chore(release): bump @mosaicstack/mosaic 0.0.37 -> 0.0.38

Custom Pi harness: native heartbeat (turn-accurate busy/ok) + model self-report
+ model-callable mosaic_mission_status tool (#602); HB reader/writer consistency
(#599); agent watch viewer-leak + workdir test settle-race fixes (#601).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RMoEx7hfdFGjUiCHuN1RRi

docs/fleet/PRD-fleet-suite.md

@@ -0,0 +1,105 @@
+# PRD — Mosaic Fleet Suite (init, configure, operate)
+
+> **Workstream:** W-FLEET (Fleet) under mission `mvp-20260312` · **Phase:** 3→4 productization
+> **North star:** [docs/fleet/north-star.md](./north-star.md) · prior: Phase-2 observability (#579), durable launch (#581), real-agent enablement (#583/#584/#586), releases 0.0.35–0.0.37
+> **Lead:** Jarvis @ `w-jarvis`. **Collaborator:** coder agent @ `dragon-lin` (jwoltje@10.1.10.37:coder0-0).
+> Owner of this file: Fleet workstream lead. Does not modify MVP single-writer control-plane files.
+
+## Mission
+
+Turn the proven fleet primitives into a **user-installable, AI-free-configurable fleet product**:
+a user runs `mosaic fleet init`, answers a few questions (general / coding / research / hybrid),
+gets a recommended set of agents plus one always-on orchestrator wired for chat-ops, and can
+operate, mutate, re-create, and observe the fleet — over tmux today and Matrix tomorrow — from
+CLI/TUI and (designed-for) the webUI.
+
+**Immediate tangible goal:** the **"Mos"** orchestrator agent running on `w-jarvis`, reachable
+in **Discord channel `1517622518662434996`** (server `1112631390438166618`). Once the fleet is
+functional, we use the fleet itself to continue the work.
+
+## Requirements
+
+### A. Configure-without-AI CLI
+| ID | Requirement |
+|---|---|
+| R1 | `mosaic fleet` command set is functional end-to-end (init/install/start/stop/status/ps/verify + agent verbs). |
+| R2 | `mosaic fleet init` is an interactive, **AI-free** CLI wizard. |
+| R3 | Init asks the **configuration type**: `general`, `coding`, `research`, `hybrid`, … (extensible). |
+| R4 | Based on the answer, the fleet is populated with a **recommended set of agents** (a preset). |
+| R5 | **Exactly one main orchestrator agent** is always configured, regardless of type. |
+| R10 | A set of **recommended configurations (presets)** ships for easy duplication. |
+| R8 | User can **re-create** the fleet when config needs change (idempotent re-init / reconfigure). |
+| R17 | Fleet controls are **simple and intuitive**. |
+
+### B. Comms & orchestrator chat-ops
+| ID | Requirement |
+|---|---|
+| R6 | Init can wire the orchestrator to a chat connector — **Telegram / Discord / Matrix / Slack** — for command + comms. |
+| R7 | Designed with the end-goal of **Matrix comms on a locally-controlled server**. |
+| R16 | Fleet supports **tmux AND Matrix** comms, **user-configurable** at init or any time. Not all users want Matrix. |
+| R19 | **"Mos" orchestrator on Discord** (`chan 1517622518662434996` / `srv 1112631390438166618`) on `w-jarvis` — the first live target. |
+
+### C. Runtime, health, lifecycle
+| ID | Requirement |
+|---|---|
+| R9 | Fleet is **mutable by the orchestrator agent** — add/remove agents per need. |
+| R13 | Fleet **gracefully handles Pi + Claude harness updates** — keep harnesses current. |
+| R14 | The **Pi harness is customized** for proper tool usage, etc. |
+| R15 | **Agent heartbeat** properly configured for **Claude AND GPT/Pi** agents. |
+
+### D. Surfaces, testing, docs
+| ID | Requirement |
+|---|---|
+| R18 | Fleet built so the **webUI can view / monitor / terminate / butt-in** on a session. |
+| R11 | Installed and **tested on both `w-jarvis` and `dragon-lin`**. |
+| R12 | **Documentation**: how to install, configure, and use the fleet. |
+
+## Architecture / approach
+
+- **Config model:** `roster.yaml` is the source of truth (already exists). Add **presets** (`general`/`coding`/`research`/`hybrid`) as shipped example rosters; `init` selects a preset, always injects the orchestrator, and writes the roster. Re-init = regenerate roster (preserve user/site overrides — mirrors install env-merge from #567).
+- **Orchestrator agent:** always present; carries the chat connector config (connector type + target IDs) so it can be commanded over chat. tmux is the substrate; the connector bridges chat ↔ the orchestrator session.
+- **Comms layers (R16):** (1) **tmux** inter-agent (`agent-send`, proven) — default, always available. (2) **chat connector** for human↔orchestrator (Discord now; Matrix the strategic target). (3) **Matrix** as the locally-controlled cross-agent bus (future). Connector is pluggable + reconfigurable.
+- **Heartbeat (R15):** runtime-agnostic launcher sidecar already covers pi/claude/codex (#584). Refine per-runtime (native HB) with the **custom Pi harness** (R14) + a Claude path.
+- **Updates (R13):** `mosaic update` (CLI) + a fleet-aware harness-update step that refreshes pi/claude/codex and re-launches agents safely (drain → update → relaunch via the durable launcher).
+- **webUI (R18):** the fleet exposes machine-readable state (`fleet ps --json` already carries tenant/host/heartbeat/managed) + control verbs (start/stop/watch/send); webUI consumes these (control plane rides federation per north star). Ensure a stable JSON contract + a terminate/attach(butt-in) path.
+
+## Phases (incremental, each shippable)
+
+| Phase | Deliverable | Notes |
+|---|---|---|
+| **F1 Presets + init wizard** | preset rosters (general/coding/research/hybrid) + always-orchestrator + AI-free `fleet init` selecting a preset; re-init idempotent | R1–R5, R8, R10, R17 |
+| **F2 Connector + Mos-on-Discord** | orchestrator chat-connector config (Discord first) + **Mos live on Discord `1517…`/`1112…`** on w-jarvis | R6, R19, partial R16 |
+| **F3 Heartbeat + harness** | HB confirmed for claude + pi/gpt; **custom Pi harness** (tool usage, native HB, model self-report); graceful harness updates | R13, R14, R15 |
+| **F4 Matrix + comms toggle** | Matrix connector (local server) + user toggle tmux/Matrix at init/anytime | R7, R16 |
+| **F5 Orchestrator-mutable fleet** | orchestrator can add/remove agents at runtime | R9 |
+| **F6 webUI hooks** | stable JSON contract + terminate/attach surface for webUI view/monitor/terminate/butt-in | R18 |
+| **F7 Test + docs** | install+test on w-jarvis AND dragon-lin; user docs (install/configure/use) | R11, R12 (runs alongside every phase) |
+
+## Work division (proposed — confirm with dragon-lin)
+
+- **Jarvis @ w-jarvis (Lead):** F1 presets+wizard, F2 connector+Mos-on-Discord, F5 mutability, F6 webUI hooks; merge authority + dual-engine reviews; co-testing on w-jarvis.
+- **coder @ dragon-lin:** F3 custom Pi harness + harness-update flow (pi/codex-savvy); plus its in-flight constitution P4–P6 (P4 installer rework underpins `fleet init`/updates — coordinate the install path). Co-testing on dragon-lin (R11).
+- **Shared:** F4 Matrix (whoever has bandwidth); F7 testing/docs continuous.
+
+## Immediate target: Mos on Discord (F2 first slice)
+
+The discord plugin is available (`~/.claude.json`). Path: configure the **orchestrator** as a durable
+fleet session running Claude Code with the discord plugin bridged to channel `1517622518662434996`
+(server `1112631390438166618`) on w-jarvis, with the existing Discord Bridge Protocol (ack within
+~3s, reply via `mcp__discord__reply`, no `AskUserQuestion`). Heartbeat via the launcher sidecar.
+
+## Success criteria
+
+- A non-AI user can `mosaic fleet init`, pick a type, and get a working fleet + orchestrator.
+- **Mos answers in Discord `1517…`** on w-jarvis.
+- Fleet runs + is observable (`fleet ps`) on **both** w-jarvis and dragon-lin.
+- Harness updates handled gracefully; HB healthy for claude + pi/gpt agents.
+- Docs let a new operator install/configure/use the fleet.
+- Re-init + orchestrator mutation work.
+
+## Assumptions (veto-able)
+
+- `ASSUMPTION:` presets ship as example rosters under the framework (`fleet/examples/*.yaml`), selected by `init`.
+- `ASSUMPTION:` chat connectors are pluggable; Discord first (target exists), Matrix is the strategic default later.
+- `ASSUMPTION:` "Mos" = a Claude Code orchestrator session with the discord plugin (reuses the documented Discord Bridge Protocol).
+- `ASSUMPTION:` per north star, runtimes default to Codex/pi-on-Codex for workers; the orchestrator "Mos" runs Claude Code (in Claude Code, which is allowed).

