feat(roles): M18 seat-role progressive capability restriction (#45)

Role contracts (roles/<role>.json): roleVersion, name bound to filename,
tools ceiling (subset of pi built-ins), network declared (none|api-only|
open; enforced when network policy lands). Strict schema, fail closed -
a non-role document refuses resolution.

mosaic-task.mjs resolve-role: config-free contract validation, emits
MOSAIC_ROLE_TOOLS / MOSAIC_ROLE_NETWORK.

agent.sh: a declared role binds to its contract. Missing/invalid contract
refuses the launch (exit 2, names the role - the under-equipped-seat
failure mode, mirroring M17 skills). Effective tools = ceiling ∩ requested
(CLI --tools or agent.json caps); no request -> ceiling stands; narrowing
and tool-free outcomes loud on stderr. Adapters unchanged; headless M9
chain (mission ∩ task) untouched.

Ships roles/researcher.json (existing seat declares the role; without the
contract the fail-closed gate would refuse its launch).

Task suite 74 -> 88: contract resolution, wrong-kind/name/network/
duplicate/unsupported/missing refusals, ceiling narrowing E2E (mock
adapter), tool-free E2E, missing-contract refusal. Test-authoring
correction recorded in BUILD-LOG (a check that registered on one path
only, caught by count arithmetic).

Suites 24/88/14/17 + verify green.
This commit is contained in:
2026-09-03 17:25:17 -05:00
parent bf56583a49
commit ca8135d70c
9 changed files with 172 additions and 6 deletions
+1 -1
View File
@@ -68,7 +68,7 @@ green at every step. Not production software — a proven foundation.
`scripts/bootstrap.sh` (idempotent) · `build.sh` · `hello.sh` ·
`verify.sh` · `run-task.sh run <task.json>` · `release.sh
package|activate|rollback|status` · `reset.sh` (**danger**: wipes the data
root; triple-safety-checked) · `mosaic-task.mjs validate|run|show|list|retry|prune` ·
root; triple-safety-checked) · `mosaic-task.mjs validate|run|show|list|retry|prune|resolve-role` ·
`agent.sh <name>` (interactive TUI agent) ·
suites: `test-config.sh`, `test-task.sh`, `test-release.sh`,
`test-conductor.sh`.
+32
View File
@@ -467,3 +467,35 @@ Release self-determination live: the system aligns itself to RELEASE without man
The conductor loop is proven end to end on a live seat; TOOLS.md covers the
tools/ tree; suites green at every gate.
## Phase 23: M18 — seat-role progressive capability restriction
- Role contracts: `roles/<role>.json` — roleVersion, name (must match
filename), tools ceiling (subset of pi built-ins), network declared
(none | api-only | open; enforcement is a later milestone). Strict
schema, fail closed; a non-role document (e.g. conductor-policy.json)
refuses.
- `mosaic-task.mjs resolve-role <file>`: config-free validation, emits
MOSAIC_ROLE_TOOLS / MOSAIC_ROLE_NETWORK for agent.sh to consume.
- `agent.sh`: a declared role binds to its contract. Missing/invalid
contract refuses the launch (exit 2, names the role) — the
under-equipped-seat failure mode, mirroring M17 skills. Effective tools
= ceiling ∩ requested (CLI --tools or agent.json caps); no request →
ceiling stands; narrowing and tool-free outcomes are loud on stderr.
Adapters unchanged (MOSAIC_TOOLS carries the effective set); headless
M9 chain (mission ∩ task) untouched.
- Ships roles/researcher.json — the existing researcher seat declares the
role; without the contract the fail-closed gate would refuse its launch.
- Suite additions (14 cases, task suite 74 → 88): contract resolution,
wrong-kind/name/network/duplicate/unsupported/missing refusals, seat
narrowing E2E via mock adapter, tool-free E2E, missing-contract refusal.
- Test-authoring correction: the first version of the missing-contract
case registered its check only on the failure path (a `|| RC=$?` chain
swallowed it on success); caught by count arithmetic (74 + 15 ≠ 88),
restructured so the case always registers.
## Result (M18)
Seat roles are ceilings, not labels: the M15 role field now resolves to a
versioned contract that seats cannot escalate past. Suites 24/88/14/17 +
verify green.
