { server, tools } = makeCapturingServer();
+    const actor = makeMemberActor('authenticated-user');
+
+    service.registerTools(server, actor);
+    const tool = getTool(tools, 'brain_list_conversations');
+
+    expect(tool.inputSchema.safeParse({ userId: 'victim-user' }).success).toBe(false);
+    await expect(tool.handler({ userId: 'victim-user' })).rejects.toThrow(
+      'MCP caller-controlled identity field is forbidden: userId',
+    );
+    expect(brain.conversations.findAll).not.toHaveBeenCalled();
+
+    await tool.handler({});
+    expect(brain.conversations.findAll).toHaveBeenCalledWith('authenticated-user');
+  });
+
+  it('scopes brain project, mission, and task reads to the authenticated actor', async () => {
+    const { service } = makeService({
+      projects: [
+        { id: 'project-owned', ownerId: 'authenticated-user', name: 'owned' },
+        { id: 'project-victim', ownerId: 'victim-user', name: 'victim' },
+      ],
+      missions: [
+        { id: 'mission-owned', projectId: 'project-owned' },
+        { id: 'mission-victim', userId: 'victim-user', projectId: 'project-victim' },
+      ],
+      tasks: [
+        { id: 'task-owned-project', projectId: 'project-owned', status: 'not-started' },
+        { id: 'task-owned-mission', missionId: 'mission-owned', status: 'not-started' },
+        { id: 'task-victim-project', projectId: 'project-victim', status: 'not-started' },
+        { id: 'task-unowned', status: 'not-started' },
+      ],
+    });
+    const { server, tools } = makeCapturingServer();
+    const actor = makeMemberActor('authenticated-user');
+
+    service.registerTools(server, actor);
+
+    const projects = JSON.parse(
+      (await getTool(tools, 'brain_list_projects').handler({})).content[0]!.text,
+    );
+    expect(projects.map((project: { id: string }) => project.id)).toEqual(['project-owned']);
+
+    const missions = JSON.parse(
+      (await getTool(tools, 'brain_list_missions').handler({})).content[0]!.text,
+    );
+    expect(missions.map((mission: { id: string }) => mission.id)).toEqual(['mission-owned']);
+
+    const tasks = JSON.parse(
+      (await getTool(tools, 'brain_list_tasks').handler({})).content[0]!.text,
+    );
+    expect(tasks.map((task: { id: string }) => task.id)).toEqual([
+      'task-owned-project',
+      'task-owned-mission',
+    ]);
+  });
+
+  it('enforces tenant boundaries for tenant-admin brain project, mission, and task reads', async () => {
+    const { service } = makeService({
+      projects: [
+        {
+          id: 'project-tenant-a',
+          ownerId: 'other-user-a',
+          teamId: 'tenant-a',
+          name: 'same tenant',
+        },
+        {
+          id: 'project-tenant-b',
+          ownerId: 'other-user-b',
+          teamId: 'tenant-b',
+          name: 'other tenant',
+        },
+      ],
+      missions: [
+        { id: 'mission-tenant-a', tenantId: 'tenant-a', projectId: 'project-tenant-a' },
+        { id: 'mission-tenant-b', tenantId: 'tenant-b', projectId: 'project-tenant-b' },
+      ],
+      tasks: [
+        { id: 'task-tenant-a', projectId: 'project-tenant-a', status: 'not-started' },
+        { id: 'task-tenant-b', projectId: 'project-tenant-b', status: 'not-started' },
+      ],
+    });
+    const { server, tools } = makeCapturingServer();
+    const actor = makeAdminActor('tenant-admin-user', 'tenant-a');
+
+    service.registerTools(server, actor);
+
+    const projects = JSON.parse(
+      (await getTool(tools, 'brain_list_projects').handler({})).content[0]!.text,
+    );
+    expect(projects.map((project: { id: string }) => project.id)).toEqual(['project-tenant-a']);
+
+    const missions = JSON.parse(
+      (await getTool(tools, 'brain_list_missions').handler({})).content[0]!.text,
+    );
+    expect(missions.map((mission: { id: string }) => mission.id)).toEqual(['mission-tenant-a']);
+
+    const tasks = JSON.parse(
+      (await getTool(tools, 'brain_list_tasks').handler({})).content[0]!.text,
+    );
+    expect(tasks.map((task: { id: string }) => task.id)).toEqual(['task-tenant-a']);
+  });
+
+  it('denies tenant-admin task writes outside the authenticated tenant', async () => {
+    const { service, brain } = makeService({
+      projects: [
+        { id: 'project-tenant-a', ownerId: 'other-user-a', teamId: 'tenant-a' },
+        { id: 'project-tenant-b', ownerId: 'other-user-b', teamId: 'tenant-b' },
+      ],
+      missions: [
+        { id: 'mission-tenant-a', tenantId: 'tenant-a', projectId: 'project-tenant-a' },
+        { id: 'mission-tenant-b', tenantId: 'tenant-b', projectId: 'project-tenant-b' },
+      ],
+      tasks: [
+        { id: 'task-tenant-a', projectId: 'project-tenant-a', status: 'not-started' },
+        { id: 'task-tenant-b', projectId: 'project-tenant-b', status: 'not-started' },
+      ],
+    });
+    const { server, tools } = makeCapturingServer();
+    const actor = makeAdminActor('tenant-admin-user', 'tenant-a');
+
+    service.registerTools(server, actor);
+
+    await expect(
+      getTool(tools, 'brain_create_task').handler({
+        title: 'unscoped tenant write',
+      }),
+    ).rejects.toThrow('MCP task scope denied');
+    expect(brain.tasks.create).not.toHaveBeenCalled();
+
+    await expect(
+      getTool(tools, 'brain_create_task').handler({
+        title: 'cross-tenant write',
+        projectId: 'project-tenant-b',
+      }),
+    ).rejects.toThrow('MCP task project scope denied');
+    expect(brain.tasks.create).not.toHaveBeenCalled();
+
+    await expect(
+      getTool(tools, 'brain_update_task').handler({
+        id: 'task-tenant-a',
+        projectId: 'project-tenant-b',
+      }),
+    ).rejects.toThrow('MCP task project scope denied');
+    expect(brain.tasks.update).not.toHaveBeenCalled();
+
+    const updateResult = await getTool(tools, 'brain_update_task').handler({
+      id: 'task-tenant-b',
+      title: 'cross-tenant update',
+    });
+    expect(updateResult.content[0]!.text).toBe('Task not found: task-tenant-b');
+    expect(brain.tasks.update).not.toHaveBeenCalled();
+
+    await getTool(tools, 'brain_create_task').handler({
+      title: 'same-tenant write',
+      projectId: 'project-tenant-a',
+    });
+    expect(brain.tasks.create).toHaveBeenCalledWith(
+      expect.objectContaining({ projectId: 'project-tenant-a', title: 'same-tenant write' }),
+    );
+  });
+
+  it('keeps admin-only coordination tools on server-derived paths', async () => {
+    const { service, coord } = makeService();
+    const { server, tools } = makeCapturingServer();
+    const member = makeMemberActor('authenticated-user');
+    const tenantAdmin = makeAdminActor('admin-user');
+    const platformAdmin = makePlatformAdminActor('platform-admin-user');
+
+    service.registerTools(server, member);
+    const memberTool = getTool(tools, 'coord_list_tasks');
+    expect(memberTool.inputSchema.safeParse({ projectPath: '/tmp/victim' }).success).toBe(false);
+    await expect(memberTool.handler({})).rejects.toThrow('MCP tool scope denied: coord:read');
+
+    tools.clear();
+    service.registerTools(server, tenantAdmin);
+    const tenantAdminTool = getTool(tools, 'coord_list_tasks');
+    await expect(tenantAdminTool.handler({})).rejects.toThrow('MCP tool scope denied: coord:read');
+
+    tools.clear();
+    service.registerTools(server, platformAdmin);
+    const platformAdminTool = getTool(tools, 'coord_list_tasks');
+    await platformAdminTool.handler({ projectPath: '/tmp/victim' });
+    expect(coord.listTasks).toHaveBeenCalledWith(process.cwd());
+  });
+
+  it('does not attach a guessed or stale-scope MCP session to another authenticated context', async () => {
+    const { service } = makeService();
+    const owner = makeMemberActor('owner-user');
+    const attacker = makeMemberActor('attacker-user');
+    const admin = makeAdminActor('admin-user');
+    const downgradedAdmin = makeMemberActor('admin-user');
+
+    const { sessionId, transport } = service.createSession(owner);
+    const staleSession = service.createSession(admin);
+
+    expect(service.getSession(sessionId, owner)).toBe(transport);
+    expect(service.getSession(sessionId, attacker)).toBeNull();
+    expect(service.getSession(sessionId, owner)).toBe(transport);
+    expect(service.getSession(staleSession.sessionId, downgradedAdmin)).toBeNull();
+    expect(service.getSession(staleSession.sessionId, admin)).toBe(staleSession.transport);
+
+    await service.onModuleDestroy();
+  });
+});

apps/gateway/src/mcp/mcp.service.ts

@@ -10,11 +10,216 @@ import { MEMORY } from '../memory/memory.tokens.js';
 import { EmbeddingService } from '../memory/embedding.service.js';
 import { CoordService } from '../coord/coord.service.js';
 
+export const MCP_CALLER_IDENTITY_FIELDS = [
+  'actorId',
+  'actor',
+  'authenticatedUserId',
+  'channel',
+  'organizationId',
+  'ownerId',
+  'sessionUserId',
+  'teamId',
+  'tenant',
+  'tenantId',
+  'user',
+  'userId',
+] as const;
+
+type McpCallerIdentityField = (typeof MCP_CALLER_IDENTITY_FIELDS)[number];
+
+export const MCP_TOOL_SCOPES = {
+  brain_list_projects: 'brain:project:read',
+  brain_get_project: 'brain:project:read',
+  brain_list_tasks: 'brain:task:read',
+  brain_create_task: 'brain:task:write',
+  brain_update_task: 'brain:task:write',
+  brain_list_missions: 'brain:mission:read',
+  brain_list_conversations: 'brain:conversation:read',
+  memory_search: 'memory:insight:read',
+  memory_get_preferences: 'memory:preference:read',
+  memory_save_preference: 'memory:preference:write',
+  memory_save_insight: 'memory:insight:write',
+  coord_mission_status: 'coord:read',
+  coord_list_tasks: 'coord:read',
+  coord_task_detail: 'coord:read',
+} as const;
+
+export type McpToolName = keyof typeof MCP_TOOL_SCOPES;
+export type McpToolScope = (typeof MCP_TOOL_SCOPES)[McpToolName];
+
+export interface McpActorContext {
+  userId: string;
+  tenantId: string;
+  role: string;
+  channel: 'mcp';
+  correlationId: string;
+  scopes: ReadonlySet<McpToolScope>;
+}
+
 interface SessionEntry {
   server: McpServer;
   transport: StreamableHTTPServerTransport;
   createdAt: Date;
+  actor: McpActorContext;
+}
+
+const GLOBAL_ADMIN_MCP_SCOPES = new Set<McpToolScope>(Object.values(MCP_TOOL_SCOPES));
+const TENANT_ADMIN_MCP_SCOPES = new Set<McpToolScope>([
+  MCP_TOOL_SCOPES.brain_list_projects,
+  MCP_TOOL_SCOPES.brain_get_project,
+  MCP_TOOL_SCOPES.brain_list_tasks,
+  MCP_TOOL_SCOPES.brain_create_task,
+  MCP_TOOL_SCOPES.brain_update_task,
+  MCP_TOOL_SCOPES.brain_list_missions,
+  MCP_TOOL_SCOPES.brain_list_conversations,
+  MCP_TOOL_SCOPES.memory_search,
+  MCP_TOOL_SCOPES.memory_get_preferences,
+  MCP_TOOL_SCOPES.memory_save_preference,
+  MCP_TOOL_SCOPES.memory_save_insight,
+]);
+const MEMBER_MCP_SCOPES = new Set<McpToolScope>([
+  MCP_TOOL_SCOPES.brain_list_projects,
+  MCP_TOOL_SCOPES.brain_get_project,
+  MCP_TOOL_SCOPES.brain_list_tasks,
+  MCP_TOOL_SCOPES.brain_list_missions,
+  MCP_TOOL_SCOPES.brain_list_conversations,
+  MCP_TOOL_SCOPES.memory_search,
+  MCP_TOOL_SCOPES.memory_get_preferences,
+  MCP_TOOL_SCOPES.memory_save_preference,
+  MCP_TOOL_SCOPES.memory_save_insight,
+]);
+
+export function deriveMcpToolScopesForUser(input: {
+  role?: string | null;
+}): ReadonlySet<McpToolScope> {
+  if (input.role === 'platform-admin' || input.role === 'super-admin') {
+    return new Set(GLOBAL_ADMIN_MCP_SCOPES);
+  }
+  if (input.role === 'admin') {
+    return new Set(TENANT_ADMIN_MCP_SCOPES);
+  }
+  return new Set(MEMBER_MCP_SCOPES);
+}
+
+export function createMcpActorContext(input: {
   userId: string;
+  tenantId?: string;
+  role?: string | null;
+  scopes?: Iterable<McpToolScope>;
+  correlationId?: string;
+}): McpActorContext {
+  const userId = input.userId.trim();
+  if (userId.length === 0) {
+    throw new Error('MCP authenticated user is required');
+  }
+
+  return {
+    userId,
+    tenantId: input.tenantId?.trim() || `user:${userId}`,
+    role: input.role ?? 'member',
+    channel: 'mcp',
+    correlationId: input.correlationId ?? randomUUID(),
+    scopes: new Set(input.scopes ?? []),
+  };
+}
+
+export function assertNoCallerControlledIdentity(params: unknown): void {
+  if (params === null || typeof params !== 'object') return;
+
+  const keys = new Set(Object.keys(params));
+  const forbidden = MCP_CALLER_IDENTITY_FIELDS.find((field: McpCallerIdentityField) =>
+    keys.has(field),
+  );
+  if (forbidden) {
+    throw new Error(`MCP caller-controlled identity field is forbidden: ${forbidden}`);
+  }
+}
+
+export function assertMcpToolAuthorized(
+  actor: McpActorContext,
+  toolName: McpToolName,
+  params: unknown,
+): void {
+  assertNoCallerControlledIdentity(params);
+  const requiredScope = MCP_TOOL_SCOPES[toolName];
+  if (!actor.scopes.has(requiredScope)) {
+    throw new Error(`MCP tool scope denied: ${requiredScope}`);
+  }
+}
+
+function strictObject<T extends z.ZodRawShape>(shape: T): z.ZodObject<T> {
+  return z.object(shape).strict();
+}
+
+type TenantScopedLike = {
+  tenantId?: string | null;
+  organizationId?: string | null;
+  teamId?: string | null;
+};
+type ProjectLike = TenantScopedLike & { id: string; ownerId?: string | null };
+type MissionLike = TenantScopedLike & {
+  id: string;
+  projectId?: string | null;
+  userId?: string | null;
+};
+type TaskLike = TenantScopedLike & {
+  projectId?: string | null;
+  missionId?: string | null;
+  userId?: string | null;
+};
+
+function isGlobalAdminActor(actor: McpActorContext): boolean {
+  return actor.role === 'platform-admin' || actor.role === 'super-admin';
+}
+
+function isTenantAdminActor(actor: McpActorContext): boolean {
+  return actor.role === 'admin';
+}
+
+function matchesTenant(actor: McpActorContext, record: TenantScopedLike): boolean {
+  return (
+    record.tenantId === actor.tenantId ||
+    record.organizationId === actor.tenantId ||
+    record.teamId === actor.tenantId
+  );
+}
+
+function filterProjectsForActor<T extends ProjectLike>(actor: McpActorContext, projects: T[]): T[] {
+  if (isGlobalAdminActor(actor)) return projects;
+  return projects.filter(
+    (project) =>
+      project.ownerId === actor.userId ||
+      (isTenantAdminActor(actor) && matchesTenant(actor, project)),
+  );
+}
+
+function filterMissionsByDirectActorScope<T extends MissionLike>(
+  actor: McpActorContext,
+  missions: T[],
+): T[] {
+  if (isGlobalAdminActor(actor)) return missions;
+  return missions.filter(
+    (mission) =>
+      mission.userId === actor.userId ||
+      (isTenantAdminActor(actor) && matchesTenant(actor, mission)),
+  );
+}
+
+function scopesEqual(left: ReadonlySet<McpToolScope>, right: ReadonlySet<McpToolScope>): boolean {
+  if (left.size !== right.size) return false;
+  for (const scope of left) {
+    if (!right.has(scope)) return false;
+  }
+  return true;
+}
+
+function sameActorAuthorization(stored: McpActorContext, current: McpActorContext): boolean {
+  return (
+    stored.userId === current.userId &&
+    stored.tenantId === current.tenantId &&
+    stored.role === current.role &&
+    scopesEqual(stored.scopes, current.scopes)
+  );
 }
 
 @Injectable()
@@ -33,13 +238,18 @@ export class McpService implements OnModuleDestroy {
    * Creates a new MCP session with its own server + transport pair.
    * Returns the transport for use by the controller.
    */
-  createSession(userId: string): { sessionId: string; transport: StreamableHTTPServerTransport } {
+  createSession(actor: McpActorContext): {
+    sessionId: string;
+    transport: StreamableHTTPServerTransport;
+  } {
     const sessionId = randomUUID();
 
     const transport = new StreamableHTTPServerTransport({
       sessionIdGenerator: () => sessionId,
       onsessioninitialized: (id) => {
-        this.logger.log(`MCP session initialized: ${id} for user ${userId}`);
+        this.logger.log(
+          `MCP session initialized: ${id} for actor=${actor.userId} tenant=${actor.tenantId} correlation=${actor.correlationId}`,
+        );
       },
     });
 
@@ -48,7 +258,7 @@ export class McpService implements OnModuleDestroy {
       { capabilities: { tools: {} } },
     );
 
-    this.registerTools(server, userId);
+    this.registerTools(server, actor);
 
     transport.onclose = () => {
       this.logger.log(`MCP session closed: ${sessionId}`);
@@ -61,31 +271,126 @@ export class McpService implements OnModuleDestroy {
       );
     });
 
-    this.sessions.set(sessionId, { server, transport, createdAt: new Date(), userId });
+    this.sessions.set(sessionId, { server, transport, createdAt: new Date(), actor });
     return { sessionId, transport };
   }
 
   /**
-   * Returns the transport for an existing session, or null if not found.
+   * Returns the transport for an existing session only when it belongs to the
+   * currently authenticated MCP actor. Guessed or cross-tenant session IDs grant
+   * no authority.
    */
-  getSession(sessionId: string): StreamableHTTPServerTransport | null {
-    return this.sessions.get(sessionId)?.transport ?? null;
+  getSession(sessionId: string, actor: McpActorContext): StreamableHTTPServerTransport | null {
+    const entry = this.sessions.get(sessionId);
+    if (!entry) return null;
+    if (!sameActorAuthorization(entry.actor, actor)) {
+      this.logger.warn(
+        `MCP session actor or scope mismatch: session=${sessionId} actor=${actor.userId} tenant=${actor.tenantId} role=${actor.role}`,
+      );
+      return null;
+    }
+    return entry.transport;
+  }
+
+  private async isProjectAuthorized(actor: McpActorContext, projectId: string): Promise<boolean> {
+    if (isGlobalAdminActor(actor)) return true;
+    const project = (await this.brain.projects.findById(projectId)) as ProjectLike | undefined;
+    return project ? filterProjectsForActor(actor, [project]).length === 1 : false;
+  }
+
+  private async filterMissionsForActor<T extends MissionLike>(
+    actor: McpActorContext,
+    missions: T[],
+  ): Promise<T[]> {
+    if (isGlobalAdminActor(actor)) return missions;
+
+    const projects = (await this.brain.projects.findAll()) as ProjectLike[];
+    const projectIds = new Set(
+      filterProjectsForActor(actor, projects).map((project) => project.id),
+    );
+
+    return missions.filter(
+      (mission) =>
+        filterMissionsByDirectActorScope(actor, [mission]).length === 1 ||
+        (typeof mission.projectId === 'string' && projectIds.has(mission.projectId)),
+    );
+  }
+
+  private async isMissionAuthorized(actor: McpActorContext, missionId: string): Promise<boolean> {
+    if (isGlobalAdminActor(actor)) return true;
+    const mission = (await this.brain.missions.findById(missionId)) as MissionLike | undefined;
+    if (!mission) return false;
+    return (await this.filterMissionsForActor(actor, [mission])).length === 1;
+  }
+
+  private async assertTaskReferencesAuthorized(
+    actor: McpActorContext,
+    refs: { projectId?: string | null; missionId?: string | null },
+  ): Promise<void> {
+    if (refs.projectId && !(await this.isProjectAuthorized(actor, refs.projectId))) {
+      throw new Error('MCP task project scope denied');
+    }
+    if (refs.missionId && !(await this.isMissionAuthorized(actor, refs.missionId))) {
+      throw new Error('MCP task mission scope denied');
+    }
+  }
+
+  private async assertTaskCreateScopeAuthorized(
+    actor: McpActorContext,
+    refs: { projectId?: string | null; missionId?: string | null },
+  ): Promise<void> {
+    if (!isGlobalAdminActor(actor) && !refs.projectId && !refs.missionId) {
+      throw new Error('MCP task scope denied');
+    }
+    await this.assertTaskReferencesAuthorized(actor, refs);
+  }
+
+  private async filterTasksForActor<T extends TaskLike>(
+    actor: McpActorContext,
+    tasks: T[],
+  ): Promise<T[]> {
+    if (isGlobalAdminActor(actor)) return tasks;
+
+    const [projects, missions] = await Promise.all([
+      this.brain.projects.findAll(),
+      this.brain.missions.findAll(),
+    ]);
+    const projectIds = new Set(
+      filterProjectsForActor(actor, projects as ProjectLike[]).map((project) => project.id),
+    );
+    const missionIds = new Set(
+      (await this.filterMissionsForActor(actor, missions as MissionLike[])).map(
+        (mission) => mission.id,
+      ),
+    );
+
+    return tasks.filter(
+      (task) =>
+        task.userId === actor.userId ||
+        (isTenantAdminActor(actor) && matchesTenant(actor, task)) ||
+        (typeof task.projectId === 'string' && projectIds.has(task.projectId)) ||
+        (typeof task.missionId === 'string' && missionIds.has(task.missionId)),
+    );
   }
 
   /**
    * Registers all platform tools on the given McpServer instance.
    */
-  private registerTools(server: McpServer, _userId: string): void {
+  registerTools(server: McpServer, actor: McpActorContext): void {
     // ─── Brain: Project tools ────────────────────────────────────────────
 
     server.registerTool(
       'brain_list_projects',
       {
         description: 'List all projects in the brain.',
-        inputSchema: z.object({}),
+        inputSchema: strictObject({}),
       },
-      async () => {
-        const projects = await this.brain.projects.findAll();
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'brain_list_projects', params);
+        const projects = filterProjectsForActor(
+          actor,
+          (await this.brain.projects.findAll()) as ProjectLike[],
+        );
         return {
           content: [{ type: 'text' as const, text: JSON.stringify(projects, null, 2) }],
         };
@@ -96,17 +401,21 @@ export class McpService implements OnModuleDestroy {
       'brain_get_project',
       {
         description: 'Get a project by ID.',
-        inputSchema: z.object({
+        inputSchema: strictObject({
           id: z.string().describe('Project ID (UUID)'),
         }),
       },
-      async ({ id }) => {
-        const project = await this.brain.projects.findById(id);
+      async ({ id, ...params }) => {
+        assertMcpToolAuthorized(actor, 'brain_get_project', params);
+        const project = (await this.brain.projects.findById(id)) as ProjectLike | undefined;
+        const authorizedProject = project ? filterProjectsForActor(actor, [project])[0] : undefined;
         return {
           content: [
             {
               type: 'text' as const,
-              text: project ? JSON.stringify(project, null, 2) : `Project not found: ${id}`,
+              text: authorizedProject
+                ? JSON.stringify(authorizedProject, null, 2)
+                : `Project not found: ${id}`,
             },
           ],
         };
@@ -119,20 +428,23 @@ export class McpService implements OnModuleDestroy {
       'brain_list_tasks',
       {
         description: 'List tasks, optionally filtered by project, mission, or status.',
-        inputSchema: z.object({
+        inputSchema: strictObject({
           projectId: z.string().optional().describe('Filter by project ID'),
           missionId: z.string().optional().describe('Filter by mission ID'),
           status: z.string().optional().describe('Filter by status'),
         }),
       },
-      async ({ projectId, missionId, status }) => {
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'brain_list_tasks', params);
+        const { projectId, missionId, status } = params;
         type TaskStatus = 'not-started' | 'in-progress' | 'blocked' | 'done' | 'cancelled';
         let tasks;
         if (projectId) tasks = await this.brain.tasks.findByProject(projectId);
         else if (missionId) tasks = await this.brain.tasks.findByMission(missionId);
         else if (status) tasks = await this.brain.tasks.findByStatus(status as TaskStatus);
         else tasks = await this.brain.tasks.findAll();
-        return { content: [{ type: 'text' as const, text: JSON.stringify(tasks, null, 2) }] };
+        const scopedTasks = await this.filterTasksForActor(actor, tasks as TaskLike[]);
+        return { content: [{ type: 'text' as const, text: JSON.stringify(scopedTasks, null, 2) }] };
       },
     );
 