packages/mosaic/framework/runtime/pi/mosaic-extension.ts

@@ -9,8 +9,16 @@
  *   4. Memory routing — remind agent to use ~/.config/mosaic/memory/
  */
 
-import type { ExtensionAPI } from '@mariozechner/pi-coding-agent';
+import type { ExtensionAPI, ExtensionContext } from '@earendil-works/pi-coding-agent';
-import { existsSync, readFileSync, writeFileSync, unlinkSync, mkdirSync } from 'node:fs';
+import { Type } from 'typebox';
+import {
+  existsSync,
+  readFileSync,
+  writeFileSync,
+  unlinkSync,
+  mkdirSync,
+  renameSync,
+} from 'node:fs';
 import { join, basename } from 'node:path';
 import { homedir } from 'node:os';
 import { execSync, spawnSync } from 'node:child_process';
@@ -25,6 +33,57 @@ const MOSAIC_HOME = process.env['MOSAIC_HOME'] ?? join(homedir(), '.config', 'mo
 // Helpers
 // ---------------------------------------------------------------------------
 
+// ---------------------------------------------------------------------------
+// Native heartbeat (fleet R14/R15)
+// ---------------------------------------------------------------------------
+// When this agent runs under the Mosaic fleet (MOSAIC_AGENT_NAME set), the
+// extension writes its OWN heartbeat in the same .hb contract `fleet ps` reads
+// (ts/pid/status[/model]) and touches a `.hb.native` precedence marker so the
+// shell sidecar defers. Native HB knows the real turn state (busy/ok), so it is
+// more accurate than the pane-PID-only sidecar fallback.
+const HB_AGENT_NAME = process.env['MOSAIC_AGENT_NAME'] ?? '';
+const HB_RUN_DIR = process.env['MOSAIC_HEARTBEAT_RUN_DIR'] ?? join(MOSAIC_HOME, 'fleet', 'run');
+const HB_INTERVAL_MS = (() => {
+  const s = Number.parseInt(process.env['MOSAIC_HEARTBEAT_INTERVAL'] ?? '', 10);
+  return Number.isFinite(s) && s > 0 ? s * 1000 : 15_000;
+})();
+
+function nativeHbEnabled(): boolean {
+  return HB_AGENT_NAME.length > 0;
+}
+
+function readModelId(ctx: ExtensionContext): string | null {
+  const m = ctx.model as unknown as { id?: string; name?: string } | undefined;
+  return m?.id ?? m?.name ?? null;
+}
+
+function writeNativeHeartbeat(status: 'ok' | 'busy', model: string | null): void {
+  if (!nativeHbEnabled()) return;
+  try {
+    mkdirSync(HB_RUN_DIR, { recursive: true });
+    const hb = join(HB_RUN_DIR, `${HB_AGENT_NAME}.hb`);
+    const lines = [`ts=${nowIso()}`, `pid=${process.pid}`, `status=${status}`];
+    if (model) lines.push(`model=${model}`);
+    const tmp = `${hb}.tmp.${process.pid}`;
+    writeFileSync(tmp, lines.join('\n') + '\n');
+    renameSync(tmp, hb); // atomic replace — fleet ps never reads a partial file
+    // Precedence marker: tells the shell sidecar that native HB is authoritative.
+    writeFileSync(join(HB_RUN_DIR, `${HB_AGENT_NAME}.hb.native`), nowIso() + '\n');
+  } catch {
+    // Best-effort: never let heartbeat I/O disrupt the Pi session.
+  }
+}
+
+function clearNativeMarker(): void {
+  if (!nativeHbEnabled()) return;
+  try {
+    const m = join(HB_RUN_DIR, `${HB_AGENT_NAME}.hb.native`);
+    if (existsSync(m)) unlinkSync(m); // native stopping — let the sidecar take over
+  } catch {
+    /* ignore */
+  }
+}
+
 function safeRead(filePath: string): string | null {
   try {
     return readFileSync(filePath, 'utf-8');
@@ -187,6 +246,9 @@ function buildMissionSummary(cwd: string, mission: ActiveMission): string {
 
 export default function register(pi: ExtensionAPI) {
   let sessionCwd = process.cwd();
+  let hbStatus: 'ok' | 'busy' = 'ok';
+  let hbModel: string | null = null;
+  let hbTimer: ReturnType<typeof setInterval> | null = null;
 
   // ── Session Start ─────────────────────────────────────────────────────
   pi.on('session_start', async (_event, ctx) => {
@@ -207,10 +269,39 @@ export default function register(pi: ExtensionAPI) {
     } else {
       ctx.ui.notify('Mosaic framework loaded', 'info');
     }
+
+    // Native heartbeat: write immediately, then on an interval. Idle = 'ok';
+    // turn_start/turn_end flip the status so `fleet ps` reflects real activity.
+    if (nativeHbEnabled()) {
+      hbModel = readModelId(ctx);
+      writeNativeHeartbeat('ok', hbModel);
+      hbTimer = setInterval(() => writeNativeHeartbeat(hbStatus, hbModel), HB_INTERVAL_MS);
+      if (typeof hbTimer.unref === 'function') hbTimer.unref();
+    }
   });
 