+1
View File
@@ -11,3 +11,4 @@ are never rewritten or removed; corrections are new entries.
| 2026-09-03 | assistant | ms-communications skill: inter-agent messaging protocol consolidated from tools/tmux/README.md and agent-send.sh (channel, preamble grammar, flip-on-reply, triage classes, etiquette, receiving protocol, delivery mechanics) | skills/ms-communications/SKILL.md created; unslop-check clean |
| 2026-09-03 | assistant (conductor) + ms-test collaborator (worker, glm-5.3-flash) | Conductor-loop calibration (#43): decompose → dispatch via agent-send.sh → receipt → line-by-line diff review (claims verified vs tool source) → suite-gated integration; CURRENT.md staleness corrected (M16/M17 late-logged, next action → M18) | docs/TOOLS.md tools/ section + suite-count fix; issue #43 closed; suites 24/74/14/17 + verify green |
| 2026-09-03 | owner + assistant (conductor) + ms-test collaborator | Skill revisions adjudicated (#44): ms-communications integrated as-authored (owner preamble restructure + collaborator delivery-discipline hunks); ms-conductor collaborator redraft integrated with conductor remediation (step 3 refusal-vs-outage distinction; preserves owner's outage-dispatch intent inside fail-closed canon); TOOLS.md gains release.sh ensure row | skills/ms-communications/SKILL.md, skills/ms-conductor/SKILL.md, docs/TOOLS.md; suites 24/74/14/17 + verify green; unslop clean ×3 |
| 2026-09-03 | assistant (conductor) | M18 seat-role progressive capability restriction (#45): roles/<role>.json contracts (strict schema, name-filename binding, network declared), mosaic-task.mjs resolve-role, agent.sh ceiling intersection with fail-closed refusals, roles/researcher.json shipped, 14 suite cases (task 74 → 88) | scripts/mosaic-task.mjs, scripts/agent.sh, scripts/test-task.sh, roles/researcher.json, docs; suites 24/88/14/17 + verify green |
+7 -1
View File
@@ -30,6 +30,7 @@ and changes nothing.
| `node scripts/mosaic-task.mjs list` | List runs | task/workspace/session columns |
| `node scripts/mosaic-task.mjs retry <runId>` | Re-execute a run's snapshot | New run dir; `retriedFrom` lineage recorded |
| `node scripts/mosaic-task.mjs prune [--keep=N] [--yes]` | Retention | Dry-run default; receipt in `runs/.pruned.log` |
| `node scripts/mosaic-task.mjs resolve-role <roleFile>` | Validate a role contract | Prints `MOSAIC_ROLE_TOOLS` / `MOSAIC_ROLE_NETWORK`; config-free |
Task fields: `prompt` (required), `mission` (path), `expectExact`,
`timeoutSeconds` (5600), `workspace` (`:run` or named), `capabilities.tools`
@@ -48,6 +49,11 @@ Launches an interactive pi TUI inside the container with the four immutable
contracts + optional mission + agent identity as its system prompt,
persistent named session, optional workspace. Exit with `/quit`.
A seat role (`agent.json` `role`) binds to `roles/<role>.json` (M18): the
contract's tools are a ceiling the seat definition or `--tools` may narrow,
never escalate past. Missing/invalid contract refuses the launch; empty
intersection = loud tool-free seat.
## Release
| Command | Purpose | Notes |
@@ -75,7 +81,7 @@ Push is never automatic.
|---|---|---|
| `scripts/reset.sh` | Delete the data root | Triple-safety-checked (path, symlink, ownership marker) |
| `scripts/test-config.sh` | Config selftests (no Docker) | 24 cases |
| `scripts/test-task.sh` | Task selftests + live cases | 74 cases |
| `scripts/test-task.sh` | Task selftests + live cases | 88 cases |
| `scripts/test-release.sh` | Release selftests | 14 cases |
| `scripts/test-conductor.sh` | Auto-apply selftests (sandboxed) | 17 cases |
| `scripts/gitea-api.sh <METHOD> <path> [body]` | Gitea API helper | Token never on argv/stdout |
+4 -3
View File
@@ -7,12 +7,11 @@ update this file to the next action). No ambiguity, no re-planning.
## Next action
M18seat-role progressive capability restriction (role ceilings ∩ mission ∩ task) — say "next" to execute.
M19harness auth tooling (pi checkpoint: auth status, provider stacking, multi-account investigation) — say "next" to execute.
## Queue (ordered per docs/plans/ROADMAP.md)
1. M19 — harness auth tooling (pi checkpoint: auth status, provider stacking, multi-account investigation)
2. Deferred by owner: CI runners (Gitea hardware slow); second real adapter; push automation
1. Deferred by owner: CI runners (Gitea hardware slow); second real adapter; push automation
## Rules
@@ -50,3 +49,5 @@ git history + Gitea issues.