@@ -140,7 +452,7 @@ export class McpService implements OnModuleDestroy {
       'brain_create_task',
       {
         description: 'Create a new task in the brain.',
-        inputSchema: z.object({
+        inputSchema: strictObject({
           title: z.string().describe('Task title'),
           description: z.string().optional().describe('Task description'),
           projectId: z.string().optional().describe('Project ID'),
@@ -149,6 +461,8 @@ export class McpService implements OnModuleDestroy {
         }),
       },
       async (params) => {
+        assertMcpToolAuthorized(actor, 'brain_create_task', params);
+        await this.assertTaskCreateScopeAuthorized(actor, params);
         type Priority = 'low' | 'medium' | 'high' | 'critical';
         const task = await this.brain.tasks.create({
           ...params,
@@ -162,7 +476,7 @@ export class McpService implements OnModuleDestroy {
       'brain_update_task',
       {
         description: 'Update an existing task.',
-        inputSchema: z.object({
+        inputSchema: strictObject({
           id: z.string().describe('Task ID'),
           title: z.string().optional(),
           description: z.string().optional(),
@@ -171,9 +485,17 @@ export class McpService implements OnModuleDestroy {
             .optional()
             .describe('not-started, in-progress, blocked, done, cancelled'),
           priority: z.string().optional(),
+          projectId: z.string().optional().describe('Project ID'),
+          missionId: z.string().optional().describe('Mission ID'),
         }),
       },
       async ({ id, ...updates }) => {
+        assertMcpToolAuthorized(actor, 'brain_update_task', updates);
+        const existing = (await this.brain.tasks.findById(id)) as TaskLike | undefined;
+        if (!existing || (await this.filterTasksForActor(actor, [existing])).length === 0) {
+          return { content: [{ type: 'text' as const, text: `Task not found: ${id}` }] };
+        }
+        await this.assertTaskReferencesAuthorized(actor, updates);
         type TaskStatus = 'not-started' | 'in-progress' | 'blocked' | 'done' | 'cancelled';
         type Priority = 'low' | 'medium' | 'high' | 'critical';
         const task = await this.brain.tasks.update(id, {
@@ -198,14 +520,19 @@ export class McpService implements OnModuleDestroy {
       'brain_list_missions',
       {
         description: 'List all missions, optionally filtered by project.',
-        inputSchema: z.object({
+        inputSchema: strictObject({
           projectId: z.string().optional().describe('Filter by project ID'),
         }),
       },
-      async ({ projectId }) => {
-        const missions = projectId
-          ? await this.brain.missions.findByProject(projectId)
-          : await this.brain.missions.findAll();
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'brain_list_missions', params);
+        const { projectId } = params;
+        const missions = await this.filterMissionsForActor(
+          actor,
+          (projectId
+            ? await this.brain.missions.findByProject(projectId)
+            : await this.brain.missions.findAll()) as MissionLike[],
+        );
         return { content: [{ type: 'text' as const, text: JSON.stringify(missions, null, 2) }] };
       },
     );
@@ -213,13 +540,12 @@ export class McpService implements OnModuleDestroy {
     server.registerTool(
       'brain_list_conversations',
       {
-        description: 'List conversations for a user.',
-        inputSchema: z.object({
-          userId: z.string().describe('User ID'),
-        }),
+        description: 'List conversations for the authenticated MCP actor.',
+        inputSchema: strictObject({}),
       },
-      async ({ userId }) => {
-        const conversations = await this.brain.conversations.findAll(userId);
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'brain_list_conversations', params);
+        const conversations = await this.brain.conversations.findAll(actor.userId);
         return {
           content: [{ type: 'text' as const, text: JSON.stringify(conversations, null, 2) }],
         };
@@ -232,14 +558,15 @@ export class McpService implements OnModuleDestroy {
       'memory_search',
       {
         description:
-          'Search across stored insights and knowledge using natural language. Returns semantically similar results.',
-        inputSchema: z.object({
-          userId: z.string().describe('User ID to search memory for'),
+          'Search stored insights and knowledge for the authenticated MCP actor using natural language.',
+        inputSchema: strictObject({
           query: z.string().describe('Natural language search query'),
           limit: z.number().optional().describe('Max results (default 5)'),
         }),
       },
-      async ({ userId, query, limit }) => {
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'memory_search', params);
+        const { query, limit } = params;
         if (!this.embeddings.available) {
           return {
             content: [
@@ -251,7 +578,11 @@ export class McpService implements OnModuleDestroy {
           };
         }
         const embedding = await this.embeddings.embed(query);
-        const results = await this.memory.insights.searchByEmbedding(userId, embedding, limit ?? 5);
+        const results = await this.memory.insights.searchByEmbedding(
+          actor.userId,
+          embedding,
+          limit ?? 5,
+        );
         return { content: [{ type: 'text' as const, text: JSON.stringify(results, null, 2) }] };
       },
     );
@@ -259,20 +590,21 @@ export class McpService implements OnModuleDestroy {
     server.registerTool(
       'memory_get_preferences',
       {
-        description: 'Retrieve stored preferences for a user.',
-        inputSchema: z.object({
-          userId: z.string().describe('User ID'),
+        description: 'Retrieve stored preferences for the authenticated MCP actor.',
+        inputSchema: strictObject({
           category: z
             .string()
             .optional()
             .describe('Filter by category: communication, coding, workflow, appearance, general'),
         }),
       },
-      async ({ userId, category }) => {
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'memory_get_preferences', params);
+        const { category } = params;
         type Cat = 'communication' | 'coding' | 'workflow' | 'appearance' | 'general';
         const prefs = category
-          ? await this.memory.preferences.findByUserAndCategory(userId, category as Cat)
-          : await this.memory.preferences.findByUser(userId);
+          ? await this.memory.preferences.findByUserAndCategory(actor.userId, category as Cat)
+          : await this.memory.preferences.findByUser(actor.userId);
         return { content: [{ type: 'text' as const, text: JSON.stringify(prefs, null, 2) }] };
       },
     );
@@ -281,9 +613,8 @@ export class McpService implements OnModuleDestroy {
       'memory_save_preference',
       {
         description:
-          'Store a learned user preference (e.g., "prefers tables over paragraphs", "timezone: America/Chicago").',
-        inputSchema: z.object({
-          userId: z.string().describe('User ID'),
+          'Store a learned preference for the authenticated MCP actor (e.g., "prefers tables over paragraphs").',
+        inputSchema: strictObject({
           key: z.string().describe('Preference key'),
           value: z.string().describe('Preference value (JSON string)'),
           category: z
@@ -292,7 +623,9 @@ export class McpService implements OnModuleDestroy {
             .describe('Category: communication, coding, workflow, appearance, general'),
         }),
       },
-      async ({ userId, key, value, category }) => {
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'memory_save_preference', params);
+        const { key, value, category } = params;
         type Cat = 'communication' | 'coding' | 'workflow' | 'appearance' | 'general';
         let parsedValue: unknown;
         try {
@@ -301,7 +634,7 @@ export class McpService implements OnModuleDestroy {
           parsedValue = value;
         }
         const pref = await this.memory.preferences.upsert({
-          userId,
+          userId: actor.userId,
           key,
           value: parsedValue,
           category: (category as Cat) ?? 'general',
@@ -315,9 +648,8 @@ export class McpService implements OnModuleDestroy {
       'memory_save_insight',
       {
         description:
-          'Store a learned insight, decision, or knowledge extracted from the current interaction.',
-        inputSchema: z.object({
-          userId: z.string().describe('User ID'),
+          'Store a learned insight, decision, or knowledge for the authenticated MCP actor.',
+        inputSchema: strictObject({
           content: z.string().describe('The insight or knowledge to store'),
           category: z
             .string()
@@ -325,11 +657,13 @@ export class McpService implements OnModuleDestroy {
             .describe('Category: decision, learning, preference, fact, pattern, general'),
         }),
       },
-      async ({ userId, content, category }) => {
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'memory_save_insight', params);
+        const { content, category } = params;
         type Cat = 'decision' | 'learning' | 'preference' | 'fact' | 'pattern' | 'general';
         const embedding = this.embeddings.available ? await this.embeddings.embed(content) : null;
         const insight = await this.memory.insights.create({
-          userId,
+          userId: actor.userId,
           content,
           embedding,
           source: 'agent',
@@ -346,16 +680,11 @@ export class McpService implements OnModuleDestroy {
       {
         description:
           'Get the current orchestration mission status including milestones, tasks, and active session.',
-        inputSchema: z.object({
-          projectPath: z
-            .string()
-            .optional()
-            .describe('Project path. Defaults to gateway working directory.'),
-        }),
+        inputSchema: strictObject({}),
       },
-      async ({ projectPath }) => {
-        const resolvedPath = projectPath ?? process.cwd();
-        const status = await this.coordService.getMissionStatus(resolvedPath);
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'coord_mission_status', params);
+        const status = await this.coordService.getMissionStatus(process.cwd());
         return {
           content: [
             {
@@ -371,16 +700,11 @@ export class McpService implements OnModuleDestroy {
       'coord_list_tasks',
       {
         description: 'List all tasks from the orchestration TASKS.md file.',
-        inputSchema: z.object({
-          projectPath: z
-            .string()
-            .optional()
-            .describe('Project path. Defaults to gateway working directory.'),
-        }),
+        inputSchema: strictObject({}),
       },
-      async ({ projectPath }) => {
-        const resolvedPath = projectPath ?? process.cwd();
-        const tasks = await this.coordService.listTasks(resolvedPath);
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'coord_list_tasks', params);
+        const tasks = await this.coordService.listTasks(process.cwd());
         return { content: [{ type: 'text' as const, text: JSON.stringify(tasks, null, 2) }] };
       },
     );
@@ -389,17 +713,14 @@ export class McpService implements OnModuleDestroy {
       'coord_task_detail',
       {
         description: 'Get detailed status for a specific orchestration task.',
-        inputSchema: z.object({
+        inputSchema: strictObject({
           taskId: z.string().describe('Task ID (e.g. P2-005)'),
-          projectPath: z
-            .string()
-            .optional()
-            .describe('Project path. Defaults to gateway working directory.'),
         }),
       },
-      async ({ taskId, projectPath }) => {
-        const resolvedPath = projectPath ?? process.cwd();
-        const detail = await this.coordService.getTaskStatus(resolvedPath, taskId);
+      async (params) => {
+        assertMcpToolAuthorized(actor, 'coord_task_detail', params);
+        const { taskId } = params;
+        const detail = await this.coordService.getTaskStatus(process.cwd(), taskId);
         return {
           content: [
             {

apps/gateway/src/memory/memory.module.ts

@@ -3,8 +3,10 @@ import {
   createMemory,
   type Memory,
   createMemoryAdapter,
+  createOperatorMemoryPlugin,
   type MemoryAdapter,
   type MemoryConfig,
+  type OperatorMemoryPlugin,
 } from '@mosaicstack/memory';
 import type { Db } from '@mosaicstack/db';
 import type { StorageAdapter } from '@mosaicstack/storage';
@@ -14,6 +16,9 @@ import { DB, STORAGE_ADAPTER } from '../database/database.module.js';
 import { MEMORY } from './memory.tokens.js';
 import { MemoryController } from './memory.controller.js';
 import { EmbeddingService } from './embedding.service.js';
+import { redactSensitiveContent } from '@mosaicstack/log';
+
+export const OPERATOR_MEMORY_PLUGIN = 'OPERATOR_MEMORY_PLUGIN';
 
 export const MEMORY_ADAPTER = 'MEMORY_ADAPTER';
 
@@ -38,9 +43,24 @@ function buildMemoryConfig(config: MosaicConfig, storageAdapter: StorageAdapter)
         createMemoryAdapter(buildMemoryConfig(config, storageAdapter)),
       inject: [MOSAIC_CONFIG, STORAGE_ADAPTER],
     },
+    {
+      provide: OPERATOR_MEMORY_PLUGIN,
+      useFactory: (adapter: MemoryAdapter): OperatorMemoryPlugin | null => {
+        const instanceId = process.env['MOSAIC_OPERATOR_MEMORY_INSTANCE_ID']?.trim();
+        const namespace = process.env['MOSAIC_OPERATOR_MEMORY_NAMESPACE']?.trim();
+        if (!instanceId || !namespace) return null;
+        return createOperatorMemoryPlugin({
+          adapter,
+          instanceId,
+          namespace,
+          redact: (content) => redactSensitiveContent(content).content,
+        });
+      },
+      inject: [MEMORY_ADAPTER],
+    },
     EmbeddingService,
   ],
   controllers: [MemoryController],
-  exports: [MEMORY, MEMORY_ADAPTER, EmbeddingService],
+  exports: [MEMORY, MEMORY_ADAPTER, OPERATOR_MEMORY_PLUGIN, EmbeddingService],
 })
 export class MemoryModule {}

apps/gateway/src/plugin/discord-ingress.security.spec.ts

@@ -0,0 +1,443 @@
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import {
+  createDiscordIngressEnvelope,
+  verifyDiscordIngressEnvelope,
+  DiscordPlugin,
+  type DiscordIngressPayload,
+  parseDiscordInteractionBindings,
+  resolveDiscordInteractionActorId,
+  resolveDiscordInteractionBinding,
+} from '@mosaicstack/discord-plugin';
+import { RuntimeProviderService } from '../agent/runtime-provider-registry.service.js';
+import { ChatGateway } from '../chat/chat.gateway.js';
+import { CommandAuthorizationService } from '../commands/command-authorization.service.js';
+import { validateDiscordServiceToken } from '../chat/chat.gateway-auth.js';
+import { DiscordReplayProtector } from './discord-replay-protector.js';
+
+const SERVICE_TOKEN = 'test-service-token';
+const ENV_KEYS = [
+  'DISCORD_SERVICE_TOKEN',
+  'DISCORD_SERVICE_USER_ID',
+  'DISCORD_SERVICE_TENANT_ID',
+  'DISCORD_INTERACTION_BINDINGS',
+  'DISCORD_ALLOWED_GUILD_IDS',
+  'DISCORD_ALLOWED_CHANNEL_IDS',
+  'DISCORD_ALLOWED_USER_IDS',
+  'MOSAIC_AGENT_NAME',
+] as const;
+const savedEnv = new Map<string, string | undefined>();
+
+function configureDiscordEnv(role: 'admin' | 'member' = 'admin'): void {
+  for (const key of ENV_KEYS) savedEnv.set(key, process.env[key]);
+  process.env['DISCORD_SERVICE_TOKEN'] = SERVICE_TOKEN;
+  process.env['DISCORD_SERVICE_USER_ID'] = 'discord-service';
+  process.env['DISCORD_SERVICE_TENANT_ID'] = 'tenant-discord';
+  process.env['MOSAIC_AGENT_NAME'] = 'Nova';
+  process.env['DISCORD_ALLOWED_GUILD_IDS'] = 'guild-001';
+  process.env['DISCORD_ALLOWED_CHANNEL_IDS'] = 'channel-001';
+  process.env['DISCORD_ALLOWED_USER_IDS'] = 'user-001';
+  process.env['DISCORD_INTERACTION_BINDINGS'] = JSON.stringify([
+    {
+      instanceId: 'Nova',
+      guildId: 'guild-001',
+      channelId: 'channel-001',
+      pairedUsers: {
+        'user-001': {
+          role: role === 'admin' ? 'admin' : 'operator',
+          mosaicUserId: 'mosaic-admin-001',
+        },
+      },
+    },
+  ]);
+}
+
+afterEach((): void => {
+  for (const key of ENV_KEYS) {
+    const value = savedEnv.get(key);
+    if (value === undefined) delete process.env[key];
+    else process.env[key] = value;
+  }
+  savedEnv.clear();
+});
+
+function commandAuthorization(role: 'admin' | 'member'): CommandAuthorizationService {
+  const entries = new Map<string, string>();
+  const db = {
+    select: () => ({ from: () => ({ where: () => ({ limit: async () => [{ role }] }) }) }),
+  };
+  const redis = {
+    get: async (key: string) => entries.get(key) ?? null,
+    set: async (key: string, value: string) => entries.set(key, value),
+    del: async (key: string) => Number(entries.delete(key)),
+  };
+  return new CommandAuthorizationService(db as never, redis);
+}
+
+function discordGateway(role: 'admin' | 'member'): {
+  gateway: ChatGateway;
+  client: { data: { discordService: boolean }; emit: ReturnType<typeof vi.fn> };
+  consumedActions: Array<{ actorId: string; correlationId: string }>;
+  durable: { getSnapshot: ReturnType<typeof vi.fn> };
+  audit: { record: ReturnType<typeof vi.fn> };
+} {
+  const authorization = commandAuthorization(role);
+  const consumedActions: Array<{ actorId: string; correlationId: string }> = [];
+  const durable = {
+    getSnapshot: vi.fn().mockResolvedValue({
+      identity: { agentName: 'Nova', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+    }),
+  };
+  const audit = { record: vi.fn().mockResolvedValue(undefined) };
+  const runtimeRegistry = new RuntimeProviderService(
+    {
+      require: () => ({
+        capabilities: async () => ({ supported: ['session.terminate'] }),
+        terminate: async () => undefined,
+      }),
+    } as never,
+    { record: async () => undefined } as never,
+    {
+      consume: async (approvalId, action) => {
+        consumedActions.push({ actorId: action.actorId, correlationId: action.correlationId });
+        return authorization.consumeRuntimeTerminationApproval(approvalId, action);
+      },
+    },
+  );
+  return {
+    gateway: new ChatGateway(
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      {} as never,
+      authorization,
+      runtimeRegistry,
+      durable as never,
+      audit as never,
+    ),
+    client: { data: { discordService: true }, emit: vi.fn() },
+    consumedActions,
+    durable,
+    audit,
+  };
+}
+
+function ingressEnvelope(
+  content: string,
+  messageId: string,
+  overrides: Partial<DiscordIngressPayload> = {},
+): ReturnType<typeof createDiscordIngressEnvelope> {
+  return createDiscordIngressEnvelope(
+    createPayload({ content, messageId, ...overrides }),
+    SERVICE_TOKEN,
+  );
+}
+
+function createPayload(overrides: Partial<DiscordIngressPayload> = {}): DiscordIngressPayload {
+  return {
+    correlationId: 'correlation-001',
+    messageId: 'discord-message-001',
+    guildId: 'guild-001',
+    channelId: 'channel-001',
+    userId: 'user-001',
+    conversationId: 'discord-channel-001',
+    content: 'hello Tess',
+    ...overrides,
+  };
+}
+
+describe('Discord ingress security', () => {
+  it('keeps legacy role-only bindings valid while withholding privileged actor identity', () => {
+    const [binding] = parseDiscordInteractionBindings(
+      JSON.stringify([
+        {
+          instanceId: 'Nova',
+          guildId: 'guild-001',
+          channelId: 'channel-001',
+          pairedUsers: { 'user-001': 'admin' },
+        },
+      ]),
+    );
+    expect(
+      resolveDiscordInteractionBinding([binding!], 'guild-001', 'channel-001', 'user-001', 'send'),
+    ).toEqual(binding);
+    expect(resolveDiscordInteractionActorId(binding!, 'user-001')).toBeNull();
+  });
+
+  it('binds a differently named configured interaction instance without code changes', () => {
+    const binding = resolveDiscordInteractionBinding(
+      [
+        {
+          instanceId: 'Nova',
+          guildId: 'guild-001',
+          channelId: 'channel-001',
+          pairedUsers: { 'user-001': { role: 'operator', mosaicUserId: 'mosaic-operator-001' } },
+        },
+      ],
+      'guild-001',
+      'channel-001',
+      'user-001',
+      'send',
+    );
+
+    expect(binding?.instanceId).toBe('Nova');
+  });
+
+  it('accepts only the configured Discord service identity', () => {
+    expect(validateDiscordServiceToken(SERVICE_TOKEN, SERVICE_TOKEN)).toBe(true);
+    expect(validateDiscordServiceToken('wrong-service-token', SERVICE_TOKEN)).toBe(false);
+    expect(validateDiscordServiceToken(undefined, SERVICE_TOKEN)).toBe(false);
+  });
+
+  it('rejects unauthenticated or tampered service envelopes', () => {
+    const envelope = createDiscordIngressEnvelope(createPayload(), SERVICE_TOKEN);
+
+    expect(verifyDiscordIngressEnvelope(envelope, SERVICE_TOKEN)).toEqual(createPayload());
+    expect(verifyDiscordIngressEnvelope(envelope, 'wrong-service-token')).toBeNull();
+    expect(
+      verifyDiscordIngressEnvelope(
+        { ...envelope, payload: { ...envelope.payload, content: 'forged command' } },
+        SERVICE_TOKEN,
+      ),
+    ).toBeNull();
+  });
+
+  it.each([
+    ['guild', { guildId: 'unlisted-guild' }],
+    ['channel', { channelId: 'unlisted-channel' }],
+    ['user', { userId: 'unlisted-user' }],
+  ])(
+    'rejects an unallowlisted Discord %s',
+    (_kind: string, overrides: Partial<DiscordIngressPayload>) => {
+      const envelope = createDiscordIngressEnvelope(createPayload(overrides), SERVICE_TOKEN);
+
+      expect(
+        verifyDiscordIngressEnvelope(envelope, SERVICE_TOKEN, {
+          guildIds: ['guild-001'],
+          channelIds: ['channel-001'],
+          userIds: ['user-001'],
+        }),
+      ).toBeNull();
+    },
+  );
+
+  it('retains Discord message and correlation IDs after authenticated allowlisted validation', () => {
+    const payload = createPayload({
+      correlationId: 'correlation-trace-123',
+      messageId: 'discord-snowflake-987',
+    });
+    const envelope = createDiscordIngressEnvelope(payload, SERVICE_TOKEN);
+
+    expect(
+      verifyDiscordIngressEnvelope(envelope, SERVICE_TOKEN, {
+        guildIds: ['guild-001'],
+        channelIds: ['channel-001'],
+        userIds: ['user-001'],
+      }),
+    ).toEqual(payload);
+  });
+
+  it('rejects a replayed Discord message ID while retaining bounded replay state', () => {
+    const replayProtector = new DiscordReplayProtector(60_000, 2);
+
+    expect(replayProtector.claim('discord-message-001')).toBe(true);
+    expect(replayProtector.claim('discord-message-001')).toBe(false);
+    expect(replayProtector.claim('discord-message-002')).toBe(true);
+    expect(replayProtector.claim('discord-message-003')).toBe(true);
+    expect(replayProtector.size).toBe(2);
+  });
+
+  it('consumes the exact target once when approval and stop are separate Discord messages', async () => {
+    configureDiscordEnv();
+    const { gateway, client, consumedActions } = discordGateway('admin');
+    await gateway.handleDiscordApproval(
+      client as never,
+      ingressEnvelope('/approve', 'approve-message', {
+        correlationId: 'approval-ingress-correlation',
+      }),
+    );
+    const approval = client.emit.mock.calls.find(
+      ([event]) => event === 'discord:approval',
+    )?.[1] as {
+      approvalId: string;
+      success: boolean;
+    };
+    expect(approval.success).toBe(true);
+
+    await gateway.handleDiscordStop(
+      client as never,
+      ingressEnvelope(`/stop ${approval.approvalId}`, 'stop-message', {
+        correlationId: 'stop-ingress-correlation',
+      }),
+    );
+    expect(client.emit).toHaveBeenCalledWith('discord:stop', {
+      correlationId: 'stop-ingress-correlation',
+      success: true,
+    });
+    expect(consumedActions).toEqual([
+      {
+        actorId: 'mosaic-admin-001',
+        correlationId: expect.stringMatching(/^discord-action:v1:/),
+      },
+    ]);
+  });
+
+  it('audits a Discord mint-side authorization denial', async () => {
+    configureDiscordEnv();
+    const { gateway, client, audit } = discordGateway('member');
+
+    await gateway.handleDiscordApproval(
+      client as never,
+      ingressEnvelope('/approve', 'denied-approve'),
+    );
+
+    expect(client.emit).toHaveBeenCalledWith('discord:approval', {
+      correlationId: 'correlation-001',
+      success: false,
+      approvalId: undefined,
+      expiresAt: undefined,
+    });
+    expect(audit.record).toHaveBeenCalledWith(
+      expect.objectContaining({
+        outcome: 'denied',
+        operation: 'session.terminate',
+        errorCode: 'policy_denied',
+      }),
+    );
+  });
+
+  it.each([
+    [
+      'binding',
+      () => {
+        process.env['MOSAIC_AGENT_NAME'] = 'Other';
+      },
+    ],
+    [
+      'durable session',
+      (durable: { getSnapshot: ReturnType<typeof vi.fn> }) => {
+        durable.getSnapshot.mockResolvedValueOnce({
+          identity: { agentName: 'Other', providerId: 'fleet', runtimeSessionId: 'runtime-1' },
+        });
+      },
+    ],
+  ])(
+    'rejects approval when the %s targets a different runtime agent',
+    async (_source, configure) => {
+      configureDiscordEnv();
+      const { gateway, client, durable } = discordGateway('admin');
+      configure(durable);
+
+      await gateway.handleDiscordApproval(
+        client as never,
+        ingressEnvelope('/approve', 'mismatched-agent-approve'),
+      );
+
+      expect(client.emit).toHaveBeenCalledWith('discord:approval', {
+        correlationId: 'correlation-001',
+        success: false,
+        approvalId: undefined,
+        expiresAt: undefined,
+      });
+    },
+  );
+
+  it('rejects unpaired and non-admin Discord users for approval and stop', async () => {
+    configureDiscordEnv();
+    const { gateway, client } = discordGateway('member');
+    await gateway.handleDiscordApproval(
+      client as never,
+      ingressEnvelope('/approve', 'member-approve'),
+    );
+    expect(client.emit).toHaveBeenCalledWith('discord:approval', {
+      correlationId: 'correlation-001',
+      success: false,
+      approvalId: undefined,
+      expiresAt: undefined,
+    });
+
+    process.env['DISCORD_INTERACTION_BINDINGS'] = JSON.stringify([]);
+    await gateway.handleDiscordStop(
+      client as never,
+      ingressEnvelope('/stop forged', 'unpaired-stop'),
+    );
+    expect(client.emit).not.toHaveBeenCalledWith('discord:stop', expect.anything());
+  });
+
+  it('rejects replaying a Discord-created termination approval', async () => {
+    configureDiscordEnv();
+    const { gateway, client } = discordGateway('admin');
+    await gateway.handleDiscordApproval(
+      client as never,
+      ingressEnvelope('/approve', 'replay-approve', {
+        correlationId: 'replay-approval-correlation',
+      }),
+    );
+    const approval = client.emit.mock.calls.find(
+      ([event]) => event === 'discord:approval',
+    )?.[1] as {
+      approvalId: string;
+    };
+    await gateway.handleDiscordStop(
+      client as never,
+      ingressEnvelope(`/stop ${approval.approvalId}`, 'replay-stop-one', {
+        correlationId: 'replay-stop-correlation-one',
+      }),
+    );
+    await gateway.handleDiscordStop(
+      client as never,
+      ingressEnvelope(`/stop ${approval.approvalId}`, 'replay-stop-two', {
+        correlationId: 'replay-stop-correlation-two',
+      }),
+    );
+    const stopResults = client.emit.mock.calls.filter(([event]) => event === 'discord:stop');
+    expect(stopResults.map(([, result]) => (result as { success: boolean }).success)).toEqual([
+      true,
+      false,
+    ]);
+  });
+
+  it('accepts a thread message through its allowed bound parent channel', () => {
+    const emitted = vi.fn();
+    const plugin = new DiscordPlugin({
+      token: 'unused',
+      gatewayUrl: 'http://unused',
+      serviceToken: SERVICE_TOKEN,
+      allowedGuildIds: ['guild-001'],
+      allowedChannelIds: ['channel-001'],
+      allowedUserIds: ['user-001'],
+      interactionBindings: [
+        {
+          instanceId: 'Nova',
+          guildId: 'guild-001',
+          channelId: 'channel-001',
+          pairedUsers: { 'user-001': { role: 'operator', mosaicUserId: 'mosaic-operator-001' } },
+        },
+      ],
+    });
+    const internals = plugin as unknown as {
+      client: { user: { id: string } };
+      socket: { connected: boolean; emit: ReturnType<typeof vi.fn> };
+      handleDiscordMessage(message: unknown): void;
+    };
+    internals.client = { user: { id: 'bot-001' } };
+    internals.socket = { connected: true, emit: emitted };
+    internals.handleDiscordMessage({
+      id: 'thread-message',
+      guildId: 'guild-001',
+      channelId: 'thread-001',
+      author: { id: 'user-001', bot: false },
+      mentions: { has: () => true },
+      content: '<@bot-001> hello from thread',
+      channel: { parentId: 'channel-001' },
+      attachments: new Map(),
+    });
+
+    const [, envelope] = emitted.mock.calls[0] as [
+      string,
+      ReturnType<typeof createDiscordIngressEnvelope>,
+    ];
+    expect(verifyDiscordIngressEnvelope(envelope, SERVICE_TOKEN)?.channelId).toBe('channel-001');
+  });
+});

apps/gateway/src/plugin/discord-replay-protector.ts

@@ -0,0 +1,40 @@
+/**
+ * Bounded replay cache for Discord's globally unique native message IDs.
+ * Durable ingress idempotency is added with Tess's canonical inbox/outbox work.
+ */
+export class DiscordReplayProtector {
+  private readonly claimedAt = new Map<string, number>();
+
+  constructor(
+    private readonly ttlMs = 15 * 60 * 1000,
+    private readonly maxEntries = 10_000,
+  ) {}
+
+  get size(): number {
+    return this.claimedAt.size;
+  }
+
+  /** Claims an ID exactly once within its bounded retention window. */
+  claim(messageId: string, now = Date.now()): boolean {
+    this.prune(now);
+    if (this.claimedAt.has(messageId)) return false;
+
+    this.claimedAt.set(messageId, now);
+    this.evictOverflow();
+    return true;
+  }
+
+  private prune(now: number): void {
+    for (const [messageId, claimedAt] of this.claimedAt) {
+      if (now - claimedAt >= this.ttlMs) this.claimedAt.delete(messageId);
+    }
+  }
+
+  private evictOverflow(): void {
+    while (this.claimedAt.size > this.maxEntries) {
+      const oldestMessageId = this.claimedAt.keys().next().value;
+      if (oldestMessageId === undefined) return;
+      this.claimedAt.delete(oldestMessageId);
+    }
+  }
+}

apps/gateway/src/plugin/plugin.module.ts

@@ -6,7 +6,7 @@ import {
   type OnModuleDestroy,
   type OnModuleInit,
 } from '@nestjs/common';
-import { DiscordPlugin } from '@mosaicstack/discord-plugin';
+import { DiscordPlugin, parseDiscordInteractionBindings } from '@mosaicstack/discord-plugin';
 import { TelegramPlugin } from '@mosaicstack/telegram-plugin';
 import { PluginService } from './plugin.service.js';
 import type { IChannelPlugin } from './plugin.interface.js';
@@ -50,19 +50,44 @@ class TelegramChannelPluginAdapter implements IChannelPlugin {
 
 const DEFAULT_GATEWAY_URL = 'http://localhost:14242';
 
+function requiredDiscordAllowlist(name: string): string[] {
+  const value = process.env[name]
+    ?.split(',')
+    .map((id: string): string => id.trim())
+    .filter((id: string): boolean => id.length > 0);
+  if (!value || value.length === 0) {
+    throw new Error(`${name} is required when DISCORD_BOT_TOKEN is configured`);
+  }
+  return value;
+}
+
 function createPluginRegistry(): IChannelPlugin[] {
   const plugins: IChannelPlugin[] = [];
   const discordToken = process.env['DISCORD_BOT_TOKEN'];
   const discordGuildId = process.env['DISCORD_GUILD_ID'];
   const discordGatewayUrl = process.env['DISCORD_GATEWAY_URL'] ?? DEFAULT_GATEWAY_URL;
+  const discordServiceToken = process.env['DISCORD_SERVICE_TOKEN'];
+  const discordServiceUserId = process.env['DISCORD_SERVICE_USER_ID'];
 
   if (discordToken) {
+    if (!discordServiceToken || !discordServiceUserId) {
+      throw new Error(
+        'DISCORD_SERVICE_TOKEN and DISCORD_SERVICE_USER_ID are required when DISCORD_BOT_TOKEN is configured',
+      );
+    }
     plugins.push(
       new DiscordChannelPluginAdapter(
         new DiscordPlugin({
           token: discordToken,
           guildId: discordGuildId,
           gatewayUrl: discordGatewayUrl,
+          serviceToken: discordServiceToken,
+          allowedGuildIds: requiredDiscordAllowlist('DISCORD_ALLOWED_GUILD_IDS'),
+          allowedChannelIds: requiredDiscordAllowlist('DISCORD_ALLOWED_CHANNEL_IDS'),
+          allowedUserIds: requiredDiscordAllowlist('DISCORD_ALLOWED_USER_IDS'),
+          interactionBindings: parseDiscordInteractionBindings(
+            process.env['DISCORD_INTERACTION_BINDINGS'],
+          ),
         }),
       ),
     );

apps/gateway/src/preferences/system-override.service.ts

@@ -1,9 +1,13 @@
 import { Injectable, Logger } from '@nestjs/common';
 import { createQueue, type QueueHandle } from '@mosaicstack/queue';
+import type { ActorTenantScope } from '../auth/session-scope.js';
 