-  // ── Session End ───────────────────────────────────────────────────────
+  // ── Turn lifecycle → accurate busy/ok heartbeat ───────────────────────
-  pi.on('session_end', async (_event, _ctx) => {
+  pi.on('turn_start', async (_event, ctx) => {
+    hbStatus = 'busy';
+    hbModel = readModelId(ctx) ?? hbModel;
+    writeNativeHeartbeat('busy', hbModel);
+  });
+  pi.on('turn_end', async (_event, ctx) => {
+    hbStatus = 'ok';
+    hbModel = readModelId(ctx) ?? hbModel;
+    writeNativeHeartbeat('ok', hbModel);
+  });
+
+  // ── Session Shutdown ──────────────────────────────────────────────────
+  // (The pi API event is 'session_shutdown'; the prior 'session_end' handler
+  // never fired — fixed here so repo hooks + lock cleanup actually run.)
+  pi.on('session_shutdown', async (_event, _ctx) => {
+    if (hbTimer) {
+      clearInterval(hbTimer);
+      hbTimer = null;
+    }
+    clearNativeMarker();
+
     // Run repo session-end hook
     runRepoHook(sessionCwd, 'session-end');
 
@@ -252,4 +343,32 @@ export default function register(pi: ExtensionAPI) {
       }
     },
   });
+
+  // ── Register mosaic_mission_status tool (model-callable) ──────────────
+  // R14 "proper tool usage": give the agent a first-class tool to load its
+  // active Mosaic mission, milestone progress, task counts, and latest
+  // scratchpad — so it self-orients on in-flight work before planning,
+  // instead of shelling out or guessing. Mirrors the /mosaic-status command
+  // but returns the summary as tool output the LLM can read.
+  pi.registerTool({
+    name: 'mosaic_mission_status',
+    label: 'Mosaic Mission Status',
+    description:
+      'Return the active Mosaic mission, milestone progress, task counts, and latest scratchpad for the current project. Returns a note when no mission is active.',
+    promptSnippet: 'Read the active Mosaic mission + task state for the current project',
+    promptGuidelines: [
+      'Use mosaic_mission_status at the start of a session or task to load the active mission, milestone progress, and open tasks before planning work.',
+    ],
+    parameters: Type.Object({}),
+    async execute(_toolCallId, _params, _signal, _onUpdate, _ctx) {
+      const mission = detectMission(sessionCwd);
+      const text = mission
+        ? buildMissionSummary(sessionCwd, mission)
+        : 'No active Mosaic mission in this project.';
+      return {
+        content: [{ type: 'text', text }],
+        details: mission ? { ...mission } : { active: false },
+      };
+    },
+  });
 }

packages/mosaic/framework/tools/fleet/start-agent-session.sh

@@ -90,11 +90,18 @@ MOSAIC_RUNTIME_BIN_PREFIX=$(_build_runtime_bin_prefix)
 #
 # We build the snippet as a double-quoted here-string embedded in a printf call
 # to avoid nested quoting problems.
+#
+# MOSAIC_AGENT_NAME must also be exported INTO the pane: panes inherit the tmux
+# server environment (not this script's, and not the systemd unit's), so the
+# name would otherwise be empty in-pane and the runtime's native heartbeat
+# (which gates on MOSAIC_AGENT_NAME) would never fire. %q-quote it so it is a
+# safe single bash token regardless of the name's characters.
+AGENT_NAME_Q=$(printf '%q' "$AGENT_NAME")
 
 if [ -n "$MOSAIC_RUNTIME_BIN_PREFIX" ]; then
-  PANE_SHELL_SNIPPET="export PATH=\"${MOSAIC_RUNTIME_BIN_PREFIX}:\${PATH}\"; exec ${MOSAIC_AGENT_COMMAND}"
+  PANE_SHELL_SNIPPET="export MOSAIC_AGENT_NAME=${AGENT_NAME_Q}; export PATH=\"${MOSAIC_RUNTIME_BIN_PREFIX}:\${PATH}\"; exec ${MOSAIC_AGENT_COMMAND}"
 else
-  PANE_SHELL_SNIPPET="exec ${MOSAIC_AGENT_COMMAND}"
+  PANE_SHELL_SNIPPET="export MOSAIC_AGENT_NAME=${AGENT_NAME_Q}; exec ${MOSAIC_AGENT_COMMAND}"
 fi
 
 mkdir -p "$MOSAIC_AGENT_WORKDIR"
@@ -129,7 +136,7 @@ _start_heartbeat_sidecar() {
   # references to any variables from this script's environment.
   local sidecar_script
   sidecar_script=$(printf \
-    'hb=%s; pid=%s; iv=%s; mkdir -p "$(dirname "$hb")"; while kill -0 "$pid" 2>/dev/null; do tmp="$hb.tmp.$$"; printf "ts=%%s\npid=%%s\nstatus=ok\n" "$(date +%%Y-%%m-%%dT%%H:%%M:%%S%%z)" "$pid" > "$tmp" && mv "$tmp" "$hb"; sleep "$iv"; done' \
+    'hb=%q; pid=%q; iv=%q; mkdir -p "$(dirname "$hb")"; while kill -0 "$pid" 2>/dev/null; do nat="$hb.native"; if [ -f "$nat" ] && [ "$(( $(date +%%s) - $(stat -c %%Y "$nat" 2>/dev/null || echo 0) ))" -lt "$(( iv * 2 ))" ]; then sleep "$iv"; continue; fi; tmp="$hb.tmp.$$"; printf "ts=%%s\npid=%%s\nstatus=ok\n" "$(date +%%Y-%%m-%%dT%%H:%%M:%%S%%z)" "$pid" > "$tmp" && mv "$tmp" "$hb"; sleep "$iv"; done' \
     "$hb_file" "$pane_pid" "$interval")
 
   # setsid + disown ensures the sidecar survives this script exiting.

packages/mosaic/framework/tools/fleet/test-start-agent-session.sh

@@ -32,8 +32,15 @@ MOSAIC_AGENT_COMMAND='bash --noprofile --norc -i' \
   "$START" "$AGENT"
 
 tmux -L "$SOCKET" has-session -t "=$AGENT:0.0" || fail "agent session was not created"
+# Retry: pane_current_path briefly reflects the tmux server's cwd until the pane
+# process establishes its own cwd (the -c start dir). Poll until it settles.
+actual_dir=""
+for _ in $(seq 1 30); do
   actual_dir=$(tmux -L "$SOCKET" display-message -p -t "=$AGENT:0.0" '#{pane_current_path}')
-[ "$actual_dir" = "$WORKDIR" ] || fail "agent workdir mismatch: $actual_dir"
+  [ "$actual_dir" = "$WORKDIR" ] && break
+  sleep 0.1
+done
+[ "$actual_dir" = "$WORKDIR" ] || fail "agent workdir mismatch: $actual_dir (expected $WORKDIR)"
 