- 2026-09-03 — M16 release self-determination (#38; `release.sh ensure` at launch, drift warnings, recursion guard) — logged late: CURRENT.md had gone stale while M16/M17 shipped; ground truth = git history
- 2026-09-03 — M17 skill lifecycle + ms-* skill set completion (#40#42; skill-lifecycle-v1); release 0.0.12 packaged, health-gated active — logged late, same staleness correction
- 2026-09-03 — conductor-loop calibration with live collaborator (#43): dispatch via agent-send.sh → receipt → line-by-line diff review → suite-gated integration; docs/TOOLS.md gains Tools (host-side) section + corrected suite counts
- 2026-09-03 — skill revisions adjudicated (#44): ms-communications integrated as-authored; ms-conductor redraft + conductor remediation (refusal vs outage); TOOLS.md release.sh ensure row
- 2026-09-03 — M18 seat-role progressive capability restriction (#45; roles resolve to contracts, ceiling ∩ seat grant, fail-closed refusals, roles/researcher.json); task suite 74 → 88
+6
View File
@@ -0,0 +1,6 @@
{
"roleVersion": 1,
"name": "researcher",
"tools": ["read", "grep", "find", "ls", "bash"],
"network": "none"
}
+23
View File
@@ -109,6 +109,29 @@ export MOSAIC_AGENT_NAME="$NAME"
[ -n "$ROLE" ] && export MOSAIC_AGENT_ROLE="$ROLE"
export MOSAIC_INTERACTIVE=1
if [ -z "$TOOLS" ] && [ -n "$DEFCAPS" ]; then TOOLS="$DEFCAPS"; fi
# Role ceiling (M18): a declared role binds to roles/<role>.json; its tools
# are a ceiling that the seat definition or CLI may narrow, never escalate
# past. A missing or invalid contract refuses the launch - a declared role
# that resolves to nothing is the under-equipped-seat failure mode.
if [ -n "$ROLE" ]; then
ROLES_DIR="${MOSAIC_ROLES_DIR:-roles}"
ROLE_FILE="$ROLES_DIR/$ROLE.json"
[ -r "$ROLE_FILE" ] || { echo "agent: role '$ROLE' is declared but has no contract: $ROLE_FILE" >&2; exit 2; }
ROLE_OUT="$(node scripts/mosaic-task.mjs resolve-role "$ROLE_FILE")" || { echo "agent: invalid role contract: $ROLE_FILE" >&2; exit 2; }
ROLE_CEILING="$(printf '%s\n' "$ROLE_OUT" | sed -n 's/^MOSAIC_ROLE_TOOLS=//p')"
if [ -n "$TOOLS" ]; then
REQUESTED_TOOLS="$TOOLS"
TOOLS="$(node -e 'const c=process.argv[1].split(",").filter(Boolean);const r=process.argv[2].split(",").filter(Boolean);process.stdout.write(r.filter(t=>c.includes(t)).join(","))' "$ROLE_CEILING" "$REQUESTED_TOOLS")"
if [ -z "$TOOLS" ]; then
echo "agent: capability policy: role '$ROLE' ceiling and requested tools have nothing in common -> tool-free seat" >&2
elif [ "$TOOLS" != "$REQUESTED_TOOLS" ]; then
echo "agent: capability policy: role '$ROLE' ceiling narrowed tools -> $TOOLS" >&2
fi
else
TOOLS="$ROLE_CEILING"
fi
fi
export MOSAIC_TOOLS="${TOOLS:+$TOOLS}"
# Skills (M17): seat definition may declare skill names; each must be
+35 -1
View File
@@ -245,6 +245,33 @@ function validateTask(document, file) {
};
}
// Role contract (M18): seat-declared role authority. The tools array is a
// ceiling — seats may narrow it, never escalate past it. network is declared
// now and enforced when network policy lands. Strict schema: unknown keys
// refuse, name must match the filename, wrong document kind refuses.