-const SESSION_SYSTEM_KEY = (sessionId: string) => `mosaic:session:${sessionId}:system`;
-const SESSION_SYSTEM_FRAGMENTS_KEY = (sessionId: string) =>
-  `mosaic:session:${sessionId}:system:fragments`;
+const scopedSessionId = (sessionId: string, scope: ActorTenantScope) =>
+  `${scope.tenantId}:${scope.userId}:${sessionId}`;
+const SESSION_SYSTEM_KEY = (sessionId: string, scope: ActorTenantScope) =>
+  `mosaic:session:${scopedSessionId(sessionId, scope)}:system`;
+const SESSION_SYSTEM_FRAGMENTS_KEY = (sessionId: string, scope: ActorTenantScope) =>
+  `mosaic:session:${scopedSessionId(sessionId, scope)}:system:fragments`;
 const SYSTEM_OVERRIDE_TTL_SECONDS = 604800; // 7 days
 
 interface OverrideFragment {
@@ -20,9 +24,9 @@ export class SystemOverrideService {
     this.handle = createQueue();
   }
 
-  async set(sessionId: string, override: string): Promise<void> {
+  async set(sessionId: string, override: string, scope: ActorTenantScope): Promise<void> {
     // Load existing fragments
-    const existing = await this.handle.redis.get(SESSION_SYSTEM_FRAGMENTS_KEY(sessionId));
+    const existing = await this.handle.redis.get(SESSION_SYSTEM_FRAGMENTS_KEY(sessionId, scope));
     const fragments: OverrideFragment[] = existing
       ? (JSON.parse(existing) as OverrideFragment[])
       : [];
@@ -37,11 +41,11 @@ export class SystemOverrideService {
     // Store both: fragments array and condensed result
     const pipeline = this.handle.redis.pipeline();
     pipeline.setex(
-      SESSION_SYSTEM_FRAGMENTS_KEY(sessionId),
+      SESSION_SYSTEM_FRAGMENTS_KEY(sessionId, scope),
       SYSTEM_OVERRIDE_TTL_SECONDS,
       JSON.stringify(fragments),
     );
-    pipeline.setex(SESSION_SYSTEM_KEY(sessionId), SYSTEM_OVERRIDE_TTL_SECONDS, condensed);
+    pipeline.setex(SESSION_SYSTEM_KEY(sessionId, scope), SYSTEM_OVERRIDE_TTL_SECONDS, condensed);
     await pipeline.exec();
 
     this.logger.debug(
@@ -49,21 +53,21 @@ export class SystemOverrideService {
     );
   }
 
-  async get(sessionId: string): Promise<string | null> {
-    return this.handle.redis.get(SESSION_SYSTEM_KEY(sessionId));
+  async get(sessionId: string, scope: ActorTenantScope): Promise<string | null> {
+    return this.handle.redis.get(SESSION_SYSTEM_KEY(sessionId, scope));
   }
 
-  async renew(sessionId: string): Promise<void> {
+  async renew(sessionId: string, scope: ActorTenantScope): Promise<void> {
     const pipeline = this.handle.redis.pipeline();
-    pipeline.expire(SESSION_SYSTEM_KEY(sessionId), SYSTEM_OVERRIDE_TTL_SECONDS);
-    pipeline.expire(SESSION_SYSTEM_FRAGMENTS_KEY(sessionId), SYSTEM_OVERRIDE_TTL_SECONDS);
+    pipeline.expire(SESSION_SYSTEM_KEY(sessionId, scope), SYSTEM_OVERRIDE_TTL_SECONDS);
+    pipeline.expire(SESSION_SYSTEM_FRAGMENTS_KEY(sessionId, scope), SYSTEM_OVERRIDE_TTL_SECONDS);
     await pipeline.exec();
   }
 
-  async clear(sessionId: string): Promise<void> {
+  async clear(sessionId: string, scope: ActorTenantScope): Promise<void> {
     await this.handle.redis.del(
-      SESSION_SYSTEM_KEY(sessionId),
-      SESSION_SYSTEM_FRAGMENTS_KEY(sessionId),
+      SESSION_SYSTEM_KEY(sessionId, scope),
+      SESSION_SYSTEM_FRAGMENTS_KEY(sessionId, scope),
     );
     this.logger.debug(`Cleared system override for session ${sessionId}`);
   }

apps/gateway/src/queue/queue.service.ts

@@ -162,6 +162,23 @@ export class QueueService implements OnModuleInit, OnModuleDestroy {
     );
   }
 
+  /**
+   * Remove every existing repeatable schedule for a job name. This supports
+   * safe retirement of previously registered system-wide jobs.
+   */
+  async removeRepeatableJobs(queueName: string, jobName: string): Promise<number> {
+    const queue = this.getQueue(queueName);
+    const jobs = await queue.getRepeatableJobs();
+    const matchingJobs = jobs.filter((job) => job.name === jobName);
+    await Promise.all(matchingJobs.map((job) => queue.removeRepeatableByKey(job.key)));
+    if (matchingJobs.length > 0) {
+      this.logger.log(
+        `Removed ${matchingJobs.length} repeatable "${jobName}" job(s) from "${queueName}"`,
+      );
+    }
+    return matchingJobs.length;
+  }
+
   /**
    * Register a Worker for the given queue name with error handling and
    * exponential backoff.

docs/MISSION-MANIFEST.md

@@ -10,9 +10,9 @@
 **Statement:** Ship a self-hosted, multi-user AI agent platform that consolidates the user's disparate jarvis-brain usage across home and USC workstations into a single coherent system reachable via three first-class surfaces — webUI, TUI, and CLI — with federation as the data-layer mechanism that makes cross-host agent sessions work in real time without copying user data across the boundary.
 **Phase:** Execution (workstream W1 in planning-complete state)
 **Current Workstream:** W1 — Federation v1
-**Progress:** 0 / 1 declared workstreams complete (more workstreams will be declared as scope is refined)
+**Progress:** 0 / 3 declared workstreams complete (more workstreams will be declared as scope is refined)
 **Status:** active (continuous since 2026-03-13)
-**Last Updated:** 2026-04-19 (manifest authored at the rollup level; install-ux-v2 archived; W1 federation planning landed via PR #468)
+**Last Updated:** 2026-07-14 (W3 Native Kanban/SOT canon independently approved under issue #751)
 **Source PRD:** [docs/PRD.md](./PRD.md) — Mosaic Stack v0.1.0
 **Scratchpad:** [docs/scratchpads/mvp-20260312.md](./scratchpads/mvp-20260312.md) (active since 2026-03-13; 14 prior sessions of phase-based execution)
 
@@ -67,10 +67,12 @@ The MVP is complete when ALL declared workstreams are complete AND every cross-c
 
 ## Workstreams
 
-| #   | ID  | Name                                        | Status            | Manifest                                                                | Notes                                               |
-| --- | --- | ------------------------------------------- | ----------------- | ----------------------------------------------------------------------- | --------------------------------------------------- |
-| W1  | FED | Federation v1                               | planning-complete | [docs/federation/MISSION-MANIFEST.md](./federation/MISSION-MANIFEST.md) | 7 milestones, ~175K tokens, issues #460–#466 filed  |
-| W2+ | TBD | (additional workstreams declared as scoped) | —                 | —                                                                       | Scope creep is expected and explicitly accommodated |
+| #   | ID   | Name                                        | Status            | Manifest                                                                              | Notes                                                    |
+| --- | ---- | ------------------------------------------- | ----------------- | ------------------------------------------------------------------------------------- | -------------------------------------------------------- |
+| W1  | FED  | Federation v1                               | planning-complete | [docs/federation/MISSION-MANIFEST.md](./federation/MISSION-MANIFEST.md)               | 7 milestones, ~175K tokens, issues #460–#466 filed       |
+| W2  | TESS | Tess interaction agent                      | planning-complete | [docs/tess/MISSION-MANIFEST.md](./tess/MISSION-MANIFEST.md)                           | 5 milestones; issue #706; M1 issue #707 ready            |
+| W3  | KBN  | Native Kanban and canonical task SOT        | planning-complete | [docs/native-kanban-sot/MISSION-MANIFEST.md](./native-kanban-sot/MISSION-MANIFEST.md) | P0–P3; issue #751; implementation held until canon merge |
+| W4+ | TBD  | (additional workstreams declared as scoped) | —                 | —                                                                                     | Scope creep is expected and explicitly accommodated      |
 
 ### Likely Additional Workstreams (Not Yet Declared)
 

docs/PRD.md

@@ -79,6 +79,102 @@ Jarvis (v0.2.0) is a self-hosted AI assistant with a Python FastAPI backend and
 
 ---
 
+## Tess Interaction Agent Workstream (TESS)
+
+### Problem and Objective
+
+Jason needs one durable, operator-facing Mosaic agent outside Hermes that is reachable through a dedicated Discord channel and CLI, can attach to and operate the Mosaic fleet and transitional Hermes agents, and preserves context across restarts and compaction. Mos remains the coding/general fleet orchestrator; Tess is the complementary human interaction, visibility, control, and migration agent.
+
+The objective is to ship **Tess** (from _tessera_, a piece of a mosaic) as a Pi-native, GPT-5.6 Sol agent with high reasoning. Tess must use Mosaic-owned contracts and plugins so Hermes can be replaced incrementally rather than becoming a permanent architectural dependency.
+
+### Scope
+
+#### In Scope
+
+1. `TESS-ARP-001`: A runtime-neutral `AgentRuntimeProvider` contract supporting `listSessions`, `streamSession`, `sendMessage`, `terminate`, `getSessionTree`, `attach`, health, capability discovery, and normalized events/errors.
+2. `TESS-PI-001`: A long-running Pi-native Tess agent profile/service pinned to GPT-5.6 Sol with high reasoning, explicit tool policy, lifecycle hooks, durable checkpoints, and restart recovery.
+3. `TESS-DSC-001`: Dedicated Discord channel binding to Tess through the Mosaic gateway, with allowlists/RBAC, thread/reply policy, streaming, attachments, approvals, and correlation IDs.
+4. `TESS-CLI-001`: `mosaic tess` CLI commands for chat, status, session listing, attach/detach, send/steer/stop, provider health, and recovery.
+5. `TESS-FLT-001`: Fleet plugin capabilities for roster/status/heartbeat inspection, message delivery, session hierarchy, safe attach, and controlled restart/recovery.
+6. `TESS-MOS-001`: Explicit Mos coordination boundary and tools: hand off orchestration requests, observe mission/task state, receive results, and never silently compete for orchestration authority.
+7. `TESS-HRM-001`: Transitional Hermes adapter for profiles/agents, sessions, streaming/messages, Kanban, skills, memory, tools, cron, and health, using capability negotiation and fail-closed unsupported operations.
+8. `TESS-MEM-001`: Unified memory/retrieval plugin with scoped search/recent/capture/stats, startup context injection, provenance, redaction, namespace isolation, and flat-file/project truth precedence.
+9. `TESS-STA-001`: Durable agent state, inbox, handoff, compaction-recovery, and resume reconstruction.
+10. `TESS-PLG-001`: Plugin/tool catalog covering runtime bootstrap, repository/PR workflow, fleet diagnostics, incident-safe read operations, Discord interaction, and extensible MCP/skill discovery.
+11. `TESS-TRN-001`: Replaceable transport providers: tmux/fleet now, Matrix/native Mosaic transport later, with no Discord/CLI business logic coupled to transport details.
+12. `TESS-SEC-001`: RBAC, per-operation authorization, explicit approval for destructive/privileged/customer-visible actions, audit events, secret/PII redaction, tenant isolation, and bounded command execution.
+13. `TESS-SEC-002`: Command execution SHALL enforce declared scope/role server-side; admin/system and destructive operations SHALL require policy-bound durable approval.
+14. `TESS-SEC-003`: Every session list/read/attach/send/terminate operation SHALL enforce server-derived owner and tenant scope; guessed or client-supplied IDs SHALL grant no authority.
+15. `TESS-SEC-004`: MCP tools SHALL derive actor/tenant from authenticated context and SHALL NOT accept caller-controlled identity fields.
+16. `TESS-SEC-005`: Discord plugin ingress SHALL authenticate service identity, enforce guild/channel/user allowlists, propagate correlation/message IDs, and reject replay.
+17. `TESS-SEC-006`: Secret/PII classification and redaction SHALL occur before persistence and before channel egress, including tool metadata and authentication flows.
+18. `TESS-SEC-007`: Approvals SHALL be one-time, expiring, actor/tenant-bound, and cryptographically bound to the exact structured action digest.
+19. `TESS-SEC-008`: Ingress, provider sends, tool side effects, and responses SHALL use durable inbox/outbox/checkpoints and idempotency records for restart-safe replay.
+20. `TESS-SEC-009`: Garbage collection and retention SHALL be session/tenant scoped unless executed as a separately authorized and audited system-wide job.
+21. `TESS-OBS-001`: Structured logs, traces, health/readiness, provider latency/errors, session lifecycle, tool audit, and actionable recovery diagnostics.
+22. `TESS-MIG-001`: Capability inventory and staged Hermes-to-Mosaic migration matrix with coexistence, cutover, rollback, and deprecation gates.
+
+#### Out of Scope
+
+1. Replacing Mos as coding/general fleet orchestrator.
+2. Making Hermes the Mosaic core or coupling Mosaic domain logic to Hermes schemas.
+3. Migrating every historical chat verbatim; only policy-compliant indexed summaries and user-selected sessions are migrated.
+4. Unrestricted shell execution from Discord.
+5. Full web UI parity in the first Tess operational milestone; gateway contracts must remain web-consumable.
+6. Replacing tmux before Matrix/native transport reaches operational parity.
+
+### Stakeholder and User Requirements
+
+- Jason must be able to converse with the same Tess session from Discord and CLI.
+- Jason must be able to see what is running, stale, blocked, or unhealthy without attaching manually to every session.
+- Jason must be able to attach to Tess and authorized fleet sessions through supported CLI controls.
+- Tess must collaborate with Mos and the fleet while preserving a single clear orchestration authority.
+- The system must migrate useful Hermes/OpenClaw capabilities intentionally, with evidence, instead of copying implementations wholesale.
+
+### Non-Functional Requirements
+
+1. **Security:** default-deny provider/tool capabilities, least privilege, no secrets in logs/prompts/commits, Discord user/channel authorization, and auditable approvals.
+2. **Reliability:** durable inbox/checkpoints; idempotent message handling; reconnect with bounded backoff; no message loss or duplicate execution across gateway restart.
+3. **Performance:** first acknowledgement within 2 seconds when connected; streamed agent output begins within 5 seconds excluding model/provider delay; status reads return within 2 seconds under nominal local conditions.
+4. **Observability:** every ingress message and resulting provider/tool operation carries a correlation ID across Discord, gateway, Tess, provider, and audit events.
+5. **Maintainability:** channel, runtime, transport, memory, and external-agent integrations remain adapter-based with contract tests.
+6. **Privacy:** only scoped context enters external runtimes; persisted messages/memories follow retention and redaction policy.
+7. **Portability:** Tess runs through Pi/Mosaic contracts and does not require Hermes to start or serve native Mosaic operations.
+
+### Acceptance Criteria
+
+1. `AC-TESS-01`: A dedicated Discord channel and `mosaic tess chat` connect to one durable Tess session and stream responses bidirectionally.
+2. `AC-TESS-02`: `mosaic tess status|sessions|tree|attach|send|stop` operate against authorized provider capabilities with stable typed outputs and actionable errors.
+3. `AC-TESS-03`: Tess runs GPT-5.6 Sol at high reasoning and its effective runtime/model/tool policy is visible through status without exposing credentials.
+4. `AC-TESS-04`: Tess can inspect and message the Mosaic fleet, hand orchestration work to Mos, and demonstrate that Tess does not independently claim Mos-owned orchestration work.
+5. `AC-TESS-05`: Hermes adapter demonstrates session listing, streaming/message delivery, hierarchy mapping, and at least one approved capability in each of Kanban, skills, memory, tools, and cron—or reports unsupported capabilities fail-closed.
+6. `AC-TESS-06`: Restart/compaction test preserves session identity, pending inbox, last durable checkpoint, and a resumable handoff without duplicate side effects.
+7. `AC-TESS-07`: Unauthorized Discord users/channels, cross-tenant access, unsafe tool calls, forged approvals, and sensitive-output cases are denied and audited.
+8. `AC-TESS-08`: tmux/fleet and Matrix/native transport implementations pass the same provider contract suite; Matrix may remain non-default until readiness gates pass.
+9. `AC-TESS-09`: Baseline quality gates, unit/integration/contract tests, Discord+CLI E2E, restart/recovery tests, independent code review, and security review are green.
+10. `AC-TESS-10`: Migration matrix documents every audited Hermes/OpenClaw capability as native, adapted, deferred, or rejected, with cutover and rollback evidence.
+11. `AC-TESS-11`: User, admin, developer, API/OpenAPI, operations/recovery, and plugin-authoring documentation is current and linked from the sitemap.
+
+### Constraints, Dependencies, Risks, and Assumptions
+
+- Dependency: Mosaic gateway remains the single API surface; Pi is the native runtime; Valkey/PostgreSQL provide canonical durable state where required.
+- Dependency: Discord bot credentials and dedicated channel ID are deployment secrets provisioned outside source control.
+- Risk: Tess could drift into a second orchestrator. Mitigation: explicit role policy, Mos handoff contract, authority checks, and E2E boundary tests.
+- Risk: broad Hermes compatibility can freeze legacy semantics into Mosaic. Mitigation: Mosaic-owned normalized contracts and capability negotiation.
+- Risk: Discord creates a privileged remote-control surface. Mitigation: pairing/allowlists, RBAC, approvals, rate limits, audit, and safe tool classes.
+- Risk: transcript ingestion can violate privacy or overload memory. Mitigation: scoped opt-in import, redacted summaries, provenance, retention, and deduplication.
+- Risk: current root filesystem has limited headroom. Mitigation: isolated worktrees, no duplicated dependency installation unless required, and cleanup only after active-lane verification.
+- `ASSUMPTION:` The public name is **Tess**, because the user requested a name and the tessera/Mosaic relationship is distinctive; config must permit later display-name changes without renaming APIs or storage keys.
+- `ASSUMPTION:` The dedicated Discord channel ID and final guild policy will be supplied/provisioned during deployment, so implementation uses explicit configuration and fail-fast startup validation.
+- `ASSUMPTION:` tmux/fleet is the production transport for the first operational milestone; Matrix/native transport is implemented behind the same contract and promoted only after parity/reliability verification.
+- `ASSUMPTION:` Project/task truth remains in canonical Mosaic/project stores; semantic memory systems are retrieval/mirror layers, not hidden authorities.
+
+### Testing and Delivery Intent
+
+Delivery uses five gated milestones: runtime contracts/security; Pi service/state; Discord/CLI; fleet/Hermes/plugin suite; migration/Matrix/recovery/qualification. Every source-code task requires tests, independent review, a PR to `main`, terminal-green CI, and issue/task closure. Production activation additionally requires a clean-host Pi launch, dedicated Discord channel smoke test, CLI attach test, restart/recovery drill, and rollback procedure.
+
+---
+
 ## Architecture
 
 ### High-Level System Diagram
@@ -1004,3 +1100,139 @@ All work is **alpha** (< 0.1.0) until Jason approves 0.1.0 beta release.
 10. ASSUMPTION: **Conversations and messages get their own PG tables** (not stored in brain's entity model). They follow a chat-specific schema with proper foreign keys to users and projects. Rationale: Chat has different access patterns (streaming, pagination, search) than brain entities.
 
 11. RESOLVED: **Pi handles all target LLM providers natively.** Anthropic, OpenAI/Codex, Z.ai, Ollama, LM Studio, and llama.cpp are all supported via Pi's built-in providers or `models.json` configuration with `openai-completions` API type. No custom provider adapters needed in @mosaicstack/agent — only configuration management.
+
+---
+
+## Fleet Declarative Configuration Management (#758)
+
+### Status and objective
+
+- **Requirement ID:** `FCM-PRD-001`
+- **Status:** approved architecture; M0 documentation gate in progress
+- **Authority:** issue #758 and the independently approved baseline plan
+
+Provide one understandable, schema-validated lifecycle for the local Mosaic fleet. The operator-owned YAML/JSON roster is the canonical desired-state input. Generated agent environment files, systemd enablement, tmux sessions, heartbeat state, and installed framework assets are derived or observed state. Mutations must pass through one shared compiler/reconciler and must be previewable, atomic where possible, recoverable, automation-safe, and non-destructive toward unmanaged resources.
+
+### Normative scope
+
+#### In scope for M0–M5
+
+1. A narrow v2 YAML/JSON roster for local tmux/systemd fleets.
+2. One executable structural contract with schema/parser parity and canonical snake_case output.
+3. Semantic validation through the existing baseline plus `roles.local` profile/persona/provision resolver; a parallel role resolver is forbidden.
+4. Canonical classes `code`, `review`, `security-review`, `validator`, `merge-gate`, `orchestrator`, `team-leader`, `enhancer`, and `interaction`, including documented legacy aliases.
+5. Read/validate/plan/apply/migrate, full local fleet-agent CRUD, lifecycle, status, verify, stable JSON output, and documented exit codes.
+6. Deterministic `.env.generated` projections, a strict non-shell `.env.local` allowlist, generation/digest stamping, and fail-closed quarantine of forbidden legacy keys.
+7. v1 inventory, preview, field-complete migration, canary cutover, rollback, compatibility aliases, and explicit disposition of every shipped example/profile.
+8. Documentation, packaging/update checks, clean-install/cold-start dogfood, and independent correctness, security, validator, and merge gates.
+
+#### Out of scope for M0–M5
+
+- Kubernetes-style resource envelopes.
+- Remote/SSH reconciliation or distributed placement mutation.
+- Connector/Matrix/Discord lifecycle mutation.
+- Secret-reference or credential-provider schema.
+- Arbitrary command or channel overrides.
+- Gateway `/api/agents` mapping, control-plane convergence, UI configuration storage, or rename of that separate DB-backed catalog.
+- Live-fleet mutation during M0.
+
+Each excluded capability requires a separate post-M5 PRD and threat model. Existing v1 remote/connector fields are inventory-only: local apply must reject them without invoking systemd or tmux.
+
+### Authority and identity decisions
+
+- The roster owns fleet membership, launch policy, and persisted lifecycle intent.
+- Role/persona contracts are product reference data; `roles.local` is the update-surviving local extension layer.
+- Tess and Ultron are configurable instance/display names, not schema identities.
+- `validator` issues the independent final validation certificate but cannot approve-to-land or merge.
+- `merge-gate` remains the sole approve-to-land and merge authority after required review, security, validation, CI, and queue gates.
+- `orchestrator` may apply validated owner-policy-compliant topology changes and grant/revoke bounded capacity leases.
+- `team-leader` may accept/release and use a named lease but cannot change global topology, re-lease capacity, or gain merge authority.
+- `review` and `security-review` provide independent correctness and security records respectively; neither authors the reviewed change.
+- `interaction` is request/status only. `enhancer` proposes fleet improvements. `code` authors implementation but cannot self-review.
+- Operator policy remains the exception, pause, and lease-revocation boundary.
+
+A capacity lease names existing agents, purpose, and expiry. It never changes class, runtime, tools, credentials, roster ownership, or merge authority.
+
+### Lifecycle and generated-state decisions
+
+The normative dimensions are `enabled`, persisted `desired_state: running|stopped`, and observed `running|stopped|error|unknown|unmanaged`.
+
+1. Fresh create defaults to enabled and stopped; `create --start` persists running.
+2. v1 migration preserves known observed running/stopped state. Unknown state blocks apply for that entry.
+3. Start/stop without `--persist` is transient and reports drift; the next apply/reboot restores persisted intent.
+4. Start/stop with `--persist` atomically updates generation and converges the local unit/session.
+5. Restart does not change desired state and rejects stopped entries unless explicitly started.
+6. Apply acts only on local, enabled, roster-owned entries. Ownership must be proven before stale projections are quarantined or removed; fuzzy names never authorize stopping an unmanaged session.
+7. Rollback restores roster/projection generation and captured unit enablement, stops processes introduced by the failed generation, and never starts an agent that was stopped before cutover.
+8. The systemd unit reads only `%i.env.generated` after migration. `.env.local` is parsed as data, cannot shadow authoritative keys, and never uses shell `source`, `eval`, expansion, or command substitution.
+9. `MOSAIC_AGENT_COMMAND`, channel flags, credential variables, and unknown agent keys are forbidden. Migration reports key names and content hashes only—never values—and blocks launch/apply until disposition.
+
+### Functional requirements
+
+| ID        | Requirement                                                                                                                                                                                  |
+| --------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| FCM-FR-01 | Show and validate YAML/JSON using one structural contract and shared semantic role/topology validation.                                                                                      |
+| FCM-FR-02 | Produce a deterministic, non-mutating desired-versus-observed plan covering roster, projections, units, sessions, installed assets, and orphans.                                             |
+| FCM-FR-03 | Apply under a lock with expected-generation checks, atomic writes/backups, ordered convergence, post-verification, and machine-readable recovery data.                                       |
+| FCM-FR-04 | Provide create, inspect, update, remove, list, start, stop, restart, status, validate, reconcile, doctor, dry-run, and automation-safe operations.                                           |
+| FCM-FR-05 | Validate duplicate names, unsupported classes/runtimes/models/options, topology cycles, missing role contracts, stale generated state, unmanaged sessions, unit drift, and socket ambiguity. |
+| FCM-FR-06 | Generate deterministic, mode-0600, digest-stamped launch projections; safely parse only allowlisted local operational overrides.                                                             |
+| FCM-FR-07 | Read v1 for one deprecation window, write v2 after migration, preserve known lifecycle intent, and provide preview/canary/rollback.                                                          |
+| FCM-FR-08 | Classify every shipped example/profile as migrated, versioned compatibility fixture, or retired with replacement.                                                                            |
+| FCM-FR-09 | Report desired, observed, generation, drift, readiness, ownership, and failing plane without exposing privileged values.                                                                     |
+| FCM-FR-10 | Keep gateway-backed `mosaic agent` records explicitly separate from local `mosaic fleet` desired state.                                                                                      |
+
+### Non-functional requirements
+
+- **Security:** fail closed on command/credential/unknown overrides; reject traversal, injection, shadowing, and unauthorized topology/lifecycle actions; never expose secret or command values.
+- **Reliability:** lock plus expected generation; temporary write, fsync, atomic rename, recoverable backup; deterministic idempotent replay; ordered rollback on partial failure.
+- **Safety:** no destructive inference from stale names; no local actions for remote/schema-only entries; stopped agents remain stopped through migration and reboot.
+- **Compatibility:** canonical snake_case serialization with bounded v1 camelCase/alias input support and explicit warnings.
+- **Observability:** stable text/JSON status, drift, plan, migration, and recovery output with documented exit codes `0` success, `2` invalid, `3` drift, `4` conflict, `5` partial failure, and `6` policy denial.
+- **Maintainability:** schema, roster load, profiles, provision, migration, and apply share the existing role-resolution implementation.
+- **Documentation:** every field, command, transition, migration rule, recovery workflow, class power, and example is linked from the fleet docs IA and validated in CI.
+
+### Acceptance criteria
+
+| ID        | Acceptance criterion                                                                                                                                                   |
+| --------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| FCM-AC-01 | `docs/PRD.md`, `docs/TASKS.md`, docs-IA checklist, and example/profile inventory are approved before implementation.                                                   |
+| FCM-AC-02 | YAML and JSON positive/negative, round-trip, unknown-field, enum, duplicate, topology, and property/fuzz tests prove schema/parser parity and canonical serialization. |
+| FCM-AC-03 | Every class resolves through the existing profile/persona resolver; authority and lease tests enforce the normative matrix.                                            |
+| FCM-AC-04 | Every shipped example/profile has a CI-valid migrate/compatibility/retire disposition with no unresolved class at M1 exit.                                             |
+| FCM-AC-05 | Validate/show/plan are deterministic and non-mutating; JSON shapes and exit codes are contract-tested.                                                                 |
+| FCM-AC-06 | Generated/local env precedence, mode, digest, forbidden shadowing, command injection, quarantine, and no-value diagnostics pass independent security tests.            |
+| FCM-AC-07 | CRUD is generation-guarded, atomic/recoverable, idempotent, concurrency-tested, and creates stopped agents unless start is explicitly persisted.                       |
+| FCM-AC-08 | Apply/lifecycle exactly implements the transition contract, including transient/persisted operations, reboot, partial failure, and rollback.                           |
+| FCM-AC-09 | Remote/schema-only and unmanaged entries receive zero local lifecycle calls; local targeting covers named and default tmux sockets exactly.                            |
+| FCM-AC-10 | Status/verify/doctor expose all state planes and actionable drift without secret, credential, or privileged command values.                                            |
+| FCM-AC-11 | v1 migration handles every mapped field, known/unknown observed state, aliases, env quarantine, current 9-managed/3-unmanaged synthetic fixture, canary, and rollback. |
+| FCM-AC-12 | `fleet add/remove` compatibility aliases and v1 reads remain for the approved deprecation window while v2 writers emit only v2.                                        |
+| FCM-AC-13 | Package/install/update tests prove schema, tools, units, roles, docs, and examples ship together while site-owned state survives.                                      |
+| FCM-AC-14 | Fleet documentation checklist is complete, links/format/examples validate, and operator recovery procedures match tested behavior.                                     |
+| FCM-AC-15 | Independent correctness review, security review, validator certificate, terminal-green CI, and merge-gate approval complete before issue closure.                      |
+
+### Risks and mitigations
+
+| Risk                                              | Mitigation / verification                                                                                |
+| ------------------------------------------------- | -------------------------------------------------------------------------------------------------------- |
+| Schema and parser drift create false validation   | One executable contract or bidirectional parity tests; shared semantic resolver.                         |
+| Apply starts intentionally stopped agents         | Persist separate desired state; migration preserves known observed state; reboot/rollback tests.         |
+| Preserved env files become a hidden control plane | Generated-only unit input; strict local allowlist; shadow rejection and quarantine before start.         |
+| Command, secret, or credential values leak        | Values never enter v2 output; key-name/hash-only diagnostics; adversarial security tests and review.     |
+| Stale artifacts cause destructive cleanup         | Proof-of-ownership requirement; unmanaged/remote zero-call tests; deterministic plan before apply.       |
+| Concurrent writers or crashes corrupt roster      | Lock, expected generation, fsync/rename, backup, failure injection, and recovery plan.                   |
+| New compiler duplicates role logic                | Hard prohibition on parallel resolver; parity tests across profile, provision, roster, migration, apply. |
+| Control-plane naming confuses automation          | Explicit local `mosaic fleet` versus gateway DB catalog documentation; no implicit mapping in M1–M5.     |
+| Legacy examples silently teach invalid classes    | Complete disposition inventory and M1 CI exit gate.                                                      |
+
+### Verification and milestone intent
+
+- **M0:** requirements, authority, lifecycle, migration mapping, TASKS DAG, docs IA, and legacy inventory approved; no implementation or live mutation.
+- **M1:** narrow v2 compiler, shared resolver, roles/aliases, validate/show/plan, and all shipped example/profile dispositions.
+- **M2:** safe launch projection and generation-guarded atomic CRUD; command/credential quarantine proven before lifecycle.
+- **M3:** local-only apply/lifecycle/status/verify/doctor implementing the full transition table.
+- **M4:** field-complete v1 migration, compatibility window, orphan inventory, canary, and rollback.
+- **M5:** accepted documentation IA, package/update checks, clean-install dogfood, independent correctness/security/validator evidence, and merge-gate release approval.
+
+Detailed delivery dependencies and acceptance mappings are canonical in `docs/TASKS.md`. Documentation acceptance is tracked in [`docs/scratchpads/758-fleet-config-docs-ia-checklist.md`](scratchpads/758-fleet-config-docs-ia-checklist.md), and shipped artifact disposition is inventoried in [`docs/tasks/758-legacy-example-profile-disposition.md`](tasks/758-legacy-example-profile-disposition.md).