 # ── Test 2: idempotency (duplicate start prints 'already running') ─────────────
 MOSAIC_TMUX_SOCKET="$SOCKET" \

packages/mosaic/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mosaicstack/mosaic",
-  "version": "0.0.36",
+  "version": "0.0.38",
   "repository": {
     "type": "git",
     "url": "https://git.mosaicstack.dev/mosaicstack/stack.git",

packages/mosaic/src/commands/fleet.spec.ts

@@ -4,6 +4,7 @@ import { dirname, join, resolve } from 'node:path';
 import { Command } from 'commander';
 import { afterEach, describe, expect, it, vi } from 'vitest';
 import {
+  addAgentToRoster,
   buildAgentSendCommand,
   buildAgentWatchAttachCommand,
   buildAgentWatchCommand,
@@ -35,9 +36,11 @@ import {
   parseTmuxListPanes,
   parseTmuxListSessions,
   registerFleetCommand,
+  removeAgentFromRoster,
   resolveFleetPaths,
   resolvePresetFilename,
   RUNTIME_ACCEPTABLE_COMMANDS,
+  serializeRosterToYaml,
   VERIFY_DEFAULT_TIMEOUT_MS,
   VERIFY_POLL_INTERVAL_MS,
   type AgentPsRow,
@@ -68,10 +71,12 @@ describe('registerFleetCommand', () => {
 
     expect(fleet).toBeDefined();
     expect(fleet!.commands.map((command) => command.name()).sort()).toEqual([
+      'add',
       'init',
       'install',
       'install-systemd',
       'ps',
+      'remove',
       'restart',
       'start',
       'status',
@@ -828,6 +833,17 @@ describe('fleet ps — heartbeat parsing', () => {
     expect(hb.pid).toBe(12345);
     expect(hb.status).toBe('ok');
     expect(hb.ageMs).toBe(10_000);
+    // No model= line in a legacy/sidecar heartbeat → model is null.
+    expect(hb.model).toBeNull();
+  });
+
+  it('parses a self-reported model id from a native heartbeat (model= line)', () => {
+    const ts = new Date(NOW - 5_000).toISOString();
+    const content = `ts=${ts}\npid=42\nstatus=busy\nmodel=openai-codex/gpt-5.5:high\n`;
+    const hb = parseHeartbeat(content, NOW);
+    expect(hb.model).toBe('openai-codex/gpt-5.5:high');
+    expect(hb.status).toBe('busy');
+    expect(hb.health).toBe('healthy');
   });
 
   it('reports stale when heartbeat is older than 3×interval', () => {
@@ -2271,6 +2287,472 @@ describe('resolvePresetFilename', () => {
   });
 });
 