function validateRole(document, file) {
rejectUnknownKeys(document, ["roleVersion", "name", "tools", "network"], "role");
if (document.roleVersion !== 1) fail(2, 'role "roleVersion" must be 1');
validateId(document.name, "role name");
const base = path.basename(file).replace(/\.json$/, "");
if (document.name !== base) fail(2, `role "name" (${document.name}) must match its filename (${base}.json)`);
if (!Array.isArray(document.tools) || document.tools.length === 0) {
fail(2, 'role "tools" must be a non-empty array of tool names');
}
const seen = new Set();
for (const tool of document.tools) {
if (!SUPPORTED_TOOLS.includes(tool)) {
fail(2, `unsupported tool: ${JSON.stringify(tool)} (supported: ${SUPPORTED_TOOLS.join(", ")})`);
}
if (seen.has(tool)) fail(2, `duplicate tool in role tools: ${tool}`);
seen.add(tool);
}
if (document.network !== undefined && !["none", "api-only", "open"].includes(document.network)) {
fail(2, 'role "network" must be one of: none, api-only, open');
}
return { roleVersion: 1, name: document.name, tools: [...seen], network: document.network ?? "none" };
}
function loadConfig() {
const proc = spawnSync(process.execPath, [path.join(PROJECT_ROOT, "scripts", "mosaic-config.mjs"), "validate"], {
cwd: PROJECT_ROOT,
@@ -640,10 +667,17 @@ switch (operation) {
case "prune":
pruneRuns(process.argv.slice(3));
break;
case "resolve-role": {
if (!target) fail(4, "usage: mosaic-task.mjs resolve-role <roleFile>");
const file = path.resolve(target);
const role = validateRole(readJsonFile(file, "role contract"), file);
process.stdout.write(`MOSAIC_ROLE_TOOLS=${role.tools.join(",")}\nMOSAIC_ROLE_NETWORK=${role.network}\n`);
process.exit(0);
}
case "retry":
if (!target) fail(4, "usage: mosaic-task.mjs retry <runId>");
retryRun(target);
break;
default:
fail(4, `unknown operation: ${JSON.stringify(operation ?? "")} (expected validate | run | show | list | retry | prune)`);
fail(4, `unknown operation: ${JSON.stringify(operation ?? "")} (expected validate | run | show | list | retry | prune | resolve-role)`);
}
+63
View File
@@ -236,6 +236,69 @@ EOF
grep -q '^MOSAIC_SKILLS=/var/lib/mosaic/skills-enabled/ms-tools$' "$SANDBOX/seat-stderr.txt" 2>/dev/null \
&& check "skill path delivered to adapter" 0 || check "skill path delivered to adapter" 1
# seat role ceiling (M18): role contracts cap seat capabilities.
# Contract validation is CLI-exercised (no docker); the seat wiring gets
# two mock-adapter launches (narrowing, tool-free) plus one pre-docker
# refusal (missing contract).
ROLE_OUT="$(node scripts/mosaic-task.mjs resolve-role roles/researcher.json 2>/dev/null)" \
&& printf '%s\n' "$ROLE_OUT" | grep -q '^MOSAIC_ROLE_TOOLS=read,grep,find,ls,bash$' \
&& printf '%s\n' "$ROLE_OUT" | grep -q '^MOSAIC_ROLE_NETWORK=none$' \
&& check "shipped researcher contract resolves (ceiling + network)" 0 || check "shipped researcher contract resolves (ceiling + network)" 1
expect_exit "resolve-role refuses a non-role document (conductor policy)" 2 -- \
node scripts/mosaic-task.mjs resolve-role roles/conductor-policy.json
mkdir -p "$SANDBOX/roles" "$SANDBOX/agents/roleseat" "$SANDBOX/no-roles"
printf '{"roleVersion":1,"name":"other","tools":["read"]}' > "$SANDBOX/roles/mismatch.json"
expect_exit "resolve-role refuses name/filename mismatch" 2 -- \
node scripts/mosaic-task.mjs resolve-role "$SANDBOX/roles/mismatch.json"
printf '{"roleVersion":1,"name":"badnet","tools":["read"],"network":"everywhere"}' > "$SANDBOX/roles/badnet.json"
expect_exit "resolve-role refuses unknown network declaration" 2 -- \
node scripts/mosaic-task.mjs resolve-role "$SANDBOX/roles/badnet.json"
printf '{"roleVersion":1,"name":"dupe","tools":["read","read"]}' > "$SANDBOX/roles/dupe.json"
expect_exit "resolve-role refuses duplicate role tool" 2 -- \
node scripts/mosaic-task.mjs resolve-role "$SANDBOX/roles/dupe.json"
printf '{"roleVersion":1,"name":"aliens","tools":["read","render3d"]}' > "$SANDBOX/roles/aliens.json"
expect_exit "resolve-role refuses unsupported tool" 2 -- \
node scripts/mosaic-task.mjs resolve-role "$SANDBOX/roles/aliens.json"
expect_exit "resolve-role refuses missing contract file" 4 -- \
node scripts/mosaic-task.mjs resolve-role "$SANDBOX/roles/absent.json"
printf '# SOUL - roleseat\n\nVerifies before claiming.\n' > "$SANDBOX/agents/roleseat/SOUL.md"
printf '{"agentVersion":1,"name":"roleseat","role":"analyst","capabilities":{"tools":["read","write","bash"]}}' > "$SANDBOX/agents/roleseat/agent.json"
printf '{"roleVersion":1,"name":"analyst","tools":["read","grep","bash"],"network":"none"}' > "$SANDBOX/roles/analyst.json"
env MOSAIC_CONFIG="$SANDBOX/mock-config.json" MOSAIC_MOCK_RESPONSE=MOCKED \
MOSAIC_AGENTS_DIR="$SANDBOX/agents" MOSAIC_ROLES_DIR="$SANDBOX/roles" \
scripts/agent.sh roleseat </dev/null >"$SANDBOX/roleseat-stdout.txt" 2>"$SANDBOX/roleseat-stderr.txt"
RC=$?