docs/SITEMAP.md

@@ -0,0 +1,50 @@
+# Documentation Sitemap
+
+## Fleet declarative configuration management
+
+- [Normative requirements](PRD.md#fleet-declarative-configuration-management-758) — issue #758 scope, authority, lifecycle, migration, acceptance, risks, and milestones.
+- [M0–M5 delivery DAG](TASKS.md#w4--fleet-declarative-configuration-management-758) — one-card/one-PR implementation order and independent gates.
+- [Documentation IA acceptance checklist](scratchpads/758-fleet-config-docs-ia-checklist.md) — required paths, owners, evidence, and exit checks.
+- [Legacy example/profile disposition inventory](tasks/758-legacy-example-profile-disposition.md) — shipped artifacts and M1 migration decisions.
+
+## Native Kanban and canonical task SOT
+
+- [Canonical requirements](requirements/native-kanban-sot.md) — ratified P0–P3 requirements and acceptance criteria.
+- [Workstream index](native-kanban-sot/INDEX.md) — artifact map, lane partition, and delivery order.
+- [Mission manifest](native-kanban-sot/MISSION-MANIFEST.md) — scope, authority, invariants, and gate model.
+- [Task decomposition](native-kanban-sot/TASKS.md) — dependency-ordered implementation slices and ownership boundaries.
+- [Frozen shared contract](native-kanban-sot/SHARED-CONTRACT.md) — schema, API, Coordinator, health, recovery, and migration contracts.
+- [Initial independent review](reports/native-kanban-sot/canon-initial-review-no-go.md) — KCR-001–016 findings that blocked the first draft.
+- [Final independent re-review](reports/native-kanban-sot/canon-final-rereview-go.md) — closure evidence and GO verdict.
+- [Ultron final gate](reports/native-kanban-sot/ultron-final-go.md) — final requirements, authority, schema, migration, recovery, and evidence review.
+
+## Tess interaction agent
+
+### Operator guides
+
+- [User guide](tess/USER-GUIDE.md) — authorized session, attach, send, stop, and handoff workflows.
+- [Admin guide](tess/ADMIN-GUIDE.md) — deployment configuration, policy, and approval controls.
+- [Developer guide](tess/DEVELOPER-GUIDE.md) — provider contracts, scope boundaries, and test workflow.
+- [Plugin guide](tess/PLUGIN-GUIDE.md) — adapter, redaction, and identity-as-data requirements.
+- [Operations guide](tess/OPERATIONS-GUIDE.md) — readiness, recovery, and incident-safe procedures.
+
+### Architecture and security
+
+- [Architecture](tess/ARCHITECTURE.md)
+- [Threat model](tess/THREAT-MODEL.md)
+- [Mos coordination boundary](tess/MOS-COORDINATION.md)
+- [Hermes runtime adapter design](tess/hermes-runtime-adapter-design.md)
+- [Operator plugin sketch](tess/M4-003-OPERATOR-PLUGIN-SKETCH.md)
+
+### API contract
+
+- [Tess OpenAPI contract](openapi-tess.yaml)
+
+### Migration and qualification
+
+- [Migration inventory](tess/M5-MIGRATION-INVENTORY.md)
+- [Cutover procedure](tess/M5-MIGRATION-CUTOVER.md)
+- [Rollback procedure](tess/M5-MIGRATION-ROLLBACK.md)
+- [Retention and deprecation evidence](tess/M5-MIGRATION-RETENTION-DEPRECATION.md)
+- [Verification matrix](tess/VERIFICATION-MATRIX.md)
+- [Documentation checklist](tess/M5-003-DOCUMENTATION-CHECKLIST.md)

docs/TASKS.md

@@ -14,9 +14,12 @@
 
 ## Workstream Rollup
 
-| id  | status            | workstream          | progress         | tasks file                                        | notes                                                           |
-| --- | ----------------- | ------------------- | ---------------- | ------------------------------------------------- | --------------------------------------------------------------- |
-| W1  | planning-complete | Federation v1 (FED) | 0 / 7 milestones | [docs/federation/TASKS.md](./federation/TASKS.md) | M1 task breakdown populated; M2–M7 deferred to mission planning |
+| id  | status            | workstream               | progress          | tasks file                                                          | notes                                                                            |
+| --- | ----------------- | ------------------------ | ----------------- | ------------------------------------------------------------------- | -------------------------------------------------------------------------------- |
+| W1  | planning-complete | Federation v1 (FED)      | 0 / 7 milestones  | [docs/federation/TASKS.md](./federation/TASKS.md)                   | M1 task breakdown populated; M2–M7 deferred to mission planning                  |
+| W2  | planning-complete | Tess interaction agent   | 0 / 5 milestones  | [docs/tess/TASKS.md](./tess/TASKS.md)                               | Issue #706; independent planning gate PASS; M1 issue #707 ready                  |
+| W3  | planning-complete | Native Kanban/SOT        | 0 / 4 phases      | [docs/native-kanban-sot/TASKS.md](./native-kanban-sot/TASKS.md)     | Issue #751; canon independently approved; implementation held until canon merges |
+| W4  | in-progress       | Fleet declarative config | M0 / 6 milestones | [W4 DAG below](#w4--fleet-declarative-configuration-management-758) | Issue #758; M0 requirements/docs only; implementation blocked on M0 gates        |
 
 ## Cross-Cutting Tracking
 
@@ -54,3 +57,100 @@ Active workstream is **W1 — Federation v1**. Workers should:
 ## P6 — Docs, compliance matrix, alpha tag (#606) — feat/p6-docs-compliance-alpha
 
 - Status: in-repo deliverables done (CONTRIBUTING.md + harness×gate compliance matrix + check-resident-budget.sh + CI wiring + ALPHA-DOD.md). Remaining: alpha tag v0.0.39-alpha (Lead, post-merge). aiguide reconcile merged (#8). Detail: scratchpads/p6-docs-compliance-alpha.md.
+
+## F3-m3 — mosaic update re-seeds framework + relaunches agents (#609) — feat/f3-m3-update-reseed
+
+- Status: implemented + tested. Closes R13: `mosaic update` now re-seeds the framework (data-safe MOSAIC_SYNC_ONLY) after the CLI install so shipped launcher/runtime changes activate; `--relaunch` restarts rostered agents; `--no-reseed` opts out. Detail: scratchpads/f3-m3-update-reseed.md.
+
+## Fleet-polish bundle — boot-survival symmetry (#611) — feat/fleet-polish-bundle
+
+- Status: MERGED to main. disable-on-remove (boot-resurrection bug, TDD) + add-enable + init-R5 hard guarantee. 4 new + 147 existing fleet tests green. Detail: scratchpads/fleet-polish-bundle.md.
+
+## Fleet enhancer role + two-agent floor (#614) — feat/fleet-enhancer-floor
+
+- Status: MERGED to main. enhancer added to 4 presets; init guarantees 1 orchestrator + >=1 enhancer; remove protects the sole enhancer; enhancer role doc. 155 fleet tests green. Detail: scratchpads/fleet-enhancer-floor.md.
+
+## F4 — Orchestrator chat connector + Matrix (#616) — feat/f4-matrix-connector
+
+- Status: Phase 1 MERGED (#617: connector interface send/subscribe/health + registry + roster schema + design). Phase 2a (#618): Matrix CS-API client + factory. 20 connector tests green; no fleet.ts changes. Remaining Phase 2: init/configure connector-selection UX + roster wiring, systemd launch wiring, Conduit deploy guide. Detail: scratchpads/f4-matrix-connector.md.
+
+## Fleet onboarding-injection — comms cheat-sheet + peer roster (#620) — feat/fleet-comms-onboarding
+
+- Status: implemented + tested. Injects # Fleet Comms (peer roster + cross-host agent-send commands + FLIP-reply + --verify) into each spawned fleet agent via composeContract; optional per-agent host/ssh/socket roster fields (socket: named → -L, unset → default socket no -L). 10 + 2 tests green. Detail: scratchpads/fleet-comms-onboarding.md.
+
+## Fleet stand-up fixes — model_hint→--model + socket-default trap (#626) — feat/fleet-standup-fixes
+
+- Status: implemented + tested. FIX1 model_hint→MOSAIC_AGENT_MODEL→--model. FIX2 absent socket = default tmux socket (no -L) across parse/spawn/systemd-unit/observe (socketArgs helper, bare-empty shellEnvValue, conditional -L). 158 fleet tests green; shipped presets unaffected (explicit socket_name). Detail: scratchpads/fleet-standup-fixes.md.
+
+## north-star doctrine consolidation — doc PR — feat/north-star-doctrine
+
+- Status: applied Mos's consolidated merge-map to docs/fleet/north-star.md (budget governance + control plane/central register + 200k cap + delegation + unified-identity Fleet + role-based naming + tmux security + drift re-captures). Doctrine only; #622/#623/#625/#628 out-of-scope. Conflict checklist green. Detail: scratchpads/north-star-doctrine.md.
+
+## #631 — re-seed preserves user fleet data (CRITICAL) — fix/631-reseed-preserves-fleet-data
+
+- Status: implemented + tested. PRIMARY: install.sh PRESERVE_PATHS += fleet/\*.yaml + fleet/agents + fleet/run (glob-aware cp-fallback); TS parity. SECONDARY: refreshActiveFleetUnits propagates unit fixes to ~/.config/systemd/user on mosaic update. bash F6 + TS + unit tests green. Detail: scratchpads/631-reseed-preserves-fleet.md.
+
+## #633 — comms-block emitter + FLEET-LAUNCH runbook — feat/633-comms-block-runbook
+
+- Status: implemented + tested (TDD). `mosaic fleet comms-block <role> [--host]` wraps resolveCommsBlock → readFleetCommsBlock; fails loud (stderr + exit 1) on unknown role / missing roster instead of silent empty. docs/fleet/FLEET-LAUNCH.md runbook: worker path + orchestrator .env fold (MOSAIC_AGENT_COMMAND; line-41 [-z] short-circuits line-44 yolo hardcode) + 3 launch gotchas + #632 preserve note + North-Star 4-field arc (harness ✅/model ✅ roster-native today; yolo + command/channels = PATH B #636). 177 fleet+comms tests green (6 new resolveCommsBlock cases). PATH A of the A→B→webUI arc. Detail: scratchpads/633-comms-block-runbook.md.
+
+---
+
+## W4 — Fleet declarative configuration management (#758)
+
+**Rules:** The table below is the canonical M0–M5 dependency DAG. Each delivery card owns one short-lived branch and one PR. Gate cards (`*-ROR`, `*-SEC`, `*-VAL`, `*-MERGE`) independently attest to the referenced delivery PR and do not author that PR. No implementation starts until `FCM-M0-MERGE` is complete. `done` requires merged PR, terminal-green CI, and linked tracking closure/evidence.
+
+| id           | status      | description                                                                                                          | issue | agent  | repo  | branch                             | depends_on                                                        | estimate | notes                                                                                                              |
+| ------------ | ----------- | -------------------------------------------------------------------------------------------------------------------- | ----- | ------ | ----- | ---------------------------------- | ----------------------------------------------------------------- | -------- | ------------------------------------------------------------------------------------------------------------------ |
+| FCM-M0-01    | in-progress | Ratify requirements, authority/lifecycle/migration decisions, DAG, docs IA checklist, and shipped artifact inventory | #758  | haiku  | stack | docs/issue-758-m0                  | —                                                                 | 18K      | One docs-only PR; maps FCM-AC-01; no source/schema/roles/examples/systemd/live changes                             |
+| FCM-M0-ROR   | not-started | Independent requirements/content review of M0 PR                                                                     | #758  | sonnet | stack | —                                  | FCM-M0-01                                                         | 8K       | Verify approved plan fidelity, DAG completeness, links, and every card→AC mapping; non-author attestation          |
+| FCM-M0-SEC   | not-started | Independent security review of authority, quarantine, lifecycle, and migration requirements                          | #758  | sonnet | stack | —                                  | FCM-M0-01                                                         | 8K       | Threat-model requirements only; verify no secret-value handling and no surprise-start path; non-author attestation |
+| FCM-M0-VAL   | not-started | Validator certificate for M0 acceptance baseline                                                                     | #758  | sonnet | stack | —                                  | FCM-M0-ROR, FCM-M0-SEC                                            | 6K       | Confirm FCM-AC-01 and no unresolved architecture blocker; validator cannot merge                                   |
+| FCM-M0-MERGE | not-started | Merge-gate approval and squash merge of M0 PR                                                                        | #758  | haiku  | stack | —                                  | FCM-M0-VAL                                                        | 3K       | Terminal-green CI required; unlocks implementation                                                                 |
+| FCM-M1-01    | not-started | Implement narrow v2 executable schema, canonical serialization, and schema/parser parity suite                       | #758  | codex  | stack | feat/fcm-v2-contract               | FCM-M0-MERGE                                                      | 30K      | One PR; FCM-AC-02; structural validation only, no lifecycle mutation                                               |
+| FCM-M1-02    | not-started | Share existing profile/persona/provision resolver for roster semantic validation and topology policy                 | #758  | codex  | stack | feat/fcm-shared-role-validation    | FCM-M1-01                                                         | 28K      | One PR; FCM-AC-03; parallel resolver forbidden                                                                     |
+| FCM-M1-03    | not-started | Add/ratify validator, team-leader, interaction role contracts, aliases, authority, and lease tests                   | #758  | codex  | stack | feat/fcm-role-authority            | FCM-M1-02                                                         | 24K      | One PR; FCM-AC-03; merge-gate remains sole merger                                                                  |
+| FCM-M1-04    | not-started | Resolve every shipped example/profile disposition and add CI validation through shared contract/resolver             | #758  | codex  | stack | feat/fcm-example-profile-migration | FCM-M1-03                                                         | 28K      | One PR; FCM-AC-04; inventory rows cannot remain decision-required                                                  |
+| FCM-M1-05    | not-started | Implement non-mutating config show, validate, and deterministic plan with stable JSON/exit codes                     | #758  | codex  | stack | feat/fcm-config-read-plan          | FCM-M1-02                                                         | 32K      | One PR; FCM-AC-05, FCM-AC-09, FCM-AC-10                                                                            |
+| FCM-M1-DOC   | not-started | Publish v2 fields, roles/leases, desired-vs-observed, and example/profile disposition docs                           | #758  | haiku  | stack | docs/fcm-m1-contract               | FCM-M1-03, FCM-M1-04, FCM-M1-05                                   | 16K      | One PR; FCM-AC-14; update sitemap and docs checklist evidence                                                      |
+| FCM-M1-ROR   | not-started | Independent correctness review of all M1 delivery PRs                                                                | #758  | sonnet | stack | —                                  | FCM-M1-01, FCM-M1-02, FCM-M1-03, FCM-M1-04, FCM-M1-05, FCM-M1-DOC | 14K      | Exact-head reviews; schema/parser/resolver parity and docs checked                                                 |
+| FCM-M1-SEC   | not-started | Independent security review of validation, authority, aliases, and input hardening                                   | #758  | sonnet | stack | —                                  | FCM-M1-01, FCM-M1-02, FCM-M1-03, FCM-M1-04, FCM-M1-05             | 12K      | Fuzz/injection/topology/policy findings must be resolved                                                           |
+| FCM-M1-VAL   | not-started | Validator certificate for M1 contract/compiler exit                                                                  | #758  | sonnet | stack | —                                  | FCM-M1-ROR, FCM-M1-SEC                                            | 8K       | Certify FCM-AC-02–05 and no mutation/lifecycle path                                                                |
+| FCM-M1-MERGE | not-started | Merge-gate approval for M1 completion                                                                                | #758  | haiku  | stack | —                                  | FCM-M1-VAL                                                        | 4K       | All M1 PRs merged, terminal-green, inventory resolved                                                              |
+| FCM-M2-01    | not-started | Implement deterministic mode-0600 `.env.generated` projection with generation/digest stamps                          | #758  | codex  | stack | feat/fcm-generated-env             | FCM-M1-MERGE                                                      | 30K      | One PR; FCM-AC-06; no unit launch migration yet                                                                    |
+| FCM-M2-02    | not-started | Implement strict data-only `.env.local` parser, shadow rejection, and forbidden legacy-key quarantine                | #758  | codex  | stack | feat/fcm-local-env-quarantine      | FCM-M2-01                                                         | 34K      | One PR; FCM-AC-06; never output values or privileged commands                                                      |
+| FCM-M2-03    | not-started | Migrate generic unit/launcher to generated input with fail-closed digest validation                                  | #758  | codex  | stack | feat/fcm-launch-chain              | FCM-M2-02                                                         | 32K      | One PR; FCM-AC-06; old `%i.env` cannot launch v2                                                                   |
+| FCM-M2-04    | not-started | Implement generation-guarded atomic fleet-agent create/get/list/update/delete and compatibility aliases              | #758  | codex  | stack | feat/fcm-atomic-crud               | FCM-M2-03                                                         | 38K      | One PR; FCM-AC-07, FCM-AC-12; create defaults stopped; no apply engine                                             |
+| FCM-M2-DOC   | not-started | Publish generated-env chain, quarantine, and CRUD operator/developer guides                                          | #758  | haiku  | stack | docs/fcm-m2-projection-crud        | FCM-M2-02, FCM-M2-03, FCM-M2-04                                   | 16K      | One PR; FCM-AC-14; synthetic values only                                                                           |
+| FCM-M2-ROR   | not-started | Independent correctness review of M2 projection and CRUD PRs                                                         | #758  | sonnet | stack | —                                  | FCM-M2-01, FCM-M2-02, FCM-M2-03, FCM-M2-04, FCM-M2-DOC            | 14K      | Crash/concurrency/idempotency/permissions review                                                                   |
+| FCM-M2-SEC   | not-started | Independent security review of launch chain, overrides, quarantine, paths, and diagnostics                           | #758  | sonnet | stack | —                                  | FCM-M2-01, FCM-M2-02, FCM-M2-03, FCM-M2-04                        | 16K      | Adversarial shell/systemd/tmux/path/secret tests; FCM-AC-06–07                                                     |
+| FCM-M2-VAL   | not-started | Validator certificate for M2 safe-projection/CRUD exit                                                               | #758  | sonnet | stack | —                                  | FCM-M2-ROR, FCM-M2-SEC                                            | 8K       | Prove no hidden launch authority or surprise starts                                                                |
+| FCM-M2-MERGE | not-started | Merge-gate approval for M2 completion                                                                                | #758  | haiku  | stack | —                                  | FCM-M2-VAL                                                        | 4K       | All M2 PRs merged and terminal-green                                                                               |
+| FCM-M3-01    | not-started | Implement locked local-only config apply with ordered convergence and machine-readable recovery                      | #758  | codex  | stack | feat/fcm-local-apply               | FCM-M2-MERGE                                                      | 40K      | One PR; FCM-AC-08–10; zero calls for remote/unmanaged entries                                                      |
+| FCM-M3-02    | not-started | Implement transient/persisted start, stop, restart, and fleet-wide lifecycle transitions                             | #758  | codex  | stack | feat/fcm-lifecycle                 | FCM-M3-01                                                         | 36K      | One PR; FCM-AC-08–09; exact socket targeting                                                                       |
+| FCM-M3-03    | not-started | Implement status, verify, and doctor desired/observed/generation/drift/readiness contracts                           | #758  | codex  | stack | feat/fcm-status-doctor             | FCM-M3-01                                                         | 30K      | One PR; FCM-AC-09–10; safe effective output only                                                                   |
+| FCM-M3-04    | not-started | Add failure-injection, reboot/linger, unmanaged ownership, socket, and rollback integration suite                    | #758  | codex  | stack | test/fcm-lifecycle-recovery        | FCM-M3-02, FCM-M3-03                                              | 32K      | One PR; FCM-AC-08–10                                                                                               |
+| FCM-M3-DOC   | not-started | Publish CLI, lifecycle, status/drift, reconcile/recover, and systemd/tmux troubleshooting docs                       | #758  | haiku  | stack | docs/fcm-m3-operations             | FCM-M3-02, FCM-M3-03, FCM-M3-04                                   | 18K      | One PR; FCM-AC-14                                                                                                  |
+| FCM-M3-ROR   | not-started | Independent correctness review of M3 lifecycle/recovery PRs                                                          | #758  | sonnet | stack | —                                  | FCM-M3-01, FCM-M3-02, FCM-M3-03, FCM-M3-04, FCM-M3-DOC            | 16K      | Exact targeting, state transitions, recovery ordering                                                              |
+| FCM-M3-SEC   | not-started | Independent security review of apply/lifecycle authority and unmanaged-resource protection                           | #758  | sonnet | stack | —                                  | FCM-M3-01, FCM-M3-02, FCM-M3-03, FCM-M3-04                        | 16K      | Policy denial, injection, TOCTOU, no-value output                                                                  |
+| FCM-M3-VAL   | not-started | Validator certificate for M3 local lifecycle exit                                                                    | #758  | sonnet | stack | —                                  | FCM-M3-ROR, FCM-M3-SEC                                            | 10K      | Certify full transition table and FCM-AC-08–10                                                                     |
+| FCM-M3-MERGE | not-started | Merge-gate approval for M3 completion                                                                                | #758  | haiku  | stack | —                                  | FCM-M3-VAL                                                        | 4K       | All M3 PRs merged and terminal-green                                                                               |
+| FCM-M4-01    | not-started | Implement field-complete v1 inventory, preview, aliases, unsupported-field reporting, and v2 writer                  | #758  | codex  | stack | feat/fcm-v1-migrator               | FCM-M3-MERGE                                                      | 38K      | One PR; FCM-AC-11–12; no mutation without `--write`                                                                |
+| FCM-M4-02    | not-started | Implement observed-state preservation, canary cutover, orphan classification, and reversible rollback                | #758  | codex  | stack | feat/fcm-migration-cutover         | FCM-M4-01                                                         | 40K      | One PR; FCM-AC-11; unknown state blocks; stopped stays stopped                                                     |
+| FCM-M4-03    | not-started | Add synthetic 9-managed/3-unmanaged migration, env quarantine, upgrade, and rollback E2E fixtures                    | #758  | codex  | stack | test/fcm-migration-e2e             | FCM-M4-02                                                         | 34K      | One PR; FCM-AC-11–12; no real credential/live-host data                                                            |
+| FCM-M4-DOC   | not-started | Publish v1→v2 field map, aliases, example disposition, backup/restore, and migration runbook                         | #758  | haiku  | stack | docs/fcm-m4-migration              | FCM-M4-01, FCM-M4-02, FCM-M4-03                                   | 18K      | One PR; FCM-AC-14                                                                                                  |
+| FCM-M4-ROR   | not-started | Independent correctness review of M4 migration/cutover PRs                                                           | #758  | sonnet | stack | —                                  | FCM-M4-01, FCM-M4-02, FCM-M4-03, FCM-M4-DOC                       | 16K      | Field completeness, state preservation, rollback fidelity                                                          |
+| FCM-M4-SEC   | not-started | Independent security review of migration inventory, quarantine, and cutover                                          | #758  | sonnet | stack | —                                  | FCM-M4-01, FCM-M4-02, FCM-M4-03                                   | 16K      | Secret-safe reporting and non-destructive ownership proof                                                          |
+| FCM-M4-VAL   | not-started | Validator certificate for M4 compatibility/migration exit                                                            | #758  | sonnet | stack | —                                  | FCM-M4-ROR, FCM-M4-SEC                                            | 10K      | Certify FCM-AC-11–12 and rollback evidence                                                                         |
+| FCM-M4-MERGE | not-started | Merge-gate approval for M4 completion                                                                                | #758  | haiku  | stack | —                                  | FCM-M4-VAL                                                        | 4K       | All M4 PRs merged and terminal-green                                                                               |
+| FCM-M5-01    | not-started | Complete fleet documentation IA, sitemap, validated examples, and checklist evidence                                 | #758  | haiku  | stack | docs/fcm-complete-ia               | FCM-M4-MERGE                                                      | 28K      | One PR; FCM-AC-14; no required checklist item incomplete                                                           |
+| FCM-M5-02    | not-started | Add package/install/update asset-drift and site-owned-state preservation qualification                               | #758  | codex  | stack | test/fcm-package-upgrade           | FCM-M4-MERGE                                                      | 32K      | One PR; FCM-AC-13                                                                                                  |
+| FCM-M5-03    | not-started | Run clean-home install, cold-start, local canary, rolling restart, failure, and rollback qualification               | #758  | codex  | stack | test/fcm-dogfood-qualification     | FCM-M5-01, FCM-M5-02                                              | 30K      | One PR; FCM-AC-08, FCM-AC-11, FCM-AC-13; synthetic harness/evidence only; never mutate production fleet            |
+| FCM-M5-ROR   | not-started | Independent final correctness and documentation review                                                               | #758  | sonnet | stack | —                                  | FCM-M5-01, FCM-M5-02, FCM-M5-03                                   | 16K      | Verify FCM-AC-01–14 evidence and docs links/examples                                                               |
+| FCM-M5-SEC   | not-started | Independent final security review and threat-gate closure                                                            | #758  | sonnet | stack | —                                  | FCM-M5-01, FCM-M5-02, FCM-M5-03                                   | 18K      | Review launch/migration/lifecycle authority, secret handling, recovery                                             |
+| FCM-M5-VAL   | not-started | Ultron/validator final acceptance certificate                                                                        | #758  | sonnet | stack | —                                  | FCM-M5-ROR, FCM-M5-SEC                                            | 12K      | Independent certificate for FCM-AC-01–15; no merge authority                                                       |
+| FCM-M5-MERGE | not-started | Merge-gate final approve-to-land, terminal CI verification, issue closure, and release handoff                       | #758  | haiku  | stack | —                                  | FCM-M5-VAL                                                        | 6K       | Sole merge path; FCM-AC-15; squash merge and close #758 after green CI                                             |
+
+### W4 acceptance mapping check
+
+Every delivery card maps to at least one `FCM-AC-*` criterion in its notes. Gate cards verify those mappings rather than introducing implementation. The detailed documentation checklist is [`docs/scratchpads/758-fleet-config-docs-ia-checklist.md`](scratchpads/758-fleet-config-docs-ia-checklist.md); the shipped artifact inventory is [`docs/tasks/758-legacy-example-profile-disposition.md`](tasks/758-legacy-example-profile-disposition.md).

docs/architecture/channel-protocol.md

@@ -232,6 +232,10 @@ The following sections document how each supported channel maps its native messa
 
 **Outbound:** Adapter calls Discord REST `POST /channels/{id}/messages`. Markdown content is sent as-is (Discord renders it). For `contentType = "code"` the adapter wraps in triple-backtick fences with the `metadata.language` tag.
 
+### Discord service ingress security
+
+The Discord adapter is an authenticated gateway service, not an anonymous Socket.IO client. It presents `DISCORD_SERVICE_TOKEN` during its `/chat` connection and signs each inbound envelope using HMAC-SHA-256. The envelope contains the Discord native message ID and a generated correlation ID. Gateway verifies the service credential, signature, and configured guild/channel/user allowlists before agent dispatch, then rejects duplicate native message IDs inside its bounded replay window. All three allowlists are default-deny and required when the Discord plugin is enabled. The service credential is injected at runtime and is never logged or included in protocol payloads.
+
 ---
 
 ### Telegram

docs/federation/TASKS.md

@@ -91,22 +91,22 @@ Goal: Two federated gateways exchange real data over mTLS. Inbound requests pass
 >
 > **Tracking issue:** #462.
 