+// ---------------------------------------------------------------------------
+// Fleet Phase F5: orchestrator-mutable fleet — pure helper tests (R9)
+// ---------------------------------------------------------------------------
+
+describe('fleet add/remove — pure helpers', () => {
+  const baseRoster: FleetRoster = {
+    version: 1,
+    transport: 'tmux',
+    tmux: { socketName: 'mosaic-factory', holderSession: '_holder' },
+    defaults: { workingDirectory: '~/src' },
+    runtimes: { codex: { resetCommand: '/clear' } },
+    agents: [
+      { name: 'orchestrator', runtime: 'claude', className: 'orchestrator' },
+      { name: 'coder0', runtime: 'codex', className: 'worker' },
+    ],
+  };
+
+  it('addAgentToRoster appends a new agent and returns a new roster object', () => {
+    const newAgent = { name: 'reviewer0', runtime: 'pi', className: 'worker' };
+    const updated = addAgentToRoster(baseRoster, newAgent);
+    expect(updated.agents).toHaveLength(3);
+    expect(updated.agents[2]).toEqual(newAgent);
+    // immutable — original unchanged
+    expect(baseRoster.agents).toHaveLength(2);
+    expect(updated).not.toBe(baseRoster);
+  });
+
+  it('addAgentToRoster throws on duplicate name', () => {
+    expect(() =>
+      addAgentToRoster(baseRoster, { name: 'coder0', runtime: 'claude', className: 'worker' }),
+    ).toThrow('Agent "coder0" already exists in the fleet roster.');
+  });
+
+  it('addAgentToRoster throws on invalid name (invalid characters)', () => {
+    expect(() =>
+      addAgentToRoster(baseRoster, { name: 'bad name!', runtime: 'claude', className: 'worker' }),
+    ).toThrow('Invalid fleet agent name');
+  });
+
+  it('addAgentToRoster throws on empty name', () => {
+    expect(() =>
+      addAgentToRoster(baseRoster, { name: '', runtime: 'claude', className: 'worker' }),
+    ).toThrow('Invalid fleet agent name');
+  });
+
+  it('removeAgentFromRoster removes the agent and returns new roster', () => {
+    const updated = removeAgentFromRoster(baseRoster, 'coder0');
+    expect(updated.agents).toHaveLength(1);
+    expect(updated.agents[0]!.name).toBe('orchestrator');
+    // immutable
+    expect(baseRoster.agents).toHaveLength(2);
+    expect(updated).not.toBe(baseRoster);
+  });
+
+  it('removeAgentFromRoster throws when agent not found', () => {
+    expect(() => removeAgentFromRoster(baseRoster, 'nonexistent')).toThrow(
+      'Agent "nonexistent" is not in the fleet roster.',
+    );
+  });
+
+  it('removeAgentFromRoster throws when removing the sole orchestrator (guard)', () => {
+    const rosterWithOnlyOrch: FleetRoster = {
+      ...baseRoster,
+      agents: [{ name: 'orchestrator', runtime: 'claude', className: 'orchestrator' }],
+    };
+    expect(() => removeAgentFromRoster(rosterWithOnlyOrch, 'orchestrator')).toThrow(
+      'sole orchestrator',
+    );
+  });
+
+  it('removeAgentFromRoster allows removing an orchestrator when another remains', () => {
+    const rosterWithTwoOrchs: FleetRoster = {
+      ...baseRoster,
+      agents: [
+        { name: 'orchestrator', runtime: 'claude', className: 'orchestrator' },
+        { name: 'orchestrator2', runtime: 'claude', className: 'orchestrator' },
+        { name: 'coder0', runtime: 'codex', className: 'worker' },
+      ],
+    };
+    const updated = removeAgentFromRoster(rosterWithTwoOrchs, 'orchestrator');
+    expect(updated.agents.map((a) => a.name)).toEqual(['orchestrator2', 'coder0']);
+  });
+
+  it('serializeRosterToYaml produces YAML that round-trips through loadFleetRoster', async () => {
+    const yaml = serializeRosterToYaml(baseRoster);
+    expect(typeof yaml).toBe('string');
+    expect(yaml).toContain('version: 1');
+    expect(yaml).toContain('name: orchestrator');
+    expect(yaml).toContain('name: coder0');
+
+    // Round-trip: write to disk and re-load
+    const dir = await mkdtemp(join(tmpdir(), 'mosaic-fleet-'));
+    const rosterPath = join(dir, 'roster.yaml');
+    try {
+      await writeFile(rosterPath, yaml);
+      const loaded = await loadFleetRoster(rosterPath);
+      expect(loaded.agents.map((a) => a.name)).toEqual(['orchestrator', 'coder0']);
+      expect(loaded.tmux.socketName).toBe('mosaic-factory');
+      expect(loaded.agents[0]!.className).toBe('orchestrator');
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+
+  it('serializeRosterToYaml round-trips optional fields (modelHint, workingDirectory)', async () => {
+    const rosterWithOptionals: FleetRoster = {
+      ...baseRoster,
+      agents: [
+        {
+          name: 'orchestrator',
+          runtime: 'claude',
+          className: 'orchestrator',
+          modelHint: 'claude-3-5-sonnet',
+          workingDirectory: '/tmp/work',
+          persistentPersona: true,
+          resetBetweenTasks: false,
+        },
+      ],
+    };
+    const yaml = serializeRosterToYaml(rosterWithOptionals);
+    expect(yaml).toContain('model_hint: claude-3-5-sonnet');
+    expect(yaml).toContain('working_directory: /tmp/work');
+    expect(yaml).toContain('persistent_persona: true');
+
+    const dir = await mkdtemp(join(tmpdir(), 'mosaic-fleet-'));
+    const rosterPath = join(dir, 'roster.yaml');
+    try {
+      await writeFile(rosterPath, yaml);
+      const loaded = await loadFleetRoster(rosterPath);
+      expect(loaded.agents[0]!.modelHint).toBe('claude-3-5-sonnet');
+      expect(loaded.agents[0]!.workingDirectory).toBe('/tmp/work');
+      expect(loaded.agents[0]!.persistentPersona).toBe(true);
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+});
+
+// ---------------------------------------------------------------------------
+// Fleet Phase F5: fleet add command tests
+// ---------------------------------------------------------------------------
+
+describe('fleet add command', () => {
+  let home: string;
+
+  afterEach(async () => {
+    if (home) {
+      await rm(home, { recursive: true, force: true });
+    }
+  });
+
+  async function makeHome(agents = ['orchestrator']): Promise<string> {
+    const dir = await mkdtemp(join(tmpdir(), 'mosaic-fleet-add-'));
+    await mkdir(join(dir, 'fleet', 'agents'), { recursive: true });
+    const agentLines = agents.map((name) => {
+      const cls = name === 'orchestrator' ? 'orchestrator' : 'worker';
+      return `  - name: ${name}\n    runtime: claude\n    class: ${cls}`;
+    });
+    await writeFile(
+      join(dir, 'fleet', 'roster.yaml'),
+      ['version: 1', 'transport: tmux', 'agents:', ...agentLines].join('\n'),
+    );
+    return dir;
+  }
+
+  it('appends agent to roster file and writes env file', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync([
+      'node',
+      'mosaic',
+      'fleet',
+      'add',
+      'coder0',
+      '--runtime',
+      'codex',
+      '--class',
+      'worker',
+    ]);
+
+    const roster = await loadFleetRoster(join(home, 'fleet', 'roster.yaml'));
+    expect(roster.agents.map((a) => a.name)).toContain('coder0');
+
+    const envContent = await readFile(join(home, 'fleet', 'agents', 'coder0.env'), 'utf8');
+    expect(envContent).toContain('MOSAIC_AGENT_NAME=coder0');
+    expect(envContent).toContain('MOSAIC_AGENT_RUNTIME=codex');
+  });
+
+  it('--no-start skips the start command', async () => {
+    home = await makeHome();
+    const calls: string[][] = [];
+    const runner: CommandRunner = async (command, args) => {
+      calls.push([command, ...args]);
+      return { stdout: '', stderr: '', exitCode: 0 };
+    };
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync([
+      'node',
+      'mosaic',
+      'fleet',
+      'add',
+      'coder0',
+      '--runtime',
+      'codex',
+      '--class',
+      'worker',
+      '--no-start',
+    ]);
+
+    // No start command should have been issued
+    const startCalls = calls.filter((c) => c.includes('start'));
+    expect(startCalls).toHaveLength(0);
+  });
+
+  it('without --no-start, issues start command for the new agent', async () => {
+    home = await makeHome();
+    const calls: string[][] = [];
+    const runner: CommandRunner = async (command, args) => {
+      calls.push([command, ...args]);
+      return { stdout: '', stderr: '', exitCode: 0 };
+    };
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync([
+      'node',
+      'mosaic',
+      'fleet',
+      'add',
+      'coder0',
+      '--runtime',
+      'codex',
+      '--class',
+      'worker',
+    ]);
+
+    expect(calls).toContainEqual(['systemctl', '--user', 'start', 'mosaic-agent@coder0.service']);
+  });
+
+  it('throws when adding a duplicate agent name', async () => {
+    home = await makeHome(['orchestrator', 'coder0']);
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await expect(
+      program.parseAsync([
+        'node',
+        'mosaic',
+        'fleet',
+        'add',
+        'coder0',
+        '--runtime',
+        'codex',
+        '--class',
+        'worker',
+      ]),
+    ).rejects.toThrow('already exists');
+  });
+
+  it('throws when runtime is invalid', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await expect(
+      program.parseAsync([
+        'node',
+        'mosaic',
+        'fleet',
+        'add',
+        'coder0',
+        '--runtime',
+        'notaruntime',
+        '--class',
+        'worker',
+      ]),
+    ).rejects.toThrow('Invalid runtime');
+  });
+
+  it('accepts optional --model and --working-dir options', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync([
+      'node',
+      'mosaic',
+      'fleet',
+      'add',
+      'coder0',
+      '--runtime',
+      'claude',
+      '--class',
+      'worker',
+      '--model',
+      'claude-sonnet',
+      '--working-dir',
+      '/tmp/work',
+    ]);
+
+    const roster = await loadFleetRoster(join(home, 'fleet', 'roster.yaml'));
+    const agent = roster.agents.find((a) => a.name === 'coder0');
+    expect(agent?.modelHint).toBe('claude-sonnet');
+    expect(agent?.workingDirectory).toBe('/tmp/work');
+  });
+});
+
+// ---------------------------------------------------------------------------
+// Fleet Phase F5: fleet remove command tests
+// ---------------------------------------------------------------------------
+
+describe('fleet remove command', () => {
+  let home: string;
+
+  afterEach(async () => {
+    if (home) {
+      await rm(home, { recursive: true, force: true });
+    }
+  });
+
+  async function makeHome(): Promise<string> {
+    const dir = await mkdtemp(join(tmpdir(), 'mosaic-fleet-remove-'));
+    await mkdir(join(dir, 'fleet', 'agents'), { recursive: true });
+    await mkdir(join(dir, 'fleet', 'run'), { recursive: true });
+    await writeFile(
+      join(dir, 'fleet', 'roster.yaml'),
+      [
+        'version: 1',
+        'transport: tmux',
+        'agents:',
+        '  - name: orchestrator',
+        '    runtime: claude',
+        '    class: orchestrator',
+        '  - name: coder0',
+        '    runtime: codex',
+        '    class: worker',
+      ].join('\n'),
+    );
+    // Create env and heartbeat files for coder0
+    await writeFile(join(dir, 'fleet', 'agents', 'coder0.env'), 'MOSAIC_AGENT_NAME=coder0\n');
+    await writeFile(join(dir, 'fleet', 'run', 'coder0.hb'), 'ts=2026-01-01T00:00:00.000Z\n');
+    return dir;
+  }
+
+  it('removes agent from roster and writes back', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'coder0']);
+
+    const roster = await loadFleetRoster(join(home, 'fleet', 'roster.yaml'));
+    expect(roster.agents.map((a) => a.name)).not.toContain('coder0');
+    expect(roster.agents.map((a) => a.name)).toContain('orchestrator');
+  });
+
+  it('stop is called before roster write (stop is the first runner call)', async () => {
+    home = await makeHome();
+    const calls: string[][] = [];
+    const runner: CommandRunner = async (command, args) => {
+      calls.push([command, ...args]);
+      return { stdout: '', stderr: '', exitCode: 0 };
+    };
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'coder0']);
+
+    expect(calls[0]).toEqual(['systemctl', '--user', 'stop', 'mosaic-agent@coder0.service']);
+  });
+
+  it('stop failure is non-fatal — warns but still removes from roster', async () => {
+    home = await makeHome();
+    const stderrMessages: string[] = [];
+    const stderrSpy = vi.spyOn(process.stderr, 'write').mockImplementation((msg) => {
+      stderrMessages.push(String(msg));
+      return true;
+    });
+
+    const runner: CommandRunner = async (command, args) => {
+      if (args.includes('stop')) {
+        return { stdout: '', stderr: 'unit not found', exitCode: 5 };
+      }
+      return { stdout: '', stderr: '', exitCode: 0 };
+    };
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    try {
+      // Must not reject
+      await expect(
+        program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'coder0']),
+      ).resolves.toBeDefined();
+
+      // Agent should be removed from roster despite stop failure
+      const roster = await loadFleetRoster(join(home, 'fleet', 'roster.yaml'));
+      expect(roster.agents.map((a) => a.name)).not.toContain('coder0');
+
+      // Warning must have been emitted
+      expect(stderrMessages.join('')).toMatch(/Warning/);
+    } finally {
+      stderrSpy.mockRestore();
+    }
+  });
+
+  it('--keep-files skips env file deletion', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'coder0', '--keep-files']);
+
+    // Env file should still exist
+    const envContent = await readFile(join(home, 'fleet', 'agents', 'coder0.env'), 'utf8');
+    expect(envContent).toContain('MOSAIC_AGENT_NAME=coder0');
+  });
+
+  it('env file is removed by default (no --keep-files)', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    await program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'coder0']);
+
+    await expect(readFile(join(home, 'fleet', 'agents', 'coder0.env'), 'utf8')).rejects.toThrow();
+  });
+
+  it('removing the sole orchestrator throws with a clear error about the guard', async () => {
+    home = await makeHome();
+    const runner: CommandRunner = async () => ({ stdout: '', stderr: '', exitCode: 0 });
+    const program = new Command();
+    program.exitOverride();
+    registerFleetCommand(program, { runner, mosaicHome: home });
+
+    // First remove the worker so only the orchestrator remains
+    await program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'coder0']);
+
+    // Now attempt to remove the sole orchestrator
+    await expect(
+      program.parseAsync(['node', 'mosaic', 'fleet', 'remove', 'orchestrator']),
+    ).rejects.toThrow('sole orchestrator');
+  });
+});
+
 describe('fleet init wizard', () => {
   let cleanup: string | undefined;
 