if [ "$RC" -eq 0 ]; then
PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET} seat launches under role ceiling"
else
FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} seat launches under role ceiling (exit $RC)" >&2
echo "ROLESEAT stderr:" >&2; cat "$SANDBOX/roleseat-stderr.txt" >&2
fi
grep -q '^MOSAIC_TOOLS=read,bash$' "$SANDBOX/roleseat-stderr.txt" 2>/dev/null \
&& check "role ceiling narrows seat tools (read,write,bash -> read,bash)" 0 || check "role ceiling narrows seat tools (read,write,bash -> read,bash)" 1
grep -q "role 'analyst' ceiling narrowed tools -> read,bash" "$SANDBOX/roleseat-stderr.txt" 2>/dev/null \
&& check "narrowing recorded loudly" 0 || check "narrowing recorded loudly" 1
env MOSAIC_CONFIG="$SANDBOX/mock-config.json" MOSAIC_MOCK_RESPONSE=MOCKED \
MOSAIC_AGENTS_DIR="$SANDBOX/agents" MOSAIC_ROLES_DIR="$SANDBOX/no-roles" \
scripts/agent.sh roleseat </dev/null >"$SANDBOX/roleseat-stdout.txt" 2>"$SANDBOX/roleseat-stderr.txt"
RC=$?
if [ "$RC" -eq 2 ]; then
PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET} missing role contract refuses launch (exit 2)"
else
FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} missing role contract refuses launch (exit $RC)" >&2
echo "ROLESEAT stderr:" >&2; cat "$SANDBOX/roleseat-stderr.txt" >&2
fi
grep -q "role 'analyst' is declared but has no contract" "$SANDBOX/roleseat-stderr.txt" 2>/dev/null \
&& check "missing-contract refusal names the role" 0 || check "missing-contract refusal names the role" 1
printf '{"agentVersion":1,"name":"roleseat","role":"analyst","capabilities":{"tools":["write","edit"]}}' > "$SANDBOX/agents/roleseat/agent.json"
env MOSAIC_CONFIG="$SANDBOX/mock-config.json" MOSAIC_MOCK_RESPONSE=MOCKED \
MOSAIC_AGENTS_DIR="$SANDBOX/agents" MOSAIC_ROLES_DIR="$SANDBOX/roles" \
scripts/agent.sh roleseat </dev/null >"$SANDBOX/roleseat-stdout.txt" 2>"$SANDBOX/roleseat-stderr.txt"
grep -q '^MOSAIC_TOOLS=$' "$SANDBOX/roleseat-stderr.txt" 2>/dev/null \
&& check "empty ceiling intersection -> tool-free seat" 0 || check "empty ceiling intersection -> tool-free seat" 1
grep -q "nothing in common -> tool-free seat" "$SANDBOX/roleseat-stderr.txt" 2>/dev/null \
&& check "tool-free outcome recorded loudly" 0 || check "tool-free outcome recorded loudly" 1
# capability policy (M9): least-privilege intersection
POL="$SANDBOX/data/workspaces"; mkdir -p "$POL"
pol_run() { # missionTools(ABSENT|json) taskTools(ABSENT|json) -> stderr MOSAIC_TOOLS value