-| id        | status      | description                                                                                                                                                                                                                                                                                            | issue | agent  | branch                               | depends_on       | estimate | notes                                                                                                                                                    |
-| --------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----- | ------ | ------------------------------------ | ---------------- | -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| FED-M3-01 | not-started | `packages/types/src/federation/` — request/response DTOs for `list`, `get`, `capabilities` verbs. Wire-format zod schemas + inferred TS types. Includes `FederationRequest`, `FederationListResponse<T>`, `FederationGetResponse<T>`, `FederationCapabilitiesResponse`, error envelope, `_source` tag. | #462  | sonnet | feat/federation-m3-types             | —                | 4K       | Reusable from gateway server + client + harness. Pure types — no I/O, no NestJS.                                                                         |
-| FED-M3-02 | not-started | `tools/federation-harness/` scaffold: `docker-compose.two-gateways.yml` (Server A + Server B + step-CA), `seed.ts` (provisions grants, peers, sample tasks/notes/credentials per scope variant), `harness.ts` helper (boots stack, returns typed clients). README documents harness use.               | #462  | sonnet | feat/federation-m3-harness           | DEPLOY-04 (soft) | 8K       | Falls back to local docker-compose if `mos-test-1/-2` not yet redeployed (DEPLOY chain blocked on IMG-FIX). Permanent test infra used by M3+.            |
-| FED-M3-03 | not-started | `apps/gateway/src/federation/server/federation-auth.guard.ts` (NestJS guard). Validates inbound client cert from Fastify TLS context, extracts `grantId` + `subjectUserId` from custom OIDs, loads grant from DB, asserts `status='active'`, attaches `FederationContext` to request.                  | #462  | sonnet | feat/federation-m3-auth-guard        | M3-01            | 8K       | Reuses OID parsing logic mirrored from `ca.service.ts` post-issuance verification. 401 on malformed/missing OIDs; 403 on revoked/expired/missing grant.  |
-| FED-M3-04 | not-started | `apps/gateway/src/federation/server/scope.service.ts`. Pipeline: (1) resource allowlist + excluded check, (2) native RBAC eval as `subjectUserId`, (3) scope filter intersection (`include_teams`, `include_personal`), (4) `max_rows_per_query` cap. Pure service — DB calls injected.                | #462  | sonnet | feat/federation-m3-scope-service     | M3-01            | 10K      | Hardest correctness target in M3. Reuses `parseFederationScope` (M2-03). Returns either `{ allowed: true, filter }` or structured deny reason for audit. |
-| FED-M3-05 | not-started | `apps/gateway/src/federation/server/verbs/list.controller.ts`. Wires AuthGuard → ScopeService → tasks/notes/memory query layer; applies row cap; tags rows with `_source`. Resource selector via path param.                                                                                           | #462  | sonnet | feat/federation-m3-verb-list         | M3-03, M3-04     | 6K       | Routes: `POST /api/federation/v1/list/:resource`. No body persistence. Audit write deferred to M4.                                                       |
-| FED-M3-06 | not-started | `apps/gateway/src/federation/server/verbs/get.controller.ts`. Single-resource fetch by id; same pipeline as list. 404 on not-found, 403 on RBAC/scope deny — both audited the same way.                                                                                                                | #462  | sonnet | feat/federation-m3-verb-get          | M3-03, M3-04     | 6K       | `POST /api/federation/v1/get/:resource/:id`. Mirrors list controller patterns.                                                                           |
-| FED-M3-07 | not-started | `apps/gateway/src/federation/server/verbs/capabilities.controller.ts`. Read-only enumeration: returns `{ resources, excluded_resources, max_rows_per_query, supported_verbs }` derived from grant scope. Always allowed for an active grant — no RBAC eval.                                            | #462  | sonnet | feat/federation-m3-verb-capabilities | M3-03            | 4K       | `GET /api/federation/v1/capabilities`. Smallest verb; useful sanity check that mTLS + auth guard work end-to-end.                                        |
-| FED-M3-08 | not-started | `apps/gateway/src/federation/client/federation-client.service.ts`. Outbound mTLS dialer: picks `(certPem, sealed clientKey)` from `federation_peers`, unwraps key, builds undici Agent with mTLS, calls peer verb, parses typed response, wraps non-2xx into `FederationClientError`.                  | #462  | sonnet | feat/federation-m3-client            | M3-01            | 8K       | Independent of server stream — can land in parallel with M3-03/04. Cert/key cached per-peer; flushed by future M5/M6 logic.                              |
-| FED-M3-09 | not-started | `apps/gateway/src/federation/client/query-source.service.ts`. Accepts `source: "local" \| "federated:<host>" \| "all"` from gateway query layer; for `"all"` fans out to local + each peer in parallel; merges results; tags every row with `_source`.                                                 | #462  | sonnet | feat/federation-m3-query-source      | M3-08            | 8K       | Per-peer failure surfaces as `_partial: true` in response, not hard failure (sets up M5 offline UX). M5 adds caching + circuit breaker on top.           |
-| FED-M3-10 | not-started | Integration tests for MILESTONES.md M3 acceptance #6 (malformed OIDs → 401; valid cert + revoked grant → 403) and #7 (`max_rows_per_query` cap). Real PG, mocked TLS context (Fastify req shim).                                                                                                       | #462  | sonnet | feat/federation-m3-integration       | M3-05, M3-06     | 8K       | Vitest profile gated by `FEDERATED_INTEGRATION=1`. Single-gateway suite; no harness required.                                                            |
-| FED-M3-11 | not-started | E2E tests for MILESTONES.md M3 acceptance #1, #2, #3, #4, #5, #8, #9, #10 (8 cases). Uses harness from M3-02; two real gateways, real Step-CA, real mTLS. Each test asserts both happy-path response and audit/no-persist invariants.                                                                  | #462  | sonnet | feat/federation-m3-e2e               | M3-02, M3-09     | 12K      | Largest single task. Each acceptance gets its own `it(...)` for clear failure attribution.                                                               |
-| FED-M3-12 | not-started | Independent security review (sonnet, not author of M3-03/04/05/06/07/08/09): focus on cert-SAN spoofing, OID extraction edge cases, scope-bypass via filter manipulation, RBAC-bypass via subjectUser swap, response leakage when scope deny.                                                          | #462  | sonnet | feat/federation-m3-security-review   | M3-11            | 10K      | Two review rounds budgeted. PRD requires explicit test for every 401/403 path — review verifies coverage.                                                |
-| FED-M3-13 | not-started | Docs update: `docs/federation/SETUP.md` mTLS handshake section, new `docs/federation/HARNESS.md` for federation-harness usage, OID reference table in SETUP.md, scope enforcement pipeline diagram. Runbook still M7-deferred.                                                                         | #462  | haiku  | feat/federation-m3-docs              | M3-12            | 5K       | One ASCII diagram for the auth-guard → scope → RBAC pipeline; helps future reviewers reason about denial paths.                                          |
-| FED-M3-14 | not-started | PR aggregate close, CI green, merge to main, close #462. Release tag `fed-v0.3.0-m3`. Update mission manifest M3 row → done; M4 row → in-progress when work begins.                                                                                                                                    | #462  | sonnet | chore/federation-m3-close            | M3-13            | 3K       | Same close pattern as M1-12 / M2-13.                                                                                                                     |
+| id        | status      | description                                                                                                                                                                                                                                                                                            | issue | agent  | branch                               | depends_on                        | estimate | notes                                                                                                                                                    |
+| --------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----- | ------ | ------------------------------------ | --------------------------------- | -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| FED-M3-01 | done        | `packages/types/src/federation/` — request/response DTOs for `list`, `get`, `capabilities` verbs. Wire-format zod schemas + inferred TS types. Includes `FederationRequest`, `FederationListResponse<T>`, `FederationGetResponse<T>`, `FederationCapabilitiesResponse`, error envelope, `_source` tag. | #462  | sonnet | feat/federation-m3-types             | —                                 | 4K       | Reusable from gateway server + client + harness. Pure types — no I/O, no NestJS.                                                                         |
+| FED-M3-02 | done        | `tools/federation-harness/` scaffold: `docker-compose.two-gateways.yml` (Server A + Server B + step-CA), `seed.ts` (provisions grants, peers, sample tasks/notes/credentials per scope variant), `harness.ts` helper (boots stack, returns typed clients). README documents harness use.               | #462  | sonnet | feat/federation-m3-harness           | DEPLOY-04 (soft)                  | 8K       | Falls back to local docker-compose if `mos-test-1/-2` not yet redeployed (DEPLOY chain blocked on IMG-FIX). Permanent test infra used by M3+.            |
+| FED-M3-03 | done        | `apps/gateway/src/federation/server/federation-auth.guard.ts` (NestJS guard). Validates inbound client cert from Fastify TLS context, extracts `grantId` + `subjectUserId` from custom OIDs, loads grant from DB, asserts `status='active'`, attaches `FederationContext` to request.                  | #462  | sonnet | feat/federation-m3-auth-guard        | M3-01                             | 8K       | Reuses OID parsing logic mirrored from `ca.service.ts` post-issuance verification. 401 on malformed/missing OIDs; 403 on revoked/expired/missing grant.  |
+| FED-M3-04 | in-progress | `apps/gateway/src/federation/server/scope.service.ts`. Pipeline: (1) resource allowlist + excluded check, (2) native RBAC eval as `subjectUserId`, (3) scope filter intersection (`include_teams`, `include_personal`), (4) `max_rows_per_query` cap. Pure service — DB calls injected.                | #462  | sonnet | feat/federation-m3-scope-service     | M3-01                             | 10K      | Hardest correctness target in M3. Reuses `parseFederationScope` (M2-03). Returns either `{ allowed: true, filter }` or structured deny reason for audit. |
+| FED-M3-05 | in-progress | `apps/gateway/src/federation/server/verbs/list.controller.ts`. Wires AuthGuard → ScopeService → tasks/notes/memory query layer; applies row cap; tags rows with `_source`. Resource selector via path param.                                                                                           | #462  | sonnet | feat/federation-m3-verb-list         | M3-03, M3-04                      | 6K       | Routes: `POST /api/federation/v1/list/:resource`. No body persistence. Audit write deferred to M4.                                                       |
+| FED-M3-06 | not-started | `apps/gateway/src/federation/server/verbs/get.controller.ts`. Single-resource fetch by id; same pipeline as list. 404 on not-found, 403 on RBAC/scope deny — both audited the same way.                                                                                                                | #462  | sonnet | feat/federation-m3-verb-get          | M3-03, M3-04                      | 6K       | `POST /api/federation/v1/get/:resource/:id`. Mirrors list controller patterns.                                                                           |
+| FED-M3-07 | done        | `apps/gateway/src/federation/server/verbs/capabilities.controller.ts`. Read-only enumeration: returns `{ resources, excluded_resources, max_rows_per_query, supported_verbs }` derived from grant scope. Always allowed for an active grant — no RBAC eval.                                            | #462  | sonnet | feat/federation-m3-verb-capabilities | M3-03                             | 4K       | `GET /api/federation/v1/capabilities`. Smallest verb; useful sanity check that mTLS + auth guard work end-to-end.                                        |
+| FED-M3-08 | done        | `apps/gateway/src/federation/client/federation-client.service.ts`. Outbound mTLS dialer: picks `(certPem, sealed clientKey)` from `federation_peers`, unwraps key, builds undici Agent with mTLS, calls peer verb, parses typed response, wraps non-2xx into `FederationClientError`.                  | #462  | sonnet | feat/federation-m3-client            | M3-01                             | 8K       | Independent of server stream — can land in parallel with M3-03/04. Cert/key cached per-peer; flushed by future M5/M6 logic.                              |
+| FED-M3-09 | done        | `apps/gateway/src/federation/client/query-source.service.ts`. Accepts `source: "local" \| "federated:<host>" \| "all"` from gateway query layer; for `"all"` fans out to local + each peer in parallel; merges results; tags every row with `_source`.                                                 | #462  | sonnet | feat/federation-m3-query-source      | M3-08                             | 8K       | Per-peer failure surfaces as `_partial: true` in response, not hard failure (sets up M5 offline UX). M5 adds caching + circuit breaker on top.           |
+| FED-M3-10 | not-started | Integration tests for MILESTONES.md M3 acceptance #6 (malformed OIDs → 401; valid cert + revoked grant → 403) and #7 (`max_rows_per_query` cap). Real PG, mocked TLS context (Fastify req shim).                                                                                                       | #462  | sonnet | feat/federation-m3-integration       | M3-05, M3-06                      | 8K       | Vitest profile gated by `FEDERATED_INTEGRATION=1`. Single-gateway suite; no harness required.                                                            |
+| FED-M3-11 | not-started | E2E tests for MILESTONES.md M3 acceptance #1, #2, #3, #4, #5, #8, #9, #10 (8 cases). Uses harness from M3-02; two real gateways, real Step-CA, real mTLS. Each test asserts both happy-path response and audit/no-persist invariants.                                                                  | #462  | sonnet | feat/federation-m3-e2e               | M3-02, M3-04, M3-05, M3-06, M3-09 | 12K      | Largest single task. Each acceptance gets its own `it(...)` for clear failure attribution.                                                               |
+| FED-M3-12 | not-started | Independent security review (sonnet, not author of M3-03/04/05/06/07/08/09): focus on cert-SAN spoofing, OID extraction edge cases, scope-bypass via filter manipulation, RBAC-bypass via subjectUser swap, response leakage when scope deny.                                                          | #462  | sonnet | feat/federation-m3-security-review   | M3-11                             | 10K      | Two review rounds budgeted. PRD requires explicit test for every 401/403 path — review verifies coverage.                                                |
+| FED-M3-13 | not-started | Docs update: `docs/federation/SETUP.md` mTLS handshake section, new `docs/federation/HARNESS.md` for federation-harness usage, OID reference table in SETUP.md, scope enforcement pipeline diagram. Runbook still M7-deferred.                                                                         | #462  | haiku  | feat/federation-m3-docs              | M3-12                             | 5K       | One ASCII diagram for the auth-guard → scope → RBAC pipeline; helps future reviewers reason about denial paths.                                          |
+| FED-M3-14 | not-started | PR aggregate close, CI green, merge to main, close #462. Release tag `fed-v0.3.0-m3`. Update mission manifest M3 row → done; M4 row → in-progress when work begins.                                                                                                                                    | #462  | sonnet | chore/federation-m3-close            | M3-13                             | 3K       | Same close pattern as M1-12 / M2-13.                                                                                                                     |
 
 **M3 estimate:** ~100K tokens (vs MILESTONES.md 40K — same per-task breakdown pattern as M1/M2: tests, review, and docs split out from implementation cost). Largest milestone in the federation mission.
 
@@ -118,6 +118,10 @@ Goal: Two federated gateways exchange real data over mTLS. Inbound requests pass
 
 **Test bed fallback:** If `mos-test-1.woltje.com` / `mos-test-2.woltje.com` are still blocked on `FED-M2-DEPLOY-IMG-FIX` when M3-11 is ready to run, the harness's local `docker-compose.two-gateways.yml` is a sufficient stand-in. Production-host validation moves to M7 acceptance suite (PRD AC-12).
 
+**Backlog sync — 2026-06-24 (orchestrator):** Status reconciled against `origin/main` (release 0.0.48). Landed on main: **FED-M3-01** (DTOs, PR #506), **FED-M3-02** (harness scaffold, PR #505), **FED-M3-03** (mTLS auth-guard, PR #509 — CRIT-1/2 + HIGH-1..4 remediated in-PR), **FED-M3-08** (outbound mTLS client, PR #508). With M3-01/03/08 merged, three cards became dependency-clear and were dispatched to the idle coder lane: **FED-M3-04** scope.service → coder0 (`feat/federation-m3-scope-service`); **FED-M3-09** query-source + **FED-M3-07** capabilities verb → coder1 (`feat/federation-m3-query-source` first). Reviewer warmed for the M3 trust-boundary PRs. Remaining blocked-by-DAG: M3-05/06 (await M3-04), M3-10 (await M3-05/06), M3-11 (await M3-09), M3-12→14 (tail). Deploy chain (DEPLOY-IMG-FIX → 03/04) still independent of M3 code — harness local docker-compose fallback covers M3-11.
+
+**Backlog sync #2 — 2026-06-24 (orchestrator):** **FED-M3-09** (query-source) merged via PR #673 and **FED-M3-07** (capabilities) merged via PR #674 — both squash-merged on independent agent review-of-record + green CI (formal Gitea approve unavailable under the shared service account; merge is not gated by the self-approve guard). **FED-M3-05** (list verb) dispatched to coder1 (based on the M3-04 branch, rebase onto main once #672 lands). **FED-M3-04** (scope.service, PR #672) is in review-changes (one include_personal no-leak test outstanding). **DAG fix:** corrected `FED-M3-11` depends_on from `M3-02, M3-09` → `M3-02, M3-04, M3-05, M3-06, M3-09` — the E2E acceptance cases (#1–#5, #8–#10) exercise list/get over mTLS, so the server verbs + scope service are hard prerequisites; the original edge set omitted them and caused a premature M3-11 dispatch. Note: M3 read-path invariant for M3-11 is **no-persist + existing enrollment audit only** — read-verb audit-log writes are deferred to M4 (see M3-05/06 notes), so M3-11 must not assert read-audit-log entries.
+
 ## Milestone 4 — search + audit + rate limit (FED-M4)
 
 _Deferred. Issue #463._

docs/fleet/FLEET-LAUNCH.md

@@ -0,0 +1,114 @@
+# Fleet Launch Runbook
+
+How every Mosaic fleet agent — workers **and** the orchestrator — is launched, and how to
+configure each one. The guiding principle: **one roster-driven launcher**. There is no bespoke
+per-agent launch script; the roster plus per-agent `.env` files are the single source of launch
+config.
+
+## The launch chain
+
+| Layer            | File                                              | Responsibility                                                                                                                                              |
+| ---------------- | ------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| systemd unit     | `mosaic-agent@<role>.service`                     | One templated unit per role; `ExecStart` runs the session launcher with the instance name `%i`. Defaults `MOSAIC_AGENT_RUNTIME=pi`, `MOSAIC_AGENT_NAME=%i`. |
+| session launcher | `tools/fleet/start-agent-session.sh <role>`       | Builds the launch command, opens the tmux pane, wires the heartbeat.                                                                                        |
+| launch command   | `mosaic yolo <runtime>` (or a per-agent override) | Replaces the pane's foreground process with the runtime, fully seeded.                                                                                      |
+| seeding          | `mosaic`'s `composeContract()`                    | Injects the Constitution/USER/TOOLS/runtime contract, `*.local` overlays, **and** the Fleet-Comms cheat-sheet — all via `--append-system-prompt`.           |
+
+Per-agent overrides live in `fleet/agents/<role>.env`, generated from `roster.yaml` by
+`generateAgentEnv` (`packages/mosaic/src/commands/fleet.ts`) and consumed by the launcher.
+
+## Worker launch path (default)
+
+1. `roster.yaml` carries each agent's `runtime` and optional `model_hint`.
+2. `generateAgentEnv` emits `fleet/agents/<role>.env` with `MOSAIC_AGENT_NAME`,
+   `MOSAIC_AGENT_RUNTIME`, and `MOSAIC_AGENT_MODEL`.
+3. `start-agent-session.sh` has no `MOSAIC_AGENT_COMMAND` set, so it falls through to the default
+   (line ~44):
+   ```sh
+   MOSAIC_AGENT_COMMAND="mosaic yolo $MOSAIC_AGENT_RUNTIME${MOSAIC_AGENT_MODEL:+ --model $MOSAIC_AGENT_MODEL}"
+   ```
+4. The launcher bakes `MOSAIC_AGENT_NAME` into the pane command (line ~118), so `composeContract`
+   can inject the Fleet-Comms cheat-sheet for that role.
+
+That is the whole worker path: roster → `.env` → `mosaic yolo <runtime>` → seeded pane.
+
+## Orchestrator fold (PATH A — ships today)
+
+The orchestrator is **just another roster agent** launched through the canonical path — not a
+snowflake script.
+
+| Piece              | Value                               |
+| ------------------ | ----------------------------------- |
+| host-side launcher | `orchestrator-launch.sh`            |
+| systemd unit       | `mosaic-fleet-orchestrator.service` |
+| tmux session       | `orchestrator` (role-named)         |
+
+Set its launch command via `fleet/agents/orchestrator.env`:
+
+```sh
+MOSAIC_AGENT_COMMAND='mosaic yolo claude --channels plugin:discord@<channel>'
+```
+
+When `MOSAIC_AGENT_COMMAND` is set, `start-agent-session.sh`'s `if [ -z "$MOSAIC_AGENT_COMMAND" ]`
+guard (line ~41) is false, so the line-44 default — **including its hardcoded `yolo`** — is skipped
+entirely. The override fully controls the runtime and flags. Routing through `mosaic yolo claude`
+(rather than a raw `claude` invocation) is what gives the orchestrator the same full
+`composeContract` seeding + Fleet-Comms cheat-sheet as every worker, with `--channels` and any
+other flags passed straight through to the `claude` binary.
+
+## Launch gotchas
+
+1. **Flag conflict.** `mosaic yolo claude` already injects `--dangerously-skip-permissions`. Do
+   **not** also pass `--permission-mode bypassPermissions` — the `claude` binary would receive both.
+   Use `mosaic yolo claude …` alone (yolo covers the unattended posture), **or** non-yolo
+   `mosaic claude --permission-mode bypassPermissions …`. Never mix the two.
+2. **`MOSAIC_AGENT_NAME` must reach the pane.** The launcher bakes it from the instance name, and
+   `composeContract` gates the Fleet-Comms block on it (`launch.ts`, in `composeContract`) — **and**
+   the role must be a member of `roster.yaml`, or the block resolves empty.
+3. **`launchRuntime` guards.** `mosaic yolo claude` runs `checkSoul` / `checkRuntime` /
+   `checkSequentialThinking`. The host needs `SOUL.md` and the sequential-thinking MCP, or the
+   launch aborts (a raw `claude` invocation skipped these checks). Dry-run the composed command in a
+   throwaway tmux session before swapping a live launcher.
+
+## Why per-agent `.env` survives upgrades (#632)
+
+`install.sh` `PRESERVE_PATHS` includes `fleet/*.yaml`, `fleet/agents`, and `fleet/run`, so
+`mosaic update`'s framework re-seed **preserves** your roster and per-agent `.env` overrides
+(glob-aware `cp` fallback; matching TS parity in `file-adapter.ts`). Before #632, an auto re-seed
+could wipe them — which is exactly why PATH A's `.env` override is safe to rely on now.
+
+## Inspecting the comms wiring
+
+- `mosaic fleet comms-block <role>` prints the Fleet-Comms cheat-sheet a given role receives at
+  launch — its `[host:session]` identity, the exact `agent-send.sh` command for each peer, and the
+  FLIP / `--verify` conventions. `--host <h>` previews a cross-host view. An unknown role or missing
+  roster **fails loud** (stderr + non-zero exit), so a typo is never a silent no-op.
+- Versus `mosaic compose-contract <runtime>`: that emits the **whole** system prompt and reads the
+  role from `MOSAIC_AGENT_NAME` (a full-prompt smoke test). `comms-block` is the targeted,
+  explicit-arg, comms-only view — e.g. `mosaic fleet comms-block coder0-0` to preview a peer.
+
+## North Star / future direction
+
+**Vision:** a webUI lets the user edit each agent's launch config — switch **harness**
+(claude / pi / codex / opencode), toggle **yolo**, pick a **model**, set a **command/channels**
+override — with no terminal.
+
+**Continuity — this is not a new launch path.** It is a data-model + UI-binding layer over the
+existing roster-driven launcher. Field-by-field status today:
+
+| Launch-config field      | Roster-native today? | Mechanism / gap                                                                                                                                          |
+| ------------------------ | -------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **harness** (`runtime`)  | ✅ end-to-end        | `roster.runtime` → `generateAgentEnv` emits `MOSAIC_AGENT_RUNTIME` → launcher line 44. UI just writes the field.                                         |
+| **model** (`model_hint`) | ✅ end-to-end        | `roster.model_hint` → `MOSAIC_AGENT_MODEL` → launcher line 44 `--model`. UI just writes the field.                                                       |
+| **yolo**                 | ❌ new               | Launcher line 44 **hardcodes** `mosaic yolo`. A non-yolo toggle needs a roster `yolo` field → emit `MOSAIC_AGENT_YOLO` → make line 44 conditional.       |
+| **command / channels**   | ❌ new               | `MOSAIC_AGENT_COMMAND` is **consumed** (launcher line ~12) but `generateAgentEnv` does not emit it. Needs a roster `command`/`channels` field → emitted. |
+
+**The arc:**
+
+- **A** — `.env` `MOSAIC_AGENT_COMMAND` hatch: manual, ships now, kept safe across upgrades by #632.
+- **B** — roster-native launch-config: harness + model are already there; add the **yolo** toggle
+  (line-44 conditional) and **command/channels** emission to complete the data model.
+- **webUI** — binds dropdowns/toggles directly to those four roster fields.
+
+PATH A's `.env` override is the **manual form** of exactly what PATH B makes roster-native and the
+webUI edits — one continuous arc, not three separate features. PATH B is tracked as #636.

docs/fleet/NORTH_STAR.md

@@ -0,0 +1,79 @@
+# Mosaic Fleet — NORTH STAR
+
+> **Generated file — do not edit by hand.**
+> Projected deterministically from [`NORTH_STAR.yaml`](./NORTH_STAR.yaml) by the pure
+> generator in `packages/mosaic/src/commands/fleet.ts` (`renderNorthStarMarkdown`).
+> Edit the YAML, then regenerate. Self-contained Mosaic — no Hermes dependency.
+
+## Mission
+
+A self-driving Mosaic system that 24/7 unattended converts a machine-readable goal set into merged, CI-green, budget-bounded change — looping plan→backlog→assign→execute→verify→merge→reassess — on Mosaic's OWN native backlog/dispatch engine. Mosaic is general-purpose: the user declares the system type they want (software delivery, personal assistant, research, business/operations, …) and the orchestrator provisions the matching persona roster and structure; the delivery fleet is one profile among many.
+
+## Substrate
+
+The Mosaic Backlog is the backlog of record + dispatch engine, built on Mosaic's native Postgres storage service (@mosaicstack/db drizzle; PGlite-embedded by default, full Postgres by config). NOT Hermes.
+
+## Standing objectives
+
+- **NS-1** — Single machine-readable source (this file) drives planning; prose docs are projections.
+- **NS-2** — Every backlog item is an independently-shippable unit with stable id, priority, depends_on DAG, represented as a Mosaic Backlog card; spend tracked as advisory projection.
+- **NS-3** — The supervisor guarantees movement: no idle agent while ready dependency-satisfied work exists; no empty backlog without a replan request; assignment via Mosaic native dispatch/claim.
+- **NS-4** — Exactly one merge-gate approver; nothing reaches main except via pr-merge.sh after pr-ci-wait.sh success; Gitea branch protection is the backstop.
+- **NS-5** — Every unit bounded by wall-clock TTL on its claim; token caps enforced only where a real meter exists, else advisory.
+- **NS-6** — Context cleared between tasks for ephemeral runners (reset_between_tasks); persona+mission re-injected per task.
+- **NS-7** — Meta-loop (session-review + enhancer) continuously proposes small fleet-improvement PRs.
+- **NS-8** — Single operator-flippable PAUSE kill-switch (fleet/run/PAUSED) honored before every dispatch and every merge.
+- **NS-9** — Mosaic is a general-purpose multi-agent system: the user declares the SYSTEM TYPE to run (e.g. software delivery, personal assistant, research, business/operations) and the orchestrator provisions the matching persona roster and org structure from a cross-domain baseline persona library; the delivery/coding fleet is one profile among many.
+
+## Success criteria
+
+- **AC-NS-1** — The supervisor keeps a two-agent floor (1 orchestrator + >=1 enhancer) healthy across reboot.
+- **AC-NS-2** — A goal added to this YAML is decomposed to cards and either merged or escalated, with no human in the loop.
+- **AC-NS-3** — No PR merges with failure/error/no-status/timeout CI, and none bypass pr-merge.sh.
+- **AC-NS-4** — TTL is enforced on claims; token caps remain advisory until a real meter exists.
+- **AC-NS-5** — Flipping fleet/run/PAUSED halts dispatch and merges within one tick.
+- **AC-NS-6** — A user can declare a system type and the fleet provisions the matching persona roster + topology from the baseline library, with no code change.
+- **AC-NS-7** — A user-customized persona (edited or added via the orchestrator) survives `mosaic update`: baseline reseed never clobbers user overrides.
+
+## Workstreams
+
+| id  | title                                                                                                       |
+| --- | ----------------------------------------------------------------------------------------------------------- |
+| A   | Substrate — Mosaic Backlog on native Postgres storage service                                               |
+| B   | Supervisor — movement guarantee, two-agent floor, dispatch/claim                                            |
+| C   | Planner — goal decomposition into independently-shippable cards                                             |
+| D   | Merge-gate — single approver, pr-merge.sh after CI wait                                                     |
+| E   | Meta-loop — session-review + enhancer improvement PRs                                                       |
+| F   | Safety-rails — TTL claims, advisory spend, PAUSE kill-switch                                                |
+| H   | Personas & system profiles — cross-domain library, system-type provisioning, update-surviving customization |
+
+## Goals (backlog projection)
+
+| id  | title                                                                                                                                             | phase | priority    | depends_on |
+| --- | ------------------------------------------------------------------------------------------------------------------------------------------------- | ----- | ----------- | ---------- |
+| A1  | Machine-readable NORTH_STAR.yaml + Markdown projection                                                                                            | 1     | must-have   | —          |
+| A2  | Mosaic Backlog schema + storage-service card store (drizzle/PGlite)                                                                               | 1     | must-have   | A1         |
+| A3a | Card lifecycle — create/claim/release with stable ids + depends_on DAG                                                                            | 1     | must-have   | A2         |
+| A3b | TTL-bounded claim enforcement (wall-clock) on cards                                                                                               | 1     | must-have   | A3a        |
+| A4  | Advisory spend projection per card (degrades to TTL, no real meter)                                                                               | 1     | should-have | A3a        |
+| B1  | Supervisor tick — readiness scan, two-agent-floor health check                                                                                    | 2     | must-have   | A3a        |
+| B2  | Native dispatch/claim — assign ready dependency-satisfied work                                                                                    | 2     | must-have   | A3b, B1    |
+| B3a | Planner decompose — goal added to YAML → cards                                                                                                    | 2     | must-have   | A2, B1     |
+| B3b | Replan request on empty backlog; escalate on no-decompose                                                                                         | 2     | should-have | B3a        |
+| G1  | PAUSE kill-switch + merge-gate honored before dispatch and merge                                                                                  | 2     | must-have   | B2         |
+| H1  | Cross-domain baseline persona library (exec, marketing, ops, research, assistant + engineering roles)                                             | 1     | must-have   | A1         |
+| H2  | System-type profiles — declarative mapping of system type to persona roster + topology                                                            | 2     | must-have   | H1         |
+| H3  | System-type provisioning — user declares type; orchestrator instantiates the matching roster + structure                                          | 2     | must-have   | H2         |
+| H4  | Update-surviving persona customization — ad-hoc edits/additions persisted in a PRESERVE-protected override layer (baseline merged with overrides) | 2     | must-have   | H1         |
+
+## Assumptions (vetoable)
+
+- **ASM-1** (vetoable) — The Mosaic Backlog on the native Postgres storage service is the backlog of record.
+- **ASM-2** (vetoable) — Claude gate roles have no native busy status, so readiness = pane-idle + heartbeat.
+- **ASM-3** (vetoable) — Two-agent floor = 1 orchestrator + >=1 enhancer.
+- **ASM-4** (vetoable) — Baseline personas ship in framework/fleet/roles/ (reseeded on update); user overrides live in a separate PRESERVE_PATHS-protected layer and win on merge.
+
+## Spend
+
+- **advisory:** true
+- No per-task token meter yet; budgets degrade to TTL. Spend is tracked only as an advisory projection alongside each card.

docs/fleet/NORTH_STAR.yaml

@@ -0,0 +1,215 @@
+# Mosaic Fleet — NORTH_STAR (machine-readable source of truth)
+#
+# This file is the single machine-readable source of truth for fleet planning.
+# Prose docs (including NORTH_STAR.md) are deterministic PROJECTIONS of this file.
+# Regenerate the Markdown projection with the pure generator in
+# packages/mosaic/src/commands/fleet.ts (renderNorthStarMarkdown). Edit the YAML,
+# never the .md.
+#
+# Self-contained Mosaic. NO Hermes runtime dependency. The backlog of record is
+# the Mosaic Backlog on Mosaic's OWN native Postgres storage service.
+
+version: 1
+
+mission: >-
+  A self-driving Mosaic system that 24/7 unattended converts a machine-readable
+  goal set into merged, CI-green, budget-bounded change — looping
+  plan→backlog→assign→execute→verify→merge→reassess — on Mosaic's OWN native
+  backlog/dispatch engine. Mosaic is general-purpose: the user declares the
+  system type they want (software delivery, personal assistant, research,
+  business/operations, …) and the orchestrator provisions the matching persona
+  roster and structure; the delivery fleet is one profile among many.
+
+substrate:
+  note: >-
+    The Mosaic Backlog is the backlog of record + dispatch engine, built on
+    Mosaic's native Postgres storage service (@mosaicstack/db drizzle;
+    PGlite-embedded by default, full Postgres by config). NOT Hermes.
+
+standing_objectives:
+  - id: NS-1
+    text: >-
+      Single machine-readable source (this file) drives planning; prose docs are
+      projections.
+  - id: NS-2
+    text: >-
+      Every backlog item is an independently-shippable unit with stable id,
+      priority, depends_on DAG, represented as a Mosaic Backlog card; spend
+      tracked as advisory projection.
+  - id: NS-3
+    text: >-
+      The supervisor guarantees movement: no idle agent while ready
+      dependency-satisfied work exists; no empty backlog without a replan
+      request; assignment via Mosaic native dispatch/claim.
+  - id: NS-4
+    text: >-
+      Exactly one merge-gate approver; nothing reaches main except via
+      pr-merge.sh after pr-ci-wait.sh success; Gitea branch protection is the
+      backstop.
+  - id: NS-5
+    text: >-
+      Every unit bounded by wall-clock TTL on its claim; token caps enforced
+      only where a real meter exists, else advisory.
+  - id: NS-6
+    text: >-
+      Context cleared between tasks for ephemeral runners
+      (reset_between_tasks); persona+mission re-injected per task.
+  - id: NS-7
+    text: >-
+      Meta-loop (session-review + enhancer) continuously proposes small
+      fleet-improvement PRs.
+  - id: NS-8
+    text: >-
+      Single operator-flippable PAUSE kill-switch (fleet/run/PAUSED) honored
+      before every dispatch and every merge.
+  - id: NS-9
+    text: >-
+      Mosaic is a general-purpose multi-agent system: the user declares the
+      SYSTEM TYPE to run (e.g. software delivery, personal assistant, research,
+      business/operations) and the orchestrator provisions the matching persona
+      roster and org structure from a cross-domain baseline persona library; the
+      delivery/coding fleet is one profile among many.
+
+success_criteria:
+  - id: AC-NS-1
+    text: >-
+      The supervisor keeps a two-agent floor (1 orchestrator + >=1 enhancer)
+      healthy across reboot.
+  - id: AC-NS-2
+    text: >-
+      A goal added to this YAML is decomposed to cards and either merged or
+      escalated, with no human in the loop.
+  - id: AC-NS-3
+    text: >-
+      No PR merges with failure/error/no-status/timeout CI, and none bypass
+      pr-merge.sh.
+  - id: AC-NS-4
+    text: >-
+      TTL is enforced on claims; token caps remain advisory until a real meter
+      exists.
+  - id: AC-NS-5
+    text: >-
+      Flipping fleet/run/PAUSED halts dispatch and merges within one tick.
+  - id: AC-NS-6
+    text: >-
+      A user can declare a system type and the fleet provisions the matching
+      persona roster + topology from the baseline library, with no code change.
+  - id: AC-NS-7
+    text: >-
+      A user-customized persona (edited or added via the orchestrator) survives
+      `mosaic update`: baseline reseed never clobbers user overrides.
+
+workstreams:
+  - id: A
+    title: Substrate — Mosaic Backlog on native Postgres storage service
+  - id: B
+    title: Supervisor — movement guarantee, two-agent floor, dispatch/claim
+  - id: C
+    title: Planner — goal decomposition into independently-shippable cards
+  - id: D
+    title: Merge-gate — single approver, pr-merge.sh after CI wait
+  - id: E
+    title: Meta-loop — session-review + enhancer improvement PRs
+  - id: F
+    title: Safety-rails — TTL claims, advisory spend, PAUSE kill-switch
+  - id: H
+    title: Personas & system profiles — cross-domain library, system-type provisioning, update-surviving customization
+
+goals:
+  - id: A1
+    title: Machine-readable NORTH_STAR.yaml + Markdown projection
+    phase: 1
+    priority: must-have
+    depends_on: []
+  - id: A2
+    title: Mosaic Backlog schema + storage-service card store (drizzle/PGlite)
+    phase: 1
+    priority: must-have
+    depends_on: [A1]
+  - id: A3a
+    title: Card lifecycle — create/claim/release with stable ids + depends_on DAG
+    phase: 1
+    priority: must-have
+    depends_on: [A2]
+  - id: A3b
+    title: TTL-bounded claim enforcement (wall-clock) on cards
+    phase: 1
+    priority: must-have
+    depends_on: [A3a]
+  - id: A4
+    title: Advisory spend projection per card (degrades to TTL, no real meter)
+    phase: 1
+    priority: should-have
+    depends_on: [A3a]
+  - id: B1
+    title: Supervisor tick — readiness scan, two-agent-floor health check
+    phase: 2
+    priority: must-have
+    depends_on: [A3a]
+  - id: B2
+    title: Native dispatch/claim — assign ready dependency-satisfied work
+    phase: 2
+    priority: must-have
+    depends_on: [A3b, B1]
+  - id: B3a
+    title: Planner decompose — goal added to YAML → cards
+    phase: 2
+    priority: must-have
+    depends_on: [A2, B1]
+  - id: B3b
+    title: Replan request on empty backlog; escalate on no-decompose
+    phase: 2
+    priority: should-have
+    depends_on: [B3a]
+  - id: G1
+    title: PAUSE kill-switch + merge-gate honored before dispatch and merge
+    phase: 2
+    priority: must-have
+    depends_on: [B2]
+  - id: H1
+    title: Cross-domain baseline persona library (exec, marketing, ops, research, assistant + engineering roles)
+    phase: 1
+    priority: must-have
+    depends_on: [A1]
+  - id: H2
+    title: System-type profiles — declarative mapping of system type to persona roster + topology
+    phase: 2
+    priority: must-have
+    depends_on: [H1]
+  - id: H3
+    title: System-type provisioning — user declares type; orchestrator instantiates the matching roster + structure
+    phase: 2
+    priority: must-have
+    depends_on: [H2]
+  - id: H4
+    title: Update-surviving persona customization — ad-hoc edits/additions persisted in a PRESERVE-protected override layer (baseline merged with overrides)
+    phase: 2
+    priority: must-have
+    depends_on: [H1]
+
+assumptions:
+  - id: ASM-1
+    vetoable: true
+    text: >-
+      The Mosaic Backlog on the native Postgres storage service is the backlog
+      of record.
+  - id: ASM-2
+    vetoable: true
+    text: >-
+      Claude gate roles have no native busy status, so readiness = pane-idle +
+      heartbeat.
+  - id: ASM-3
+    vetoable: true
+    text: 'Two-agent floor = 1 orchestrator + >=1 enhancer.'
+  - id: ASM-4
+    vetoable: true
+    text: >-
+      Baseline personas ship in framework/fleet/roles/ (reseeded on update);
+      user overrides live in a separate PRESERVE_PATHS-protected layer and win
+      on merge.
+
+spend:
+  advisory: true
+  note: >-
+    No per-task token meter yet; budgets degrade to TTL. Spend is tracked only
+    as an advisory projection alongside each card.

docs/fleet/PRD.md

@@ -7,10 +7,10 @@
 
 ## Problem
 
-The durable tmux fleet runs on the isolated `mosaic-factory` socket. That isolation
+The durable tmux fleet runs on the isolated `mosaic-fleet` socket. That isolation
 (which protects the operator's default tmux) makes the fleet **invisible** to default
 tooling, and truth is split across three planes no single command joins — systemd
-(`systemctl --user`), tmux (`-L mosaic-factory`), and the process tree (`pstree`).
+(`systemctl --user`), tmux (`-L mosaic-fleet`), and the process tree (`pstree`).
 `agent tail` (`capture-pane`) returns **blank for full-screen TUIs**, and `agent send`
 confirms only keystroke injection, not acceptance. Net: the operator has near-zero
 observability and no safe way to watch a session.
@@ -56,7 +56,7 @@ observability and no safe way to watch a session.
 
 ## Acceptance criteria
 
-- `mosaic fleet ps` shows all 5 live sessions on `mosaic-factory` with correct
+- `mosaic fleet ps` shows all 5 live sessions on `mosaic-fleet` with correct
   pane/pid/idle and flags the dogfood **drift** (`canary-pi` runtime=pi but pane runs
   `dogfood-agent.py`) and the **boot-enable** gap (active but disabled).
 - Killing one agent's pane flips its row to dead/stale within one `interval`.
@@ -72,7 +72,7 @@ observability and no safe way to watch a session.
 - Unit/CLI specs in `packages/mosaic/src/commands/fleet.spec.ts` (and a new
   `fleet-ps`/`watch`/`send-verify` spec) using the injected `CommandRunner` to assert
   exact tmux/systemd command construction and JSON shape (tenant+host present).
-- Situational: run against the live `mosaic-factory` fleet; capture `fleet ps` output,
+- Situational: run against the live `mosaic-fleet` fleet; capture `fleet ps` output,
   a kill-and-detect cycle, a read-only `watch`, and a `send --verify` pass/fail pair.
 
 ## Known limitations

docs/fleet/TASKS.md

@@ -7,18 +7,18 @@
 > Mission: `mvp-20260312` · PRD: [docs/fleet/PRD.md](./PRD.md) · North star: [docs/fleet/north-star.md](./north-star.md)
 > Status: `not-started` | `in-progress` | `done` | `blocked` | `failed`
 
-| id            | status      | description                                                                                                        | depends_on            | agent       | pr  | notes                                                                                                                         |
-| ------------- | ----------- | ------------------------------------------------------------------------------------------------------------------ | --------------------- | ----------- | --- | ----------------------------------------------------------------------------------------------------------------------------- |
-| FLEET-OBS-000 | done        | Plan: north-star + Phase-2 PRD + workstream scaffolding                                                            | —                     | lead        | —   | persisted 2026-06-20 on `feat/fleet-observability`                                                                            |
-| FLEET-OBS-001 | done        | Heartbeat protocol v1 spec finalized in PRD + framework doc                                                        | FLEET-OBS-000         | lead        | —   | file-based `~/.config/mosaic/fleet/run/<agent>.hb`; spec in PRD                                                               |
-| FLEET-OBS-002 | in-progress | Implement heartbeat responder in `dogfood-agent.py`                                                                | FLEET-OBS-001         | fleet-coder | —   | dispatched to ad-hoc `mosaic yolo` fleet agent (dogfood)                                                                      |
-| FLEET-OBS-003 | done        | `mosaic fleet ps` — join systemd+tmux+proc+idle+heartbeat; tenant+host tagged; drift + boot-enable flags; `--json` | FLEET-OBS-001         | worker      | —   | commit ab47831; LIVE-verified on mosaic-factory; caught canary-pi DRIFT + BOOT-ENABLE. Polish: idleSeconds parse returns null |
-| FLEET-OBS-004 | done        | `mosaic agent watch <name>` — read-only join (no resize, no keystrokes)                                            | FLEET-OBS-000         | worker      | —   | `attach -r`; verb wired                                                                                                       |
-| FLEET-OBS-005 | done        | `mosaic agent send --verify` — delivery/acceptance receipt                                                         | FLEET-OBS-000         | worker      | —   | --verify flag; draft-heuristic verify                                                                                         |
-| FLEET-OBS-006 | done        | CLI specs for ps/watch/send-verify (tenant+host shape, command construction)                                       | FLEET-OBS-003,004,005 | worker      | —   | 62 tests green (31 new); re-verified by lead                                                                                  |
-| FLEET-OBS-007 | not-started | Framework doc: fleet observability guide + verbs                                                                   | FLEET-OBS-003,004,005 | lead        | —   | `docs/guides/` or `framework/tools/.../README`                                                                                |
-| FLEET-OBS-008 | not-started | Independent review + dogfood verification on live fleet                                                            | FLEET-OBS-002..007    | reviewer    | —   | author ≠ reviewer; capture evidence in scratchpad                                                                             |
-| FLEET-OBS-009 | not-started | Open PR → green CI (queue guard) → squash-merge → close `fleet-observability-1`                                    | FLEET-OBS-008         | lead        | —   | trunk merge; no direct push to main                                                                                           |
+| id            | status      | description                                                                                                        | depends_on            | agent       | pr  | notes                                                                                                                       |
+| ------------- | ----------- | ------------------------------------------------------------------------------------------------------------------ | --------------------- | ----------- | --- | --------------------------------------------------------------------------------------------------------------------------- |
+| FLEET-OBS-000 | done        | Plan: north-star + Phase-2 PRD + workstream scaffolding                                                            | —                     | lead        | —   | persisted 2026-06-20 on `feat/fleet-observability`                                                                          |
+| FLEET-OBS-001 | done        | Heartbeat protocol v1 spec finalized in PRD + framework doc                                                        | FLEET-OBS-000         | lead        | —   | file-based `~/.config/mosaic/fleet/run/<agent>.hb`; spec in PRD                                                             |
+| FLEET-OBS-002 | in-progress | Implement heartbeat responder in `dogfood-agent.py`                                                                | FLEET-OBS-001         | fleet-coder | —   | dispatched to ad-hoc `mosaic yolo` fleet agent (dogfood)                                                                    |
+| FLEET-OBS-003 | done        | `mosaic fleet ps` — join systemd+tmux+proc+idle+heartbeat; tenant+host tagged; drift + boot-enable flags; `--json` | FLEET-OBS-001         | worker      | —   | commit ab47831; LIVE-verified on mosaic-fleet; caught canary-pi DRIFT + BOOT-ENABLE. Polish: idleSeconds parse returns null |
+| FLEET-OBS-004 | done        | `mosaic agent watch <name>` — read-only join (no resize, no keystrokes)                                            | FLEET-OBS-000         | worker      | —   | `attach -r`; verb wired                                                                                                     |
+| FLEET-OBS-005 | done        | `mosaic agent send --verify` — delivery/acceptance receipt                                                         | FLEET-OBS-000         | worker      | —   | --verify flag; draft-heuristic verify                                                                                       |
+| FLEET-OBS-006 | done        | CLI specs for ps/watch/send-verify (tenant+host shape, command construction)                                       | FLEET-OBS-003,004,005 | worker      | —   | 62 tests green (31 new); re-verified by lead                                                                                |
+| FLEET-OBS-007 | not-started | Framework doc: fleet observability guide + verbs                                                                   | FLEET-OBS-003,004,005 | lead        | —   | `docs/guides/` or `framework/tools/.../README`                                                                              |
+| FLEET-OBS-008 | not-started | Independent review + dogfood verification on live fleet                                                            | FLEET-OBS-002..007    | reviewer    | —   | author ≠ reviewer; capture evidence in scratchpad                                                                           |
+| FLEET-OBS-009 | not-started | Open PR → green CI (queue guard) → squash-merge → close `fleet-observability-1`                                    | FLEET-OBS-008         | lead        | —   | trunk merge; no direct push to main                                                                                         |
 
 ## Proposed MVP rollup row (for the MVP orchestrator — not written by this workstream)
 

docs/fleet/backlog-conventions.md

@@ -0,0 +1,138 @@
+# Fleet Backlog Conventions
+
+The **backlog** is Mosaic's native backlog-of-record for fleet work. It is built
+end-to-end on Mosaic's own storage layer (`@mosaicstack/db`, drizzle/Postgres)
+and surfaced as `mosaic fleet backlog <sub> --json`.
+
+> **Mosaic-native, no Hermes.** This backlog REPLACES the former Hermes adapter.
+> There is **no** runtime dependency on Hermes, `hermes kanban`, or `~/.hermes`
+> anywhere in this feature. Anything previously delegated to Hermes is recreated
+> here on Mosaic's own Postgres storage layer.
+
+## Storage tier — PGlite by default, Postgres by config
+
+The backlog uses the existing Mosaic storage layer; there is **no** new database
+engine (no sqlite, no raw client).
+
+| Condition                      | Tier                 | Data location                    |
+| ------------------------------ | -------------------- | -------------------------------- |
+| `DATABASE_URL` set             | Full server Postgres | the configured database          |
+| `PGLITE_DATA_DIR` set (no URL) | Embedded PGlite      | that directory                   |
+| neither (default)              | Embedded PGlite      | `~/.config/mosaic/fleet/backlog` |
+
+PGlite is real Postgres semantics in-process — including the row locks the atomic
+claim relies on — so the **same code** runs on a laptop (embedded, single-host
+default) and on a full Postgres deployment. Switching tiers is config-only.
+
+The schema (`backlog` table) is created automatically on first CLI use:
+`runMigrations()` for Postgres, `runPgliteMigrations()` for embedded PGlite.
+
+### Update safety
+
+The embedded PGlite store lives under `~/.config/mosaic/fleet/backlog`, which is
+listed in `PRESERVE_PATHS` in `packages/mosaic/framework/install.sh`. This means
+`mosaic update` (which runs the framework sync with `rsync --delete`) will **not**
+wipe the operator's backlog — same protection as the roster, per-agent env, and
+heartbeat run dir.
+
+## Card schema
+
+A card is one row in the `backlog` table:
+
+| Column              | Type                | Notes                                                         |
+| ------------------- | ------------------- | ------------------------------------------------------------- |
+| `id`                | text (PK)           | Stable, caller-supplied id (e.g. `A4`, `fleet-001`).          |
+| `title`             | text                | Required.                                                     |
+| `body`              | text (nullable)     | Free-form description.                                        |
+| `phase`             | text (nullable)     | Board/phase grouping (see below).                             |
+| `priority`          | int (default 0)     | **Higher = sooner.** Claim picks the max-priority ready card. |
+| `status`            | enum                | `ready` \| `claimed` \| `blocked` \| `done`.                  |
+| `depends_on`        | jsonb `string[]`    | DAG edges — ids of cards this one depends on.                 |
+| `claim_owner`       | text (nullable)     | Owner token of the active claim.                              |
+| `claim_ttl_seconds` | int (nullable)      | TTL of the active claim.                                      |
+| `claimed_at`        | timestamptz (null)  | When the claim was taken. `claimed_at + ttl` = expiry.        |
+| `attempts`          | int (default 0)     | Incremented each time the card is claimed.                    |
+| `idempotency_key`   | text (unique, null) | Dedups `create`; NULLs are distinct in Postgres.              |
+| `acceptance`        | jsonb (nullable)    | Acceptance criteria (array of strings or object).             |
+| `created_at`        | timestamptz         |                                                               |
+| `updated_at`        | timestamptz         |                                                               |
+
+`depends_on` is modeled as a `jsonb` array column rather than a separate edge
+table. Justification: it matches the repo's existing style (e.g. `tasks.tags`,
+`agents.skills`, `routing_rules.conditions` are all jsonb arrays), keeps a card
+self-contained, and the DAG is small (per-card dependency lists), so a join table
+would add ceremony without benefit.
+
+### Board / phase convention
+
+`phase` is a free-form grouping string used as the board column / milestone label
+(e.g. `M1`, `fleet`, `infra`). `list --phase <phase>` filters to one board lane.
+`priority` orders cards **within** the ready pool regardless of phase.
+
+## Status lifecycle
+
+```
+            create
+              │
+              ▼
+   ┌──────► ready ───── claim ─────► claimed ───── complete ─────► done
+   │          │                         │
+   │       block                  reclaim (TTL expiry or --id)
+   │          ▼                         │
+   │       blocked                      └──────────────────────────┘ (back to ready)
+   └──────────┘  (reclaim / re-create can return a card to ready)
+```
+
+- **ready** — eligible to be claimed once every `depends_on` card is `done`.
+- **claimed** — a worker holds it; `claim_owner` + `claimed_at` set.
+- **blocked** — explicitly parked; never auto-claimed.
+- **done** — completed; satisfies dependents.
+
+## Atomic claim (`FOR UPDATE SKIP LOCKED`) + TTL
+
+`claim` is atomic. Inside a single transaction it locks candidate `ready` rows
+with `SELECT ... FOR UPDATE SKIP LOCKED` (via the drizzle `sql` operator), picks
+the highest-priority deps-satisfied card, and flips it to `claimed`. Because a row
+already locked by a concurrent claimer is **skipped**, two claimers can **never**
+both win the same card — the loser falls through to the next candidate or gets
+`null`. (Proven by the concurrency tests in `packages/db/src/backlog.spec.ts`.)
+
+- **Deps gate:** a card is only claimable when every id in `depends_on` is `done`.
+- **TTL:** `claim --ttl <sec>` (default **900s**) records `claim_ttl_seconds`.
+- **reclaim:** releases claims whose `claimed_at + ttl` is in the past (expired)
+  back to `ready`, clearing the claim fields. `reclaim --id <id>` force-releases a
+  specific card regardless of expiry. This is how a crashed worker's card returns
+  to the pool.
+
+## CLI — `mosaic fleet backlog <sub> --json`
+
+All subcommands support `--json`.
+
+| Subcommand                                                                                    | Purpose                                                                                   |
+| --------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- |
+| `create --id --title [--body --phase --priority --depends-on --acceptance --idempotency-key]` | Create a card; `idempotency_key` dedups (repeat returns the existing card).               |
+| `list [--status --phase --ready-only]`                                                        | List cards. `--ready-only` = status `ready` AND all deps `done`.                          |
+| `claim --owner [--ttl <sec> --id <id>]`                                                       | Atomically claim the highest-priority ready card (or `--id`). Returns the card or `null`. |
+| `reclaim [--id <id>]`                                                                         | Release expired claims (or a specific card) back to `ready`.                              |
+| `link --from --to`                                                                            | Add a `depends_on` edge (`--from` depends on `--to`).                                     |
+| `stats`                                                                                       | Counts by status, oldest-ready age, expired-claim count.                                  |
+| `block --id`                                                                                  | Set a card to `blocked`.                                                                  |
+| `complete --id`                                                                               | Set a card to `done` (releases any claim).                                                |
+
+### Example
+
+```sh
+# Seed two cards, the second depends on the first.
+mosaic fleet backlog create --id A1 --title "schema" --priority 5
+mosaic fleet backlog create --id A2 --title "service" --depends-on A1 --priority 9
+
+# A2 is gated on A1, so claim returns A1 first.
+mosaic fleet backlog claim --owner worker-1 --ttl 600 --json
+
+# Finish A1; now A2 is ready.
+mosaic fleet backlog complete --id A1
+mosaic fleet backlog list --ready-only --json
+
+# Recover stalled work.
+mosaic fleet backlog reclaim --json
+```

docs/fleet/f4-matrix-connector.md

@@ -0,0 +1,92 @@
+# F4 — Orchestrator chat connector + Matrix (local homeserver)
+
+> **Issue:** #616 · **Doctrine:** `docs/fleet/north-star.md` (#613) — orchestrator-chat-connector decision.
+> **Status:** Phase 1 (abstraction + scaffold) in this PR; Phase 2+ are follow-ups (below).
+
+## Goal
+
+The fleet **orchestrator** is the operator's single point of contact. The north-star makes the
+chat channel a **user-chosen connector** — tmux today, Discord live ("Mos"), with Matrix /
+Telegram / Slack configurable. F4 adds **Matrix** (local homeserver) as a **peer** connector and,
+first, the small **connector abstraction** that makes connectors pluggable without touching fleet
+core.
+
+## The abstraction (Phase 1 — this PR)
+
+Connectors implement one small, uniform interface (`src/fleet/connectors/types.ts`):
+
+```ts
+interface OrchestratorConnector {
+  readonly kind: 'tmux' | 'discord' | 'matrix';
+  send(message: OutboundMessage): Promise<SendResult>; // orchestrator → human
+  subscribe(handler: (m: InboundMessage) => void): Unsubscribe; // human → orchestrator
+  health(): Promise<ConnectorHealth>; // reachable + authenticated
+}
+```
+
+- **send / subscribe / health** — the only surface fleet core depends on. `SendResult` is the
+  ack half; `health()` is the liveness half.
+- **Thread-aware by metadata** — `OutboundMessage.threadId` / `InboundMessage.threadId` are
+  optional, so thread-capable connectors (Matrix rooms/threads, the future first-party Mosaic
+  Discord plugin) fit **without an interface change**.
+- **Registry** (`registry.ts`) — implementations register a factory by kind; `createConnector(config)`
+  resolves one from roster config. Phase 1 ships the registry + `resolveConnectorKind` (defaults
+  `tmux` when a roster declares no connector — **back-compat**); the factories land in Phase 2.
+
+### Config model
+
+A roster may carry an optional `connector` block (`roster.schema.json`); absent ⇒ tmux.
+
+```yaml
+connector:
+  kind: matrix # tmux | discord | matrix
+  matrix:
+    homeserver_url: https://matrix.example.internal
+    user_id: '@mos:example.internal'
+    room_id: '!abc:example.internal'
+```
+
+**Secrets are never in the roster.** `MATRIX_ACCESS_TOKEN` / `DISCORD_BOT_TOKEN` come from the
+environment (the gateway env-config pattern that already masks them). The sanitization gate would
+reject a token committed to a shipped file anyway.
+
+## Matrix connector (Phase 2)
+
+The connector speaks the **Matrix client-server API** directly over HTTPS (`fetch` — no SDK needed
+for MVP), so it is **homeserver-agnostic**:
+
+| Op          | Matrix CS-API                                                            |
+| ----------- | ------------------------------------------------------------------------ |
+| `send`      | `PUT /_matrix/client/v3/rooms/{roomId}/send/m.room.message/{txnId}`      |
+| `subscribe` | `GET /_matrix/client/v3/sync` (long-poll, `since` token) → room timeline |
+| `health`    | `GET /_matrix/client/versions` (reachable) + `…/account/whoami` (authed) |
+| threads     | `m.thread` relations ↔ `threadId`                                        |
+
+## Local homeserver (infra, not connector code)
+
+Strategic default: a **self-hosted** homeserver on our own infra — no third-party gateway.
+
+- **Default: Conduit** (Rust, single binary, low resource) — trivial to stand up for a fleet/dev
+  homeserver.
+- **Alternative: Synapse** (mature, feature-complete) for scale.
+
+The connector only needs `homeserver_url` + `user_id` + `room_id` + an access token, so the
+homeserver choice is a **deployment** concern (a Phase-2 deploy guide), not connector code.
+
+## Phasing
+
+| Phase | Scope                                                                                   | This PR |
+| ----- | --------------------------------------------------------------------------------------- | ------- |
+| **1** | Connector interface + types, registry + kind resolution, roster `connector` schema, doc | ✅ yes  |
+| 2     | Matrix CS-API client (fetch-based send/sync/health) + registered factory + tests        | follow  |
+| 2     | `fleet init` / `configure` connector-selection UX; roster parse wires the block         | follow  |
+| 2     | systemd launch wiring so the orchestrator starts on the chosen connector                | follow  |
+| 3     | Conduit deploy guide; first-party Mosaic Discord (threads) registers as a connector     | follow  |
+
+## Back-compat & boundaries
+
+- Existing rosters (no `connector`) resolve to tmux — **zero change**.
+- Fleet core never branches on connector kind; it depends only on the interface.
+- Cross-host reach rides the **federation** layer (W1), not a bespoke broker (north-star assumption).
+- Phase 1 touches **no** `fleet.ts` core (a self-contained `connectors/` module), so it is
+  independent of the in-flight fleet-config PRs.

docs/fleet/north-star.md

@@ -55,14 +55,22 @@ The Fleet inherits — does not re-invent — the MVP's hard requirements:
 
 One **definition** is the source of truth; the **session** is how it runs.
 
-| Layer                            | Owner                                                                                       | Phase-2 reality                                        | Destination                                             |
-| -------------------------------- | ------------------------------------------------------------------------------------------- | ------------------------------------------------------ | ------------------------------------------------------- |
-| **Definition + identity + auth** | gateway / `mosaic-as` (scoped tokens, #541)                                                 | `roster.yaml` (tenant-tagged)                          | one definition; `mosaic agent --new` materializes it    |
-| **Tenancy boundary**             | **Linux uid per tenant** (linger, own `systemd --user`, own socket, own `~/.config/mosaic`) | one tenant: `jarvis` = tenant zero                     | uid-per-tenant; federation aggregates across hosts      |
-| **Runtime**                      | per-tenant tmux session on isolated socket                                                  | dogfood stub sessions (live now on `mosaic-factory`)   | claude/codex/pi/opencode TUIs                           |
-| **Liveness**                     | **heartbeat protocol** every runtime answers                                                | protocol defined + dogfood stub answers it             | all runtimes answer; "healthy" ≠ "pane alive"           |
-| **Observation**                  | read-only `watch` (native tmux) + `pipe-pane` stream                                        | CLI `watch`/`ps`; explicit opt-in `attach` for control | + auth-gated webUI streams                              |
-| **Control plane**                | **federation** across hosts × tenants                                                       | records already carry `tenant_id` + `host`             | federated gateways expose fleet state; webUI in Phase 5 |
+| Layer                            | Owner                                                                                       | Phase-2 reality                                        | Destination                                                                                                                         |
+| -------------------------------- | ------------------------------------------------------------------------------------------- | ------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------- |
+| **Definition + identity + auth** | gateway / `mosaic-as` (scoped tokens, #541)                                                 | `roster.yaml` (tenant-tagged)                          | one definition; `mosaic agent --new` materializes it                                                                                |
+| **Tenancy boundary**             | **Linux uid per tenant** (linger, own `systemd --user`, own socket, own `~/.config/mosaic`) | one tenant: `jarvis` = tenant zero                     | uid-per-tenant; federation aggregates across hosts                                                                                  |
+| **Runtime**                      | per-tenant tmux session on isolated socket                                                  | dogfood stub sessions (live now on `mosaic-factory`)   | claude/codex/pi/opencode TUIs                                                                                                       |
+| **Liveness**                     | **heartbeat protocol** every runtime answers                                                | protocol defined + dogfood stub answers it             | all runtimes answer; "healthy" ≠ "pane alive"                                                                                       |
+| **Observation**                  | read-only `watch` (native tmux) + `pipe-pane` stream                                        | CLI `watch`/`ps`; explicit opt-in `attach` for control | + auth-gated webUI streams                                                                                                          |
+| **Control plane**                | **federation** across hosts × tenants                                                       | records already carry `tenant_id` + `host`             | federated gateways expose fleet state; webUI in Phase 5                                                                             |
+| **Central register**             | Postgres `fleet` schema (gateway instance); access via gateway API only                     | _none in PoC_ (files + `roster.yaml`)                  | agents, missions, tasks, heartbeats, spend — single network-accessible SSOT; docs = generated projections                           |
+| **Budget / spend governance**    | **per-tenant budget policy** ingested by the orchestrator + routing layer                   | none today (spend is unmetered)                        | usage-vs-limit feedback ingested; spend auto-paced to the limit window; per-provider/per-account/concurrency/API-$ budgets enforced |
+
+> **PoC socket hygiene:** the PoC fleet runs on the **default tmux socket** (no `-L`).
+> The named production-isolation socket is **`mosaic-fleet`** (matches the product brand);
+> an absent roster `socket_name` means the default socket everywhere (spawn, `fleet ps`,
+> onboarding cheat-sheet). The legacy dogfood canary still runs on the old `mosaic-factory`
+> socket pending migration.
 
 ## Operating model (inherited, not reinvented)
 
@@ -73,6 +81,37 @@ diff-sanity → squash-merge → verify), **decide-and-inform** cadence, and a d
 this model. See `mosaicstack-aiguide` whitepapers 01 (inter-agent comms) and 03
 (orchestration model) for the rationale.
 
+## Fleet roster — the two-agent floor and the role library
+
+A fleet is **never a single agent**. The minimum viable fleet is **two**:
+
+| Role             | Mandate                                                                                                                                                                                                                                                                                                                                                                 | Boundaries                                                                                 |
+| ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------ |
+| **Orchestrator** | The user's **single point of contact**. Owns the general flow, keeps agentic actions on-target, and **adds/removes agents from the fleet at will** to meet goals and user needs. Exactly **one** per fleet (the existing R5 invariant).                                                                                                                                 | Delegates source work; never the sole worker.                                              |
+| **Enhancer**     | The fleet's **continuous-improvement loop**. Monitors fleet activity, analyzes for enhancements/optimizations, builds a **plan of remediation**, and — **with the orchestrator** — upgrades fleet capability: tool creation/repair, skills, harness improvements, and **bug reports filed to Mosaic Stack** for proper remediation. Recommends which agents are needed. | **Does not code, review code, or perform delivery tasks.** Improvement and diagnosis only. |
+
+> **Why two, not one:** the orchestrator drives delivery; the enhancer makes the fleet
+> _get better at delivering_ over time. The enhancer is how the fleet self-heals its tools,
+> skills, and harnesses, and how real defects flow back to Mosaic Stack as bug reports.
+> Together they are the irreducible core — every other role is added on demand.
+
+A **general** fleet starts at this floor: the orchestrator (advised by the enhancer)
+materializes whatever roles prove necessary over the mission's life. Specialized presets
+(coding, research, etc.) seed additional roles up front, but all reduce to the same two-agent
+spine plus an on-demand **role library**:
+
+| Role profile        | Purpose                                                                           |
+| ------------------- | --------------------------------------------------------------------------------- |
+| **orchestrator**    | point of contact, flow control, fleet composition (1 per fleet)                   |
+| **enhancer**        | fleet monitoring, optimization, tool/skill/harness upgrades, upstream bug reports |
+| **coder**           | implementation (worker; stops at PR-open)                                         |
+| **code review**     | independent code review gate                                                      |
+| **security review** | security/auth/secret review gate                                                  |
+| **research**        | investigation, synthesis, options analysis                                        |
+| **board**           | deliberation panel — moonshot, contrarian, technical, business, financial lenses  |
+| **operations**      | infra, deploy, health, incident response                                          |
+| _…extensible_       | new profiles added as missions demand (orchestrator + enhancer decide)            |
+
 ## Invariants — "maximal vision, incremental delivery, zero foreclosure"
 
 Every artifact, starting Phase 2, MUST:
@@ -82,6 +121,67 @@ Every artifact, starting Phase 2, MUST:
 3. Define **healthy = answered a heartbeat within N seconds**, never just "pane alive".
 4. Make **observation read-only by default**; control is an explicit, separate, opt-in verb.
 
+> **OPS INVARIANT — runtime agents need a real TTY.** Claude/Codex/pi/opencode agents
+> cannot be bare-launched from a systemd `ExecStart`; a durable harness with a real PTY is
+> required. This is **why `start-agent-session.sh` launches into tmux** and uses a
+> `MOSAIC_AGENT_COMMAND` override rather than running the runtime directly under systemd.
+
+## Budget & token governance (first-class fleet concern)
+
+Spend is a fleet-level resource, not a per-agent afterthought. The fleet treats token
+and API-dollar budget the way it treats liveness: a signal every runtime exposes and the
+control plane is accountable for. This rides the same primitives as everything else —
+`tenant_id` + `host` on every spend record, **read-only metering by default**, and the
+**federation** layer as the cross-host aggregation point (W1) — so budgeting is zero-foreclosure
+from day one even while one tenant exists.
+
+**Two spend regimes, one policy surface:**
+
+| Regime                                                  | Feedback signal                                                          | Fleet obligation                                                                                        |
+| ------------------------------------------------------- | ------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------- |
+| **OAuth-subscription runtimes** (Claude sub, Codex sub) | runtime exposes **current-usage-vs-limit** within a rolling limit window | **ingest** the signal per sub-account; **auto-pace** agentic spend so the window is not exhausted early |
+| **API-token runtimes** (metered per token)              | provider billing / token counts                                          | enforce **hard $-spend ceilings**; on breach, **downgrade → queue → refuse** (below)                    |
+
+**Auto-pacing law (OAuth subs) — EVEN-SPREAD default (Jason override, 2026-06-22):** the fleet
+paces agentic token spend to consume the limit window **evenly over remaining time**:
+target rate = _(remaining usage available)_ ÷ _(remaining time in the window)_. Example: 100% of
+a 7-day window = **~14.285%/day**; the system tracks current usage and continuously re-splits the
+remainder evenly to hold pace. **Anticipated token-spend-per-task is the budgeting informant** —
+tasks are scheduled against the daily pace, not run until the quota is gone. Rationale: spreading
+delivery evenly beats rapidly exhausting usage and losing **multiple days of momentum**.
+**Rapid pacing / overspend requires EXPLICIT user authorization;** absent it, even-spread holds.
+Pacing is a control-plane decision, surfaced read-only before it throttles a lane.
+
+**Hard-cap breach behavior (ladder):** when a budget ceiling is hit mid-work, the fleet
+**downgrades first** (opus → sonnet → haiku, then Claude → Codex), **queues** the lane at the
+cheapest floor until the window resets, and **refuses** only as a last resort. Refusal is never
+the first response to a breach.
+
+**Spend accounting, learning & telemetry:**
+
+- **Multi-subscription auto-routing:** a tenant with multiple subscriptions may let the fleet
+  **auto-route work to the account with the most available usage** (within budget policy).
+- **Historical spend learning:** every task's token spend is **recorded**; historical data
+  continuously updates known **spend-per-task**, **typical daily spend**, and projections — so
+  estimates self-correct and pacing stays on target.
+- **Projected + actual spend on artifacts (Mosaic Stack mandate):** PRDs, missions, and task
+  decomposition **MUST note projected AND actual token spend** — a Mosaic Stack process standard
+  (template-level), tracked separately as **#622**.
+- **Anonymized telemetry → mosaicstack.dev:** spend data is reported (anonymous) to the
+  mosaicstack.dev telemetry endpoint so other agents/fleets budget and optimize from real,
+  anonymized data. Product workstream, tracked separately as **#623**.
+
+**User-settable budgets (the policy surface).** A tenant operator can set budgets for every
+configured **provider** (per-provider ceilings), the **account-to-task mapping**, the **agentic
+routing flow**, **concurrency** (the spend multiplier), and **hard API-token $-limits**. Budgets
+are enforced at the orchestrator + routing boundary, not inside individual workers (a worker never
+decides its own budget — see delegation discipline).
+
+**Budget CLI UX (#558):** `mosaic budget set --reset-at` sets the window reset; reset-datetimes
+carry **confidence tags** (`user` / `provider` / `estimated` / `unknown`); and **urgency/criticality
+is a dispatch-gate modifier** — high-urgency work may override even-spread pacing **within
+authorization**. (Also feeds the budgeting workstream, not only this doc.)
+
 ## Observation model
 
 | Verb                                | Behavior                                                                                           |
@@ -96,15 +196,83 @@ Every artifact, starting Phase 2, MUST:
 > (blank for full-screen TUIs), and `attach` is read-write + resizes the session. The
 > verbs above restore "join and observe" safely.
 
+## Control plane & central register
+
+### Why the register must be Postgres
+
+The fleet is multi-host (w-jarvis + dragon-lin + future). A SQLite file is a local
+file — it is not a network service and cannot be shared across hosts. Beyond topology,
+Postgres MVCC eliminates the concurrent-writer corruption class Hermes hit with SQLite
+under multi-agent access.
+
+Access is exclusively through the **gateway API** (`apps/gateway` — typed, auth-gated,
+scoped tokens). No agent or dispatcher pane ever holds a raw DB credential; a
+compromised pane cannot corrupt or exfiltrate the register.
+
+### Architecture (layers)
+
+| Layer                  | Responsibility                                                                                                | Implementation                                                                                                                                                                                            |
+| ---------------------- | ------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **Register**           | Source of truth: agents, missions, tasks, heartbeats, spend                                                   | Postgres `fleet` schema — existing stack instance (`@mosaicstack/db`)                                                                                                                                     |
+| **Access**             | Typed, auth-gated API                                                                                         | Gateway `fleet/*` routes                                                                                                                                                                                  |
+| **Dispatcher**         | Brief classification, BOD review, planning/coding/review/test/deploy sequencing + gates → fleet task dispatch | **forge pipeline engine** (`runPipeline`/`resumePipeline`, brief classifier, BOD) **+ thin `forge-exec` adapter → `agent-send.sh`**; NOT a new daemon — forge is reused, only stage→agent dispatch is new |
+| **Orchestrator (Mos)** | Goals, missions, judgment, user/PA interface                                                                  | Context-light; sets intent → re-engages only for decisions                                                                                                                                                |
+
+### Dispatcher = forge (reuse, do not rebuild)
+
+The dispatcher is **not new work**: it is `@mosaicstack/forge`, a fully-implemented
+software-factory pipeline engine (brief → Board-of-Directors review → 3 planning stages →
+coding → review/remediation → testing → deploy). Forge already provides
+`runPipeline`/`resumePipeline`, a brief classifier, and a BOD persona loader, so the fleet
+does **not** re-implement sequencing, gate logic, or brief classification. The only new
+fleet-owned code is a thin **`forge-exec` TaskExecutor adapter** (`ForgeTask` →
+`agent-send.sh` to a named agent) — forge's single missing piece — tracked as a Gitea
+issue and built post-PoC. The Postgres register backs forge's pipeline state (durable
+`resumePipeline`, cross-host) in addition to cross-project missions/tasks/Kanban. The
+north-star **'board' role IS forge's Board-of-Directors** — reused from forge, not a new
+role implementation.
+
+### Docs as projections
+
+`docs/TASKS.md` and `MISSION-MANIFEST.md` are **generated projections** of the DB,
+not hand-maintained. The dispatcher (or a scheduled job) renders Markdown from
+`fleet.*` tables and commits the output. DB is authoritative; docs are for human
+reference.
+
+### Spend
+
+`fleet.spend_ledger` records projected and actual token spend per agent/mission/task
+(ties to issue #622). The dispatcher enforces budget caps before dispatching. Mos reads
+the roll-up via API — no raw DB access, no context-bloating dumps.
+
+### Federation
+
+Cross-host fleet state flows through federated gateway queries (existing
+`federation_peers` / `federation_grants` machinery). This is the existing north-star
+invariant: **control plane rides federation (W1), not a bespoke broker.** No new
+broker introduced.
+
+### Scope
+
+This is Phase 4–5 of this roadmap, materialized. It MUST NOT block the PoC (which
+runs correctly on files + `roster.yaml`). Begin when Phase 2 heartbeat protocol is
+stable and concurrent-agent count makes file coordination the bottleneck.
+
+### Open sub-decision
+
+Dedicated Postgres **instance** vs. dedicated **schema** in the existing instance.
+Recommendation: dedicated schema, existing instance (a migration file, not new infra);
+re-evaluate if isolation or write-volume demands it.
+
 ## Phased roadmap
 
-| Phase                  | Outcome                                                                                                                                      | Status  |
-| ---------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
-| 0–1                    | tmux PoC, hardening, published CLI v0.0.34 (#565–#568)                                                                                       | ✅ done |
-| **2 — Observability**  | `fleet ps` (host+tenant aware join), heartbeat protocol + dogfood stub answers it, `agent watch` (read-only), `agent send --verify` receipts | ▶ now   |
-| 3 — Real runtimes      | claude/codex/pi/opencode answer heartbeat; **hybrid lifecycle** (core always-on: orchestrator+reviewer; ephemeral workers per lane)          | planned |
-| 4 — Unified definition | one agent schema in gateway; `mosaic agent --new` → materialized per-tenant session; uid-tenant provisioning                                 | planned |
-| 5 — Control plane      | federation-backed cross-host × cross-tenant fleet view; **webUI** (surface chosen then) for MVP-X1 parity                                    | planned |
+| Phase                  | Outcome                                                                                                                                                                                                  | Status  |
+| ---------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
+| 0–1                    | tmux PoC, hardening, published CLI v0.0.34 (#565–#568)                                                                                                                                                   | ✅ done |
+| **2 — Observability**  | `fleet ps` (host+tenant aware join), heartbeat protocol + dogfood stub answers it, `agent watch` (read-only), `agent send --verify` receipts                                                             | ▶ now   |
+| 3 — Real runtimes      | claude/codex/pi/opencode answer heartbeat; **hybrid lifecycle** (core always-on: **orchestrator + enhancer**; ephemeral workers per lane)                                                                | planned |
+| 4 — Unified definition | one agent schema in gateway; `mosaic agent --new` → materialized per-tenant session; uid-tenant provisioning; **`fleet` schema migration + `forge-exec` TaskExecutor adapter (forge → `agent-send.sh`)** | planned |
+| 5 — Control plane      | federation-backed cross-host × cross-tenant fleet view; **webUI** (surface chosen then) for MVP-X1 parity; **central register live (spend ledger, docs-as-projections, multi-host Kanban)**              | planned |
 
 ## Decisions of record (2026-06-20, with Jason)
 
@@ -121,6 +289,108 @@ Every artifact, starting Phase 2, MUST:
   runtime-bin on PATH (baked into the pane command) + boot-survival (`enable` + linger),
   which `fleet init` should automate.
 
+## Decisions of record (2026-06-22, with Jason)
+
+- **Two-agent floor:** every fleet has, at minimum, an **orchestrator** and an **enhancer**.
+  The orchestrator is the user's point of contact and composes the fleet; the enhancer runs the
+  continuous-improvement loop (monitor → analyze → remediate → upgrade tools/skills/harness →
+  file Mosaic Stack bug reports) and **does not code or review**.
+- **Role library:** orchestrator, enhancer, coder, code review, security review, research,
+  board (moonshot/contrarian/technical/business/financial), operations — extensible; the
+  orchestrator (advised by the enhancer) adds roles as missions demand.
+- **Orchestrator chat connector:** the orchestrator is reachable over a user-chosen connector
+  (tmux now; Telegram/Discord/Matrix/Slack configurable). Validated live: **"Mos" orchestrator
+  on Discord** via the Claude Code discord channel plugin (w-jarvis).
+- **Session context cap = 200k tokens (GLOBAL to all Claude sessions):** Claude Code sessions are
+  capped at a **max 200k-token context window**. Long-running sessions extended toward 1M tokens
+  have proven **worse in practice** (degraded steering, off-plan divergence); 200k is the standard.
+  **Enforcement split:** the _window_ lives in **`~/.claude/settings.json`** (host-global) as
+  `"autoCompactWindow": 200000` + `"autoCompactEnabled": true`; the _1M-disable_ lives in **launch
+  ENV** (`CLAUDE_CODE_DISABLE_1M_CONTEXT=1`, plus `CLAUDE_CODE_AUTO_COMPACT_WINDOW=200000`) wherever
+  a `[1m]` model can be selected (`mos-claude.service` + the fleet Claude launcher), so every Claude
+  agent is capped at spawn. (settings = window; env = 1M-disable.)
+- **Worker context bound (#8):** workers are kept context-bounded via the **ephemeral-per-lane
+  lifecycle + native compaction**, not via the 200k knob. The explicit `autoCompactWindow` 200k knob
+  **stays Claude-specific** — the _principle_ (bounded context) extends to workers, the _knob_ does not.
+- **Orchestrator delegation discipline:** the orchestrator **delegates all delivery work** to
+  subagents / workflows / ultracode / coder agents and confines its own context to \*\*orchestration
+  - the personal-assistant lane\*\*. Keeping delivery out of the orchestrator's window keeps its
+    context unpolluted and measurably reduces off-plan divergence. The orchestrator coordinates and
+    decides; it does not implement.
+- **Budget governance is fleet doctrine:** token/API-dollar budgeting is a first-class fleet concern
+  (see "Budget & token governance"). OAuth-sub usage-vs-limit feedback is ingested per account, spend
+  is **auto-paced EVEN-SPREAD over remaining time** (rapid/overspend only on explicit authorization),
+  spend is **tracked historically** to self-correct per-task/daily estimates, multi-sub tenants may
+  **auto-route by available usage**, and operators set budgets per provider, per account-to-task
+  mapping, per routing flow, per concurrency level, and as hard API-$ ceilings.
+- **Spend accounting is a Mosaic Stack process mandate:** PRDs, missions, and task decomposition
+  **MUST carry projected + actual token spend**; used locally for pacing and reported as **anonymized
+  telemetry to mosaicstack.dev**. The template standard (#622) and telemetry product (#623) are
+  tracked separately.
+- **Unified identity = "Fleet" (Jason, 2026-06-22):** the product is **Mosaic Fleet** — one unified
+  user-facing identity and CLI surface. **forge** is the Fleet's **internal** delivery/orchestration
+  engine (not a separate product); the control-plane **Postgres register is the Fleet's register**;
+  workers/runtime are the **Fleet substrate**. **"factory" is RETIRED as a product term** — it was
+  only ever the software-factory concept (which forge implements) and the old `mosaic-factory` tmux
+  socket name. The production-isolation socket is now **`mosaic-fleet`** (matches the product brand);
+  the legacy dogfood canary remains on the old `mosaic-factory` socket pending migration. **Code stays
+  layered** (forge + fleet + control-plane as internal layers);
+  only the **identity + CLI surface unify under Fleet.**
+- **Role-based session naming (Jason, 2026-06-22):** agent tmux sessions are named by **role**
+  (`orchestrator`, `enhancer`, `research`, `coder0-0`, …), not by persona. **Persona lives in
+  `SOUL.md`**; the front-end / Discord presents a **friendly alias** (e.g. "Mos" = the orchestrator's
+  alias). The session name is the stable addressing handle; the alias is presentation.
+
+### Control plane & central register
+
+- **Store:** Postgres (existing stack instance, dedicated `fleet` schema via `@mosaicstack/db`). SQLite rejected: (1) it is a local file — structurally incompatible with a multi-host fleet; (2) concurrent multi-agent writes caused repeated corruption in Hermes. "SQLite + access service" rejected as reinventing a DB server badly; "LLM agent gating DB access" rejected as slow, expensive, and a single point of failure.
+- **Access:** gateway API only (`apps/gateway`, `fleet/*` routes). No raw DB credentials in any agent/dispatcher pane — directly mitigates the tmux attack-surface concern.
+- **Dispatcher = forge (reuse, not a new build):** the dispatcher IS `@mosaicstack/forge`'s pipeline engine (`runPipeline`/`resumePipeline` + brief classifier + BOD persona loader), a fully-implemented software-factory pipeline (brief → BOD review → 3 planning stages → coding → review/remediation → testing → deploy). We do **not** design/build a new dispatcher and do **not** re-implement sequencing, gate logic, or brief classification. The only new fleet-owned piece is a thin **`forge-exec` TaskExecutor adapter** (suggested package `packages/forge-exec`) mapping a `ForgeTask` → `agent-send.sh` dispatch to a named fleet agent — forge's single missing piece. It is tracked as a Gitea issue and built **post-PoC** (not now).
+- **Register backs forge:** the Postgres `fleet` register is genuinely new (neither forge nor the fleet has cross-project state). It BACKS forge's pipeline state (durable `resumePipeline`, cross-host) plus cross-project missions/tasks/Kanban.
+- **'board' role = forge BOD:** the north-star role-library 'board' role IS forge's Board-of-Directors — reused, not reinvented.
+- **Orchestration vs. dispatch:** Orchestrator (Mos) sets intent and handles judgment; forge works the mechanical pipeline (sequencing, gates, status transitions, spend ledger). LLM escalation reserved for judgment: mission decomposition, re-planning on failure.
+- **Spend in the register:** `fleet.spend_ledger` tracks projected vs. actual tokens per agent/mission/task; ties to issue #622.
+- **Docs as projections:** `docs/TASKS.md` and `MISSION-MANIFEST.md` become generated exports of the DB, not hand-maintained.
+- **Sub-decision pending:** dedicated schema in existing PG instance (recommended) vs. dedicated PG instance. Revisit if isolation or write-volume demands it.
+
+## Decisions of record (2026-06-24, with Jason)
+
+- **Per-agent model switch (operator-configurable, NOT a global lock):** model selection is
+  **per-agent**, never a host-global pin. Claude sessions MUST NOT be locked to a single model in
+  `~/.claude/settings.json`; each agent chooses its model independently. The plumbing already exists —
+  roster `model_hint` → `MOSAIC_AGENT_MODEL` → `start-agent-session.sh` appends `--model <hint>` to that
+  agent's harness (claude or pi); settable today via `mosaic fleet add|edit <agent> --model <hint>`.
+  **North-star target:** surface this as a **per-agent model switch in the webUI** (with CLI/TUI parity
+  per MVP-X1) — read the roster, expose a per-agent model dropdown, write `model_hint` back, and restart
+  that one agent to apply. Unset = inherit the harness default. This **composes with** the budget
+  downgrade ladder (opus → sonnet → haiku, then Claude → Codex): the operator sets the per-agent model
+  _intent/ceiling_; budget pacing may downgrade within policy. Tracked as a Fleet `TASKS.md` entry under
+  the Phase-5 webUI surface.
+- **Orchestrator runtime (confirmed live):** the **orchestrator and enhancer run Claude Opus 4.8 in the
+  Claude Code harness**; only workers (coder/reviewer) run pi/gpt-5.5. Consistent with the 2026-06-20
+  "Claude reserved for Claude Code only" decision (the orchestrator runs _in_ Claude Code, not an
+  alternate Claude harness). Pi/gpt-5.5 as the orchestrator is permitted **only if proven** at least as
+  satisfactory; absent that proof, the orchestrator stays on Claude Opus 4.8.
+
+## Future enhancements (north-star, post-MVP — not on the MVP track)
+
+- **Mosaic Claude Discord Plugin** — a first-party Mosaic Discord connector that properly
+  implements the basic Discord functions **and native Discord threads**. Threads let a user
+  separate conversation topics with the orchestrator (the pattern proven by the Hermes agent).
+  A major enhancement over the current third-party channel plugin; **not required for the MVP**,
+  but a committed north-star target. `ASSUMPTION:` ships as a Mosaic-owned plugin so the fleet
+  controls Discord UX (threads, reactions, attachments, per-thread context) end-to-end.
+- **Matrix on a local homeserver — strategic future transport.** **F4 (in progress) IS the Matrix
+  connector**: an orchestrator chat connector speaking the Matrix client-server API against a
+  self-hosted homeserver (Conduit default, Synapse alt). Matrix is named here as the strategic
+  future transport — peer to tmux/Discord, not superseded by them.
+- **tmux fleet attack-surface hardening.** Many always-on tmux sessions are an attack surface;
+  `tmux send-keys` / socket access could enable malicious action against agents directly.
+  Mitigations to build toward: socket ownership/perms, per-tenant socket isolation (already an
+  invariant), authenticated `agent-send`, and an audit of who can write to any pane. **Post-MVP
+  unless a P0 surfaces.** The control-plane register reinforces this (gateway-API access = no raw
+  DB creds in panes). A not-started risk-assessment + mitigation-plan task rides the Fleet `TASKS.md`.
+
 ## Assumptions (veto-able)
 
 - `ASSUMPTION:` first-class runtimes = claude, codex, pi, opencode; a "role" (analyst,
@@ -131,3 +401,30 @@ Every artifact, starting Phase 2, MUST:
 - `ASSUMPTION:` Fleet is workstream **W-FLEET** under `mvp-20260312`; a rollup row in
   `docs/TASKS.md` and a workstream declaration in `MISSION-MANIFEST.md` are proposed to
   the MVP orchestrator, not written by this workstream.
+- `ASSUMPTION:` OAuth-subscription runtimes (Claude sub, Codex sub) expose a machine-readable
+  current-usage-vs-limit signal the fleet can poll/ingest; if a provider exposes no such signal,
+  that provider's accounts fall back to API-style hard-ceiling budgeting only (no auto-pacing).
+- `ASSUMPTION:` budget policy lives at the orchestrator + routing layer and is surfaced through the
+  same CLI→TUI→webUI parity (MVP-X1) as the rest of fleet state — not a separate budgeting daemon.
+- `ASSUMPTION:` the 200k session cap is enforced by Claude Code settings/env composition (model
+  variant + `autoCompactWindow`), not by a Mosaic wrapper; a wrapper is the fallback only if the
+  harness later removes those knobs.
+- `ASSUMPTION:` The central register (Postgres `fleet` schema + gateway API + forge as dispatcher) is
+  the Phase 4–5 control plane, begun after Phase 2 observability is proven. It is a dedicated
+  **W-FLEET** sub-workstream entry, not a separate mission. The dispatcher is `@mosaicstack/forge`
+  (reused, not a new daemon); the only new fleet-owned code is the thin **`forge-exec` TaskExecutor
+  adapter** (suggested package `packages/forge-exec`, `ForgeTask` → `agent-send.sh`), tracked as a
+  Gitea issue and built post-PoC.
+
+---
+
+> **Release procedure (drift re-capture, 2026-06-22):** `mosaic update` only propagates new fleet
+> commands when the **CLI version is bumped** — without a version bump, fleet command changes never
+> reach installed hosts. The release/version-bump procedure (bump → publish → `mosaic update`
+> [→ `--relaunch`]) must be documented so fleet changes actually land. (Also feeds the budgeting
+> workstream.)
+>
+> **Tracked separately (not in scope for this doc PR):** **#622** PRD/mission/task projected+actual
+> spend template standard · **#623** anonymized spend telemetry → mosaicstack.dev (product) ·
+> **#625** `tenant_id` roster-schema field (multi-tenant; invariant #1 home) · **#628** `forge-exec`
+> TaskExecutor adapter (post-PoC). This PR records **doctrine only** — no implementation.

docs/guides/admin-guide.md

@@ -293,13 +293,28 @@ Each OIDC provider requires its client ID, client secret, and issuer URL togethe
 
 ### Plugins
 
-| Variable               | Description                                                                |
-| ---------------------- | -------------------------------------------------------------------------- |
-| `DISCORD_BOT_TOKEN`    | Discord bot token (enables Discord plugin)                                 |
-| `DISCORD_GUILD_ID`     | Discord guild/server ID                                                    |
-| `DISCORD_GATEWAY_URL`  | Gateway URL for Discord plugin to call (default: `http://localhost:14242`) |
-| `TELEGRAM_BOT_TOKEN`   | Telegram bot token (enables Telegram plugin)                               |
-| `TELEGRAM_GATEWAY_URL` | Gateway URL for Telegram plugin to call                                    |
+| Variable                      | Description                                                                                                                               |
+| ----------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------- |
+| `DISCORD_BOT_TOKEN`           | Discord bot token (enables Discord plugin)                                                                                                |
+| `DISCORD_SERVICE_TOKEN`       | Required high-entropy service credential used to authenticate and sign Discord ingress; inject through the approved secret mechanism only |
+| `DISCORD_SERVICE_USER_ID`     | Required Mosaic service-principal user ID that owns persisted Discord conversations; the original Discord user ID remains audit metadata  |
+| `DISCORD_GUILD_ID`            | Discord guild/server ID                                                                                                                   |
+| `DISCORD_GATEWAY_URL`         | Gateway URL for Discord plugin to call (default: `http://localhost:14242`)                                                                |
+| `DISCORD_ALLOWED_GUILD_IDS`   | Required comma-separated Discord guild snowflake allowlist; default-deny                                                                  |
+| `DISCORD_ALLOWED_CHANNEL_IDS` | Required comma-separated Discord channel snowflake allowlist; default-deny                                                                |
+| `DISCORD_ALLOWED_USER_IDS`    | Required comma-separated Discord user snowflake allowlist; default-deny                                                                   |
+| `TELEGRAM_BOT_TOKEN`          | Telegram bot token (enables Telegram plugin)                                                                                              |
+| `TELEGRAM_GATEWAY_URL`        | Gateway URL for Telegram plugin to call                                                                                                   |
+
+### Discord ingress security
+
+When `DISCORD_BOT_TOKEN` is configured, `DISCORD_SERVICE_TOKEN`, `DISCORD_SERVICE_USER_ID`, and all three Discord allowlists are required. Gateway startup fails rather than enabling a broad or unauthenticated remote-control surface. The service user ID identifies a provisioned Mosaic service principal for persistence; the original Discord user ID is retained in ingress audit metadata. The service token is a secret supplied by the approved runtime secret mechanism and is never committed or logged.
+
+Inbound Discord messages must originate from an allowed guild, channel, and user, mention the bot, and carry a signed envelope containing the native Discord message ID and a generated correlation ID. The gateway validates the service identity, envelope signature, and allowlists again before dispatching. Replayed Discord message IDs are rejected during the bounded ingress replay window. Durable inbox/idempotency retention is introduced with Tess durable state.
+
+### Session retention and garbage collection
+
+Session cleanup is scoped to one session identifier and only removes that session's Valkey keys and demotes that session's hot logs. Gateway startup and scheduled jobs do not perform global session cleanup; startup removes legacy repeatable `session-gc` schedules created by older deployments. The `/gc` command is intentionally disabled until a distinct global-retention job supplies explicit authorization and audit evidence. This prevents one tenant or session's cleanup from changing another's retained data.
 
 ### Observability
 

docs/guides/fleet-local-canary.md

@@ -1,7 +1,7 @@
 # Local Fleet Canary
 
 The local fleet canary runs a small tmux-backed Mosaic agent fleet on an
-isolated tmux socket. The default socket is `mosaic-factory`; the commands do
+isolated tmux socket. The default socket is `mosaic-fleet`; the commands do
 not use or stop the default tmux server.
 
 ## Files
@@ -67,7 +67,7 @@ mosaic agent tail canary-pi -n 80
 
 These commands read the roster and target the configured tmux socket. The
 generated systemd agent services use `start-agent-session.sh`; message delivery
-uses the tmux send tools with `-L mosaic-factory`.
+uses the tmux send tools with `-L mosaic-fleet`.
 
 `mosaic agent send` is operator-origin traffic unless a caller explicitly says
 otherwise. The CLI always passes a deterministic source label to
@@ -82,7 +82,7 @@ impersonating a known handoff lane. The lower-level inter-agent wrapper
 Use these checks before expanding the roster:
 
 ```bash
-tmux -L mosaic-factory ls
+tmux -L mosaic-fleet ls
 tmux ls
 mosaic fleet verify
 systemctl --user status mosaic-tmux-holder.service
@@ -90,7 +90,7 @@ systemctl --user status mosaic-tmux-holder.service
 
 Expected results:
 
-- `tmux -L mosaic-factory ls` shows `_holder` and roster agent sessions.
+- `tmux -L mosaic-fleet ls` shows `_holder` and roster agent sessions.
 - `tmux ls` shows only the default tmux server sessions and is not changed by
   fleet start/stop operations.
 - `mosaic fleet verify` checks exact session targets on the isolated socket.
@@ -108,7 +108,7 @@ Run this checklist before cutting or dogfooding a fleet release:
   repeated `start` against the named socket; verify the default tmux server is
   unchanged.
 - Liveness verification: run `mosaic fleet verify` and confirm roster sessions
-  with `tmux -L mosaic-factory ls` or exact `has-session` checks.
+  with `tmux -L mosaic-fleet ls` or exact `has-session` checks.
 - Package dry-run: run `npm pack --dry-run --json` from `packages/mosaic` and
   confirm `framework/fleet`, `framework/systemd/user`,
   `framework/tools/fleet`, and `framework/tools/tmux` assets are included.
@@ -140,5 +140,5 @@ This rollback leaves the default tmux server untouched. If a canary session is
 still present after service stop, remove only the isolated socket server:
 
 ```bash
-tmux -L mosaic-factory kill-server
+tmux -L mosaic-fleet kill-server
 ```

docs/native-kanban-sot/DOCUMENTATION-CHECKLIST.md

@@ -0,0 +1,36 @@
+# Documentation Completion Checklist — Native Kanban/SOT Canon
+
+**Tracking:** Mosaic Stack issue #751
+**Scope:** Requirements and contract publication only; runtime implementation follows in separate slices.
+
+## Required artifacts
+
+- [x] Project `docs/PRD.md` exists; the workstream requirements refine its task/project-management scope.
+- [x] Canonical workstream requirements published at `docs/requirements/native-kanban-sot.md`.
+- [x] Mission manifest, task decomposition, frozen shared contract, and typed contract declarations included.
+- [x] `docs/SITEMAP.md` updated.
+- [x] Independent initial review and final GO report stored under `docs/reports/native-kanban-sot/`.
+- [x] Task scratchpad stored under `docs/scratchpads/`.
+- [ ] User/Admin/Developer guides — N/A for canon-only publication; required in implementation slices that change behavior or operations.
+- [ ] OpenAPI and endpoint index — N/A until KBN-105 freezes implementation-ready endpoint contracts.
+
+## Structural and root hygiene
+
+- [x] Canonical requirements are under `docs/requirements/`.
+- [x] Workstream artifacts are under `docs/native-kanban-sot/`.
+- [x] Review reports are under `docs/reports/native-kanban-sot/`.
+- [x] No new unscoped document was added to the `docs/` root.
+- [x] Root mission/task rollups link to the workstream.
+
+## Review gate
+
+- [x] Author and independent reviewer are different agents.
+- [x] KCR-001–016 closure was independently verified.
+- [x] Ultron final gate returned GO with zero BLOCKER/HIGH findings.
+- [x] Formatter, lint, typecheck, strict contract TypeScript, link, scope, and invariant publication validation passed in the current Stack toolchain.
+- [ ] PR review, CI, squash merge, and issue closure remain required before publication completion.
+
+## Publishing
+
+- [x] Canonical source remains in-repository.
+- [x] No external publishing platform is required for this internal architecture contract.

docs/native-kanban-sot/INDEX.md

@@ -0,0 +1,64 @@
+# Native Kanban/SOT Canon
+
+**Status:** KCR-001–016 independently cleared; canonical publication is in progress under issue [#751](https://git.mosaicstack.dev/mosaicstack/stack/issues/751)
+**Date:** 2026-07-14
+**Implementation hold:** no feature implementation starts until this canon is squash-merged to `main` with terminal-green CI; after merge, every slice remains held until its KBN prerequisite graph is satisfied.
+
+## Artifacts
+
+| Artifact                                                                                 | Purpose                                                                                                                                                                                               |
+| ---------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| [Canonical requirements](../requirements/native-kanban-sot.md)                           | Canonical P0–P3 requirements, all seven ratified decisions, fixed invariants, thin MVP, recovery tiers, non-goals, and per-requirement acceptance criteria                                            |
+| [`MISSION-MANIFEST.md`](./MISSION-MANIFEST.md)                                           | Mission/authority boundaries, exact role chain, gate model, mandatory SecReview triggers, Certifier final/no-merge rule, and collision-free slice ownership                                           |
+| [`TASKS.md`](./TASKS.md)                                                                 | Dependency-ordered, bounded P0–P3 slices with IN/OUT scope, dependencies, shared contracts, file ownership, evidence, and USC coder2/3/4/5 parallelization                                            |
+| [`SHARED-CONTRACT.md`](./SHARED-CONTRACT.md)                                             | Remediated v1 integration contract: proof authority, exact failures/routes/DTOs/MCP ownership, concrete current-main field migration map, relational invariants, Coordinator split, recovery delivery |
+| [`contracts/kanban-schema.v1.ts`](./contracts/kanban-schema.v1.ts)                       | Drizzle target declarations including exact owner/principal membership, project congruence, tags/archive, proposals, persisted assignments, monotonic fences, durable retry, immutable evidence/audit |
+| [`contracts/mechanical-coordinator.v1.ts`](./contracts/mechanical-coordinator.v1.ts)     | Pure snapshot decision engine separated from persistence/service adapter; ID-bound approvals, bigint-safe fences, durable retry/quarantine, artifact-backed checkpoints, exact failures               |
+| [`contracts/health-state.v1.ts`](./contracts/health-state.v1.ts)                         | Discriminated public health, separate branded transaction-local write proof, and non-overlapping denial/transport/version-conflict mappings                                                           |
+| [`contracts/recovery-posture.v1.ts`](./contracts/recovery-posture.v1.ts)                 | Provider-neutral shape schema plus normative runtime refinement, cross-field constraints, and Lite/Standard/High-assurance defaults                                                                   |
+| [`tsconfig.json`](./tsconfig.json)                                                       | Strict no-emit project scope for linting and compiling the four frozen TypeScript contracts against the current Stack Drizzle declarations                                                            |
+| [`DOCUMENTATION-CHECKLIST.md`](./DOCUMENTATION-CHECKLIST.md)                             | Publication documentation gate and implementation-slice deferrals                                                                                                                                     |
+| [Initial independent review](../reports/native-kanban-sot/canon-initial-review-no-go.md) | KCR-001–016 findings that blocked the first draft                                                                                                                                                     |
+| [Final independent re-review](../reports/native-kanban-sot/canon-final-rereview-go.md)   | Closure matrix, reproducible validation evidence, and GO verdict                                                                                                                                      |
+| [Ultron final gate](../reports/native-kanban-sot/ultron-final-go.md)                     | Final requirements, authority, schema, migration, recovery, decomposition, and evidence review GO                                                                                                     |
+
+## Recommended USC lane partition
+
+| Lane       | Natural seam                         | Exclusive ownership                                                                                                                                     |
+| ---------- | ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **coder2** | Schema + migrations + recovery slice | Unified Drizzle schema, migration SQL/meta/journal/tests, then recovery parser/mechanism/runbook files                                                  |
+| **coder3** | Domain + Gateway + MCP server        | Workspace-safe repositories, DTOs/controllers/services, exact `apps/gateway/src/mcp/**` files, health proof, proposals, Coordinator persistence adapter |
+| **coder4** | Pure Coordinator + tooling           | `packages/coord` mechanical engine, CLI/MCP consumers, generated projection, one-way importer and cutover tooling; lane-serialized internally           |
+| **coder5** | Web                                  | Tasks/Projects Kanban/List/detail and later Coordinator/migration-review UI                                                                             |
+| **Mos**    | Serialized integration               | Canon publication, frozen-contract changes, shared-root/exports, integration gates, merge authority                                                     |
+
+The safe order is KBN-010 → KBN-100 → KBN-105, then coder3 Gateway/MCP server, coder4 CLI/projection, coder5 web, and coder2 recovery can proceed on disjoint files. coder4 then runs pure Coordinator → importer → cutover tooling serially. No two active slices edit the same files.
+
+## Recovery defaults
+
+| Tier           | RPO / RTO    | WAL / PITR          | Base backup | Restore / break-glass   | Off-cluster                                     |
+| -------------- | ------------ | ------------------- | ----------- | ----------------------- | ----------------------------------------------- |
+| Lite           | 24h / 24h    | disabled / disabled | daily       | quarterly / annual      | encrypted separate target                       |
+| Standard       | 1h / 8h      | q15m / 14d          | daily       | quarterly / semiannual  | encrypted separate object storage               |
+| High-assurance | **15m / 4h** | **q5m / 35d**       | **daily**   | **monthly / quarterly** | **encrypted base+WAL, separate failure domain** |
+
+These knobs affect recovery posture only. PostgreSQL remains the sole writable SOT in every tier. Fail-closed writes, generated-file non-authority, attributable post-recovery proposals, non-LLM Coordinator limits, and Certifier final-gate/no-merge authority are fixed for every tier.
+
+## Non-blocking implementation sub-decisions for Mos
+
+The source plan and ratified seven decisions resolve all build-blocking product choices. The following implementation-local selections remain for the owning slices/Mos and must not weaken v1:
+
+1. Exact PostgreSQL write-health probe SQL and bounded proof lifetime; authority and failures are frozen.
+2. Dependency-cycle serialization mechanism (recursive CTE plus transaction/advisory lock or equivalent); required behavior is frozen.
+3. Whether RLS lands in the first migration or immediately after the tested session-context pattern; workspace constraints/repository authorization are required from migration one.
+4. Concrete off-cluster backup provider/bucket and selected production recovery tier; High-assurance minima are frozen if selected.
+5. Cutover reconciliation thresholds and stabilization duration, to be owner-approved before P3 execution.
+
+None authorizes a second writer, dual sync, LLM scheduling, Coordinator gate waiver/merge, or Certifier merge authority.
+
+## Publication validation evidence
+
+- Concrete TypeScript contracts are formatted with repository Prettier.
+- All four contracts pass strict TypeScript no-emit checking against the current Stack Drizzle toolchain.
+- Contract remediation and KCR-001–016 traceability are recorded in the issue scratchpad and linked review reports.
+- Independent re-review returned GO with KCR-001–016 closed; implementation remains held until canon merge and the dependency-ordered KBN prerequisites complete.

docs/native-kanban-sot/MISSION-MANIFEST.md

@@ -0,0 +1,195 @@
+# Mission Manifest — Mosaic Native Kanban and Canonical Task SOT P0–P3
+
+**Mission status:** CANON INDEPENDENTLY APPROVED; publication in progress under issue [#751](https://git.mosaicstack.dev/mosaicstack/stack/issues/751)
+**Date:** 2026-07-14
+**Human decision owner:** Jason
+**Orchestrator/publication owner:** web1 control plane (`mos-claude`; `mosaic-100` acting during Claude quota outage)
+**Execution topology:** USC web1, partitioned across collision-free GPT coder2/3/4/5 lanes
+**Canonical requirements:** [`../requirements/native-kanban-sot.md`](../requirements/native-kanban-sot.md)
+**Frozen integration contract:** `SHARED-CONTRACT.md` and `contracts/*.v1.ts`
+
+## 1. Mission statement
+
+Extend current `mosaicstack/stack` main into the sole native control plane for workspace-scoped project, mission, milestone, task, dependency, assignment, lease, approval, evidence, and audit state. First deliver a thin writable Kanban/List vertical slice; then add deterministic mechanical coordination and execute a one-way migration/cutover from jarvis-brain/Vikunja project/task stores.
+
+Success means every user, agent, orchestrator, specialist, and UI sees and mutates the same PostgreSQL aggregate revisions through typed Gateway commands, with no writable fallback and no hidden second authority.
+
+## 2. Scope boundaries
+
+### In scope
+
+- Current Drizzle/PostgreSQL schema extension and migrations.
+- Workspace tenancy and authorization from the first migration.
+- Projects, missions, milestones, tasks, normalized tags, dependencies, assignments, durable execution/quarantine state, links, immutable artifacts/evidence joins, outage change proposals, events, approvals, leases, checkpoints, and transactional outbox.
+- NestJS Gateway queries and explicit lifecycle commands.
+- MCP/CLI agent surfaces and generated read-only projections.
+- Thin writable Next.js Tasks Kanban/List, task detail, minimal Projects CRUD, filters, dependency readiness, ownership/lease separation, and audit timeline.
+- Non-LLM Mechanical Coordinator eligibility, proposal, approval-policy, lease/fence, heartbeat, retry, expiry, quarantine, and restart recovery.
+- Planning, Enhance, Coder, Review, SecReview, PR-Monitor, and Certifier role/gate representation.
+- One-way shadow importer, reconciliation, write freeze, final delta, cutover, rollback package, and legacy read-only stabilization.
+- Recovery-posture configuration and health-state/fail-closed contract.
+
+### Out of scope
+
+- Greenfield services, Prisma runtime revival, or jarvis-brain flat files as runtime storage.
+- Writable Markdown/JSON/Valkey/browser/provider fallback.
+- Gitea issue/PR replacement or generic bidirectional provider sync.
+- Calendar, email, GLPI cache, CRM, billing, time tracking, personal-brain migration.
+- LLM scheduling or scope interpretation by the Coordinator.
+- Autonomous gate waiver, certification, merge, release, deployment, or issue closure by Coordinator.
+- Merge authority for Certifier.
+- P4 full portfolio/mission designer and P5 fleet-scale policy unless separately released.
+
+## 3. Fixed invariants
+
+Every deployment MUST preserve all of the following:
+
+1. PostgreSQL is the sole writable SOT.
+2. Drizzle on current stack main is the only persistence foundation.
+3. Mutations fail closed when DB write-health cannot be proven `healthy`.
+4. No file, Valkey, browser, queue, provider, or human note becomes a fallback writer.
+5. `TASKS.md`, `mission.json`, and every file export are generated, read-only, non-authoritative, and never import sources.
+6. Human outage notes become attributable post-recovery proposals only.
+7. Workspace is the hard tenant; Team is intra-workspace authorization.
+8. Valkey is expendable; PostgreSQL owns state, leases, fencing, audit, and outbox.
+9. Mechanical Coordinator is deterministic/non-LLM and cannot invent scope, waive gates, certify, or merge.
+10. Certifier is the final independent quality gate and has no merge authority.
+11. Mutations use idempotency and optimistic aggregate versions; worker commands also require a current fencing token.
+12. Recovery tier changes only backup/recovery posture, never authority or gate semantics.
+
+## 4. Configurable recovery posture
+
+Deployments select Lite, Standard, or High-assurance defaults from [`../requirements/native-kanban-sot.md`](../requirements/native-kanban-sot.md) and `contracts/recovery-posture.v1.ts`. Configurable fields are limited to:
+
+- backup/base-backup cadence;
+- RPO and RTO targets;
+- PITR retention;
+- WAL archive cadence;
+- restore-test frequency;
+- break-glass drill frequency;
+- encrypted off-cluster storage.
+
+High-assurance defaults are fixed reference values: RPO 15 minutes, RTO 4 hours, encrypted off-cluster WAL every 5 minutes with 35-day PITR, daily base backup, monthly restore test, and quarterly break-glass drill.
+
+## 5. Canonical role map
+
+```text
+User
+  ↓ objectives, constraints, ratified decisions
+Interaction Layer
+  ↓ workspace/project context; no scheduling authority
+Portfolio Orchestrator
+  ↓ approved mission, cross-project priority/capacity
+Project Sub-Orchestrator
+  ↓ decomposition, DAG, acceptance, release, routing policy, overrides
+Gateway
+  ↓ authenticated/authorized typed commands
+Project/Task Domain Services
+  ↓ transactional state + semantic event + outbox
+Mechanical Coordinator
+  ↓ deterministic eligibility/proposal/lease/fence/retry/quarantine
+Specialists
+  Planning → Enhance → Coder → Review → conditional SecReview → remediation
+  ↓ complete evidence bundle
+Certifier
+  ↓ final pass/reject/escalate; NO merge authority
+Project Sub-Orchestrator / control plane
+  ↓ merge authority after all gates
+Post-merge validation
+```
+
+### Authority table
+
+| Role/layer               | Owns                                                                                                        | Explicitly cannot do                               |
+| ------------------------ | ----------------------------------------------------------------------------------------------------------- | -------------------------------------------------- |
+| User                     | Objectives, constraints, Jason-owned decisions                                                              | Direct DB/file authority bypass                    |
+| Interaction              | Conversation and context resolution                                                                         | Schedule, approve, lease, certify                  |
+| Portfolio Orchestrator   | Mission approval, cross-project priority/capacity/global holds                                              | Implement or self-certify specialist work          |
+| Project Sub-Orchestrator | Task decomposition/DAG/acceptance, release to ready, routing policy, overrides, remediation, merge go-ahead | Bypass required independent gates                  |
+| Gateway                  | Identity, tenancy, DTO validation, commands, state-machine enforcement                                      | Accept file edits or client SQL as mutations       |
+| Domain services          | Transactional business invariants, semantic events/outbox                                                   | Depend on Valkey/files for committed truth         |
+| Mechanical Coordinator   | Eligibility, dependencies, proposal, approved routing, lease/fence, heartbeat, retry/quarantine             | Invent/alter scope, waive gates, certify, merge    |
+| Specialists              | Bounded planning/implementation/review artifacts under a task lease                                         | Modify another lane's owned files or self-approve  |
+| Certifier                | Final independent evidence/traceability/gate decision                                                       | Merge, close provider issue, release, waive policy |
+
+## 6. Gate model
+
+### Mandatory gates
+
+1. Requirements/contract freeze before parallel implementation.
+2. P0 schema/authority threat model and tenant isolation review.
+3. Author and reviewer MUST be different principals/sessions.
+4. Functional review validates requirements, endpoint registry, concurrency, and negative paths.
+5. **Mandatory SecReview (`secrev`)** for any auth, authorization, tenant, service-token, secret, database schema/migration, data-integrity, import/cutover, audit, lease/fencing, recovery, or destructive-retirement surface.
+6. Review findings enter bounded remediation owned by the implementation lane.
+7. Raising reviewer re-verifies remediation.
+8. Certifier performs the final independent evidence and traceability gate.
+9. Merge authority remains with `mos-claude`/Project Sub-Orchestrator control plane after gates pass.
+10. Post-merge CI and situational validation must be terminal green before closure.
+
+### Gate outcomes
+
+- **PASS:** evidence complete; next authority may proceed.
+- **REJECT:** findings are explicit and route to remediation.
+- **ESCALATE:** policy/owner decision required; no implicit waiver.
+
+No role can transform a missing gate into a warning by changing status, editing a projection, or writing Valkey.
+
+## 7. Slice ownership rules
+
+1. USC web1 is the sole execution environment; coder2/3/4/5 are independent bounded lanes under Mos.
+2. Every slice has one named file-tree owner and an explicit IN/OUT boundary in `TASKS.md`.
+3. Two active slices MUST NOT edit the same source file, migration file, generated snapshot, lockfile, or API contract.
+4. coder2 exclusively owns `packages/db/src/schema.ts`, `packages/db/drizzle/**`, migration journal/meta/tests, then its disjoint recovery-parser/runbook slice. All schema requests serialize through coder2.
+5. Frozen `contracts/*.v1.ts` are read-only inputs during implementation. Contract changes require Mos approval, a version bump/amendment, and coordinated rebase before work resumes.
+6. coder3 exclusively owns Gateway DTO/controllers/services and the enumerated `apps/gateway/src/mcp/**` server files. coder4 owns CLI/projection clients and never edits MCP server files. Web consumers use the exact KBN-105 endpoint/DTO freeze.
+7. coder4 executes one lane order: CLI/projection → pure Coordinator → importer → cutover. The pure Coordinator under `packages/coord` does not load IDs or access DB, Gateway, Valkey, recovery I/O, or web files; coder3 owns the persistence/service adapter.
+8. Migration/import tooling calls Gateway/migration-only approved ports and does not add a second database model.
+9. Each lane commits only its owned files and reports any needed cross-slice change as a contract-change request instead of editing another lane's tree.
+10. Cross-review is mandatory: no lane reviews its own changes. Recommended ring is coder2 ← coder5, coder3 ← coder2, coder4 ← coder3, coder5 ← coder4, followed by independent SecReview where triggered and Certifier final.
+11. Integration-only edits are a separate serialized slice after component lanes are green; no opportunistic merge-conflict resolution may alter semantics.
+
+## 8. Delivery phases and exit gates
+
+### P0 — Canon and authority foundation
+
+- Publish this canon, frozen schema/ports/health/recovery contracts, threat model, authorization matrix, exact endpoint/DTO registry, concrete current-main field-by-field migration map, and standards amendment.
+- Build hold remains active until independent author≠reviewer re-review returns GO on health proof/failures, approval binding, fencing, tenant relationships, proposals, migration map, slice ordering/API freeze, recovery validation, and vocabulary alignment.
+- Exit: no unresolved second writer or contract blocker, tenant boundary frozen, all seven decisions traceable, and independent re-review GO recorded.
+
+### P1 — Thin native MVP
+
+- Schema/migration, tenant-safe Gateway, CLI/MCP/projection, writable Kanban/List/Projects, dependencies/readiness/audit.
+- Exit: same revision across web/CLI/MCP/projection; cross-workspace tests fail closed; generated files cannot mutate state.
+
+### P2 — Mechanical coordination
+
+- Agent/session registry, deterministic engine, approval queue, PostgreSQL leases/fencing/checkpoints/outbox, retry/quarantine, operations UI.
+- Exit: one lease winner, stale tokens rejected, dependencies/approvals enforced, DB/Valkey fault semantics proven, Certifier gate has no merge authority.
+
+### P3 — Shadow migration and cutover
+
+- Importer, lineage, reconciliation, reviewer UI, write freeze, final delta, Gateway switch, legacy read-only, stabilization and rollback package.
+- Exit: signed reconciliation, zero active legacy writers, scoped Gateway identities, imported backlog cannot dispatch accidentally.
+
+## 9. Evidence required for mission closure
+
+- Requirement-to-test/evidence matrix.
+- Schema/migration and N-1 rolling-deploy proof.
+- Cross-workspace API/repository/import/Coordinator negative tests.
+- Health-state and fail-closed fault injection.
+- Valkey-loss/outbox replay and Coordinator restart tests.
+- Concurrent lease and stale fencing tests.
+- Endpoint-registry alignment across web/CLI/MCP/Gateway.
+- Accessible real-Gateway Kanban journeys.
+- Generated projection tamper/no-import proof.
+- One-way migration dry-run/apply/verify and field reconciliation.
+- Author-independent functional review and required SecReview.
+- Certifier final decision and evidence bundle.
+- Merged main SHA, terminal green CI, closed linked task/issue, and post-merge situational validation under orchestrator ownership.
+
+## 10. Change control
+
+This manifest is derived from the ratified source plan. Any change to SOT authority, workspace tenancy, fixed statuses, Coordinator/Certifier authority, health-state semantics, schema v1, migration direction, or recovery-tier field set is a contract change. Contract changes require Jason/Mos authorization and cannot be inferred by an implementation lane.
+
+No coder lane may start while the build hold is active. KBN-010 must complete before KBN-100; KBN-105 exact endpoint/DTO freeze must complete before any API consumer implementation.