@@ -2421,3 +2903,33 @@ describe('fleet init wizard', () => {
     expect(content).toContain('name: coder0');
   });
 });
+
+describe('fleet ps — heartbeat path resolution', () => {
+  const savedRunDir = process.env.MOSAIC_HEARTBEAT_RUN_DIR;
+  const savedHome = process.env.MOSAIC_HOME;
+  afterEach(() => {
+    if (savedRunDir === undefined) delete process.env.MOSAIC_HEARTBEAT_RUN_DIR;
+    else process.env.MOSAIC_HEARTBEAT_RUN_DIR = savedRunDir;
+    if (savedHome === undefined) delete process.env.MOSAIC_HOME;
+    else process.env.MOSAIC_HOME = savedHome;
+  });
+
+  it('honors MOSAIC_HEARTBEAT_RUN_DIR (matches the writer sidecar override)', () => {
+    process.env.MOSAIC_HEARTBEAT_RUN_DIR = '/run/hb';
+    expect(heartbeatPath('agent-x', '/any/home')).toBe(join('/run/hb', 'agent-x.hb'));
+  });
+
+  it('honors MOSAIC_HOME when no explicit mosaicHome is given', () => {
+    delete process.env.MOSAIC_HEARTBEAT_RUN_DIR;
+    process.env.MOSAIC_HOME = '/custom/mhome';
+    expect(heartbeatPath('agent-y')).toBe(join('/custom/mhome', 'fleet', 'run', 'agent-y.hb'));
+  });
+
+  it('falls back to <mosaicHome>/fleet/run by default', () => {
+    delete process.env.MOSAIC_HEARTBEAT_RUN_DIR;
+    delete process.env.MOSAIC_HOME;
+    expect(heartbeatPath('agent-z', '/home/u/.config/mosaic')).toBe(
+      join('/home/u/.config/mosaic', 'fleet', 'run', 'agent-z.hb'),
+    );
+  });
+});

packages/mosaic/src/commands/fleet.ts

@@ -1,5 +1,5 @@
 import { constants } from 'node:fs';
-import { access, chmod, copyFile, mkdir, readFile, writeFile } from 'node:fs/promises';
+import { access, chmod, copyFile, mkdir, readFile, unlink, writeFile } from 'node:fs/promises';
 import { homedir, hostname, userInfo } from 'node:os';
 import { dirname, join, resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
@@ -152,13 +152,16 @@ export function resolveFleetPaths(mosaicHome = defaultMosaicHome()): FleetPaths
 }
 
 function defaultMosaicHome(): string {
-  return join(homedir(), '.config', 'mosaic');
+  // Honor MOSAIC_HOME so the reader matches the writer sidecar (and the launcher),
+  // even when MOSAIC_HOME is set in the shell without an explicit --mosaic-home flag.
+  return process.env.MOSAIC_HOME ?? join(homedir(), '.config', 'mosaic');
 }
 
