Merge M9: mission-level capability policy

Closes #30

docs/plans/CURRENT.md

@@ -7,13 +7,14 @@ update this file to the next action). No ambiguity, no re-planning.
 
 ## Next action
 
-Gitea issue: (none queued — awaiting owner direction)
+Owner decision on M9: mission-level capability policy (design sketch: missions may declare default tool sets; tasks inherit unless overridden; conductor validates the merge). Say "next" to proceed or name a different target.
 
 ## Queue (ordered, not started)
 
-1. Owner review of M5–M8 (workspaces, capabilities, sessions, conductor loop)
-2. M8 follow-up: retry lineage (retriedFrom field) + relative-mission-path retry limitation
-3. M9 candidates: mission-level capability policy; run-record retention; decision on auto-apply policy for worker patches
+1. M9: mission-level capability policy
+2. Run-record retention/pruning policy
+3. Second real adapter (parked — owner focused on Pi)
+4. Auto-apply policy for worker patches (deferred until capability policy exists)
 
 ## Rules
 
@@ -31,3 +32,5 @@ Gitea issue: (none queued — awaiting owner direction)
 - 2026-09-03 — M6 named sessions + resume (#22, #23) — merged, teach/recall verified
 - 2026-09-03 — M7 run inspection + release 0.0.6 (#24) — merged, activated
 - 2026-09-03 — M8 conductor loop + worker-built retry (#25, #26, #27) — merged; worker authored retry in 2 refinement rounds, conductor fixed a 3-line interpolation rename; live retry verified
+- 2026-09-03 — retry lineage + relative mission resolution (#28) — merged, 36/24/14 suites + verify green
+- 2026-09-03 — M8 conductor loop + worker-built retry (#25, #26, #27) — merged; worker authored retry in 2 refinement rounds, conductor fixed a 3-line interpolation rename; live retry verified

scripts/mosaic-task.mjs

@@ -84,7 +84,7 @@ function validateId(value, what) {
 
 function validateMission(document, file) {
   if (!isPlainObject(document)) fail(2, "mission must be a JSON object");
-  rejectUnknownKeys(document, ["missionVersion", "id", "objective", "directives"], "mission");
+  rejectUnknownKeys(document, ["missionVersion", "id", "objective", "directives", "capabilities"], "mission");
   if (document.missionVersion !== 1) {
     fail(2, `unsupported missionVersion: ${JSON.stringify(document.missionVersion)} (supported: 1)`);
   }
@@ -102,7 +102,29 @@ function validateMission(document, file) {
       return d;
     });
   }
-  return { missionVersion: document.missionVersion, id: document.id, objective: document.objective, directives };
+
+  // Governing capability constraints (M9): same validation as task
+  // capabilities; semantically these BOUND tasks (least-privilege
+  // intersection at run time), never grant beyond them.
+  let capabilities = null;
+  if (document.capabilities !== undefined && document.capabilities !== null) {
+    if (!isPlainObject(document.capabilities)) fail(2, 'mission "capabilities" must be a JSON object');
+    rejectUnknownKeys(document.capabilities, ["tools"], 'mission "capabilities"');
+    if (!Array.isArray(document.capabilities.tools) || document.capabilities.tools.length === 0) {
+      fail(2, 'mission "capabilities.tools" must be a non-empty array of tool names');
+    }
+    const seen = new Set();
+    for (const tool of document.capabilities.tools) {
+      if (!SUPPORTED_TOOLS.includes(tool)) {
+        fail(2, `unsupported mission tool: ${JSON.stringify(tool)} (supported: ${SUPPORTED_TOOLS.join(", ")})`);
+      }
+      if (seen.has(tool)) fail(2, `duplicate tool in mission capabilities.tools: ${tool}`);
+      seen.add(tool);
+    }
+    capabilities = { tools: [...seen] };
+  }
+
+  return { missionVersion: document.missionVersion, id: document.id, objective: document.objective, directives, capabilities };
 }
 
 function validateTask(document, file) {
@@ -234,7 +256,7 @@ function writeOnce(file, content) {
   }
 }
 
-function runTask(taskFile) {
+function runTask(taskFile, options = {}) {
   const resolved = JSON.parse(
     spawnSync(process.execPath, [path.join(PROJECT_ROOT, "scripts", "mosaic-config.mjs"), "validate"], {
       cwd: PROJECT_ROOT,
@@ -291,7 +313,24 @@ function runTask(taskFile) {
     workspaceContainerPath = `/var/lib/mosaic/workspaces/${task.workspace}`;
   }
   if (workspaceContainerPath) spawnEnv.MOSAIC_WORKSPACE = workspaceContainerPath;
-  spawnEnv.MOSAIC_TOOLS = task.tools ? task.tools.join(",") : "";
+
+  // Capability policy (M9): least-privilege intersection. A task may narrow
+  // a mission's tool grant, never widen it. Empty intersection = tool-free.
+  let effectiveTools = task.tools;
+  let policyNote = null;
+  if (task.missionSnapshot?.capabilities) {
+    const missionTools = task.missionSnapshot.capabilities.tools;
+    if (effectiveTools) {
+      effectiveTools = effectiveTools.filter((t) => missionTools.includes(t));
+      if (effectiveTools.length === 0) {
+        policyNote = `capability policy: mission ${task.missionSnapshot.id} and task request no tools in common -> tool-free run`;
+      }
+    } else {
+      effectiveTools = [...missionTools];
+    }
+  }
+  if (policyNote) process.stderr.write(`mosaic-task: ${policyNote}\n`);
+  spawnEnv.MOSAIC_TOOLS = effectiveTools ? effectiveTools.join(",") : "";
 
   // Session (M6): persistent named session dir, passed as container path.
   if (task.session) {
@@ -345,8 +384,9 @@ function runTask(taskFile) {
     request: task.prompt,
     response,
     expectedExact: expected,
+    ...(options.retriedFrom ? { retriedFrom: options.retriedFrom } : {}),
     workspace: task.workspace,
-    tools: task.tools,
+    tools: effectiveTools,
     session: task.session,
     exitCode: proc.status,
     signal: proc.signal ?? null,
@@ -425,6 +465,7 @@ function showRun(runId) {
         `request:   ${JSON.stringify(result.request)}`,
         `response:  ${JSON.stringify(result.response)}`,
         result.expectedExact !== null && result.expectedExact !== undefined ? `expected:  ${JSON.stringify(result.expectedExact)}` : null,
+        result.retriedFrom ? `retriedFrom: ${result