 function assertDefaultMosaicHomeForSystemd(mosaicHome: string): void {
-  if (resolve(mosaicHome) !== resolve(defaultMosaicHome())) {
+  const literalHome = join(homedir(), '.config', 'mosaic');
+  if (resolve(mosaicHome) !== resolve(literalHome)) {
     throw new Error(
-      `install-systemd only supports the default Mosaic home (${defaultMosaicHome()}) because the user systemd units use %h/.config/mosaic paths.`,
+      `install-systemd only supports the default Mosaic home (${literalHome}) because the user systemd units use %h/.config/mosaic paths.`,
     );
   }
 }
@@ -387,6 +390,8 @@ export interface HeartbeatInfo {
   /** healthy | stale | unknown */
   health: 'healthy' | 'stale' | 'unknown';
   ageMs: number | null;
+  /** Model id the runtime self-reported in its heartbeat (native HB only), else null. */
+  model: string | null;
 }
 
 export interface AgentPsRow {
@@ -475,7 +480,10 @@ export function parseTmuxListSessions(output: string): string[] {
  * Returns the heartbeat file path for an agent.
  */
 export function heartbeatPath(agentName: string, mosaicHome = defaultMosaicHome()): string {
-  return join(mosaicHome, 'fleet', 'run', `${agentName}.hb`);
+  // Honor MOSAIC_HEARTBEAT_RUN_DIR (the writer sidecar's override); otherwise the
+  // canonical <mosaicHome>/fleet/run. Keeps reader and writer on the same path.
+  const runDir = process.env.MOSAIC_HEARTBEAT_RUN_DIR ?? join(mosaicHome, 'fleet', 'run');
+  return join(runDir, `${agentName}.hb`);
 }
 
 /**
@@ -484,15 +492,17 @@ export function heartbeatPath(agentName: string, mosaicHome = defaultMosaicHome(
  *   ts=<iso8601>
  *   pid=<pid>
  *   status=<ok|busy>
+ *   model=<model-id>   (optional — native runtime heartbeats self-report it)
  */
 export function parseHeartbeat(content: string | null, nowMs = Date.now()): HeartbeatInfo {
   if (content === null) {
-    return { ts: null, pid: null, status: null, health: 'unknown', ageMs: null };
+    return { ts: null, pid: null, status: null, health: 'unknown', ageMs: null, model: null };
   }
   const lines = content.split('\n');
   let ts: Date | null = null;
   let pid: number | null = null;
   let status: 'ok' | 'busy' | null = null;
+  let model: string | null = null;
   for (const line of lines) {
     const [key, ...rest] = line.split('=');
     const val = rest.join('=').trim();
@@ -504,6 +514,8 @@ export function parseHeartbeat(content: string | null, nowMs = Date.now()): Hear
       if (Number.isFinite(n)) pid = n;
     } else if (key === 'status' && (val === 'ok' || val === 'busy')) {
       status = val;
+    } else if (key === 'model' && val) {
+      model = val;
     }
   }
   const thresholdMs = heartbeatIntervalMs() * HEARTBEAT_HEALTHY_MULTIPLIER;
@@ -513,7 +525,7 @@ export function parseHeartbeat(content: string | null, nowMs = Date.now()): Hear
     ageMs = nowMs - ts.getTime();
     health = ageMs <= thresholdMs ? 'healthy' : 'stale';
   }
-  return { ts, pid, status, health, ageMs };
+  return { ts, pid, status, health, ageMs, model };
 }
 
 /**
@@ -1117,6 +1129,7 @@ export function registerFleetCommand(program: Command, deps: FleetCommandDeps =
         'PID'.padEnd(8),
         'IDLE'.padEnd(8),
         'HB'.padEnd(12),
+        'MODEL'.padEnd(22),
         'FLAGS',
       ].join(' ');
       console.log(header);
@@ -1131,6 +1144,7 @@ export function registerFleetCommand(program: Command, deps: FleetCommandDeps =
           row.heartbeat.ageMs !== null
             ? `${Math.round(row.heartbeat.ageMs / 1000)}s/${row.heartbeat.health}`
             : `unknown`;
+        const model = row.heartbeat.model ?? '-';
         const flags: string[] = [];
         if (!row.managed) flags.push('UNMANAGED');
         if (row.driftFlag) flags.push('DRIFT');
@@ -1147,12 +1161,119 @@ export function registerFleetCommand(program: Command, deps: FleetCommandDeps =
             pid.padEnd(8),
             idle.padEnd(8),
             hbAge.padEnd(12),
+            model.padEnd(22),
             flags.join(','),
           ].join(' '),
         );
       }
     });
 
+  cmd
+    .command('add <name>')
+    .description('Add a new agent to the fleet roster and optionally start it')
+    .requiredOption('--runtime <runtime>', `Agent runtime (${VALID_FLEET_RUNTIMES.join(', ')})`)
+    .requiredOption('--class <class>', 'Agent class (e.g. worker, orchestrator, canary)')
+    .option('--model <hint>', 'Model hint for the agent')
+    .option('--working-dir <path>', 'Working directory for the agent')
+    .option('--no-start', 'Skip starting the agent after adding')
+    .action(
+      async (
+        name: string,
+        opts: {
+          runtime: string;
+          class: string;
+          model?: string;
+          workingDir?: string;
+          start: boolean;
+        },
+      ) => {
+        if (!VALID_FLEET_RUNTIMES.includes(opts.runtime)) {
+          throw new Error(
+            `Invalid runtime "${opts.runtime}". Valid runtimes: ${VALID_FLEET_RUNTIMES.join(', ')}.`,
+          );
+        }
+        const commandOpts = cmd.opts<{ mosaicHome: string; roster?: string }>();
+        const activePaths = resolveFleetPaths(commandOpts.mosaicHome);
+        const rosterPath = await resolveRosterPath(commandOpts.mosaicHome, commandOpts.roster);
+        const roster = await loadFleetRoster(rosterPath);
+
+        const newAgent: FleetAgent = {
+          name,
+          runtime: opts.runtime,
+          className: opts.class,
+          ...(opts.workingDir !== undefined && { workingDirectory: opts.workingDir }),
+          ...(opts.model !== undefined && { modelHint: opts.model }),
+        };
+
+        const updatedRoster = addAgentToRoster(roster, newAgent);
+        await writeFile(rosterPath, serializeRosterToYaml(updatedRoster));
+
+        const envPath = join(activePaths.agentEnvDir, `${name}.env`);
+        const existingEnv = (await canRead(envPath)) ? await readFile(envPath, 'utf8') : undefined;
+        await mkdir(activePaths.agentEnvDir, { recursive: true });
+        await writeFile(
+          envPath,
+          mergeAgentEnv(generateAgentEnv(updatedRoster, newAgent), existingEnv),
+        );
+
+        console.log(`Added ${name} (${opts.runtime}/${opts.class}) to the fleet.`);
+
+        if (opts.start !== false) {
+          await runChecked(runner, buildFleetServiceCommand('start', name));
+          console.log(`Started mosaic-agent@${name}.service.`);
+        } else {
+          console.log(`Agent queued (--no-start); run: mosaic fleet start ${name}`);
+        }
+      },
+    );
+
+  cmd
+    .command('remove <name>')
+    .description('Remove an agent from the fleet roster')
+    .option('--keep-files', 'Skip deleting env and heartbeat files')
+    .action(async (name: string, opts: { keepFiles?: boolean }) => {
+      const commandOpts = cmd.opts<{ mosaicHome: string; roster?: string }>();
+      const activePaths = resolveFleetPaths(commandOpts.mosaicHome);
+      const rosterPath = await resolveRosterPath(commandOpts.mosaicHome, commandOpts.roster);
+      const roster = await loadFleetRoster(rosterPath);
+
+      // Guard: throws if removing leaves 0 orchestrators or agent not in roster
+      const updatedRoster = removeAgentFromRoster(roster, name);
+
+      // Stop agent (non-fatal)
+      try {
+        const stopResult = await runner(...splitCommand(buildFleetServiceCommand('stop', name)));
+        if (stopResult.exitCode !== 0) {
+          process.stderr.write(
+            `Warning: could not stop mosaic-agent@${name}.service: ${stopResult.stderr || stopResult.stdout || 'non-zero exit'}\n`,
+          );
+        }
+      } catch (err) {
+        process.stderr.write(
+          `Warning: stop command failed for ${name}: ${err instanceof Error ? err.message : String(err)}\n`,
+        );
+      }
+
+      // Write updated roster
+      await writeFile(rosterPath, serializeRosterToYaml(updatedRoster));
+
+      // Delete env and heartbeat files (best-effort, non-fatal)
+      if (!opts.keepFiles) {
+        try {
+          await unlink(join(activePaths.agentEnvDir, `${name}.env`));
+        } catch {
+          // best-effort
+        }
+        try {
+          await unlink(heartbeatPath(name, activePaths.mosaicHome));
+        } catch {
+          // best-effort
+        }
+      }
+
+      console.log(`Removed ${name} from the fleet.`);
+    });
+
   return cmd;
 }
 
@@ -1332,15 +1453,19 @@ export function registerFleetAgentCommands(
 
       await runChecked(runner, buildAgentWatchCreateViewerCommand(agent, viewerName, socketName));
 
+      let exitCode = 0;
+      try {
         const [bin, args] = splitCommand(buildAgentWatchAttachCommand(viewerName, socketName));
-      const exitCode = await iRunner(bin, args);
+        exitCode = await iRunner(bin, args);
-
+      } finally {
-      // Best-effort cleanup of the viewer session regardless of how the user detached.
+        // ALWAYS clean up the viewer session — even if attach threw or the process was
-      // Errors here are intentionally suppressed — the agent session is unaffected.
+        // interrupted — so stale grouped *-watch-* sessions never accumulate. Errors here
+        // are intentionally suppressed; the agent session is unaffected.
         const killResult = await runner(
           ...splitCommand(buildAgentWatchKillViewerCommand(viewerName, socketName)),
         );
-      void killResult; // result is intentionally ignored
+        void killResult;
+      }
 
       if (exitCode !== 0) {
         process.exitCode = exitCode;
@@ -1769,6 +1894,105 @@ export function countOrchestrators(roster: FleetRoster): number {
   return roster.agents.filter((a) => a.className === 'orchestrator').length;
 }
 
+/** Valid runtime identifiers for fleet agents. */
+export const VALID_FLEET_RUNTIMES: readonly string[] = [
+  'pi',
+  'claude',
+  'codex',
+  'opencode',
+  'dogfood',
+];
+
+/**
+ * Add a new agent to a fleet roster (immutable — returns a new FleetRoster).
+ * Throws on invalid name, duplicate name.
+ */
+export function addAgentToRoster(roster: FleetRoster, agent: FleetAgent): FleetRoster {
+  if (!agent.name || !/^[A-Za-z0-9_.-]+$/.test(agent.name)) {
+    throw new Error(`Invalid fleet agent name: ${agent.name || '<empty>'}`);
+  }
+  if (roster.agents.some((a) => a.name === agent.name)) {
+    throw new Error(`Agent "${agent.name}" already exists in the fleet roster.`);
+  }
+  return {
+    ...roster,
+    agents: [...roster.agents, agent],
+  };
+}
+
+/**
+ * Remove an agent from a fleet roster (immutable — returns a new FleetRoster).
+ * Throws if the agent is not found, or if removal would leave zero orchestrators.
+ */
+export function removeAgentFromRoster(roster: FleetRoster, name: string): FleetRoster {
+  const agent = roster.agents.find((a) => a.name === name);
+  if (!agent) {
+    throw new Error(`Agent "${name}" is not in the fleet roster.`);
+  }
+  const remaining = roster.agents.filter((a) => a.name !== name);
+  const remainingOrchCount = remaining.filter((a) => a.className === 'orchestrator').length;
+  if (remainingOrchCount === 0 && agent.className === 'orchestrator') {
+    throw new Error(
+      `Cannot remove agent "${name}": it is the sole orchestrator. Add another orchestrator first (R5).`,
+    );
+  }
+  return {
+    ...roster,
+    agents: remaining,
+  };
+}
+
+/**
+ * Serialize a FleetRoster to YAML text (snake_case keys).
+ * The output is parseable by loadFleetRoster.
+ */
+export function serializeRosterToYaml(roster: FleetRoster): string {
+  const agents = roster.agents.map((agent) => {
+    const raw: Record<string, unknown> = {
+      name: agent.name,
+      runtime: agent.runtime,
+      class: agent.className,
+    };
+    if (agent.workingDirectory !== undefined) {
+      raw['working_directory'] = agent.workingDirectory;
+    }
+    if (agent.modelHint !== undefined) {
+      raw['model_hint'] = agent.modelHint;
+    }
+    if (agent.persistentPersona !== undefined) {
+      raw['persistent_persona'] = agent.persistentPersona;
+    }
+    if (agent.resetBetweenTasks !== undefined) {
+      raw['reset_between_tasks'] = agent.resetBetweenTasks;
+    }
+    if (agent.kickstartTemplate !== undefined) {
+      raw['kickstart_template'] = agent.kickstartTemplate;
+    }
+    return raw;
+  });
+
+  const runtimes: Record<string, { reset_command: string }> = {};
+  for (const [runtime, config] of Object.entries(roster.runtimes)) {
+    runtimes[runtime] = { reset_command: config.resetCommand };
+  }
+
+  const raw: Record<string, unknown> = {
+    version: roster.version,
+    transport: roster.transport,
+    tmux: {
+      socket_name: roster.tmux.socketName,
+      holder_session: roster.tmux.holderSession,
+    },
+    defaults: {
+      working_directory: roster.defaults.workingDirectory,
+    },
+    runtimes,
+    agents,
+  };
+
+  return YAML.stringify(raw);
+}
+
 /**
  * Prompt interactively for a fleet profile via stdin readline.
  * AI-free: no LLM calls — pure readline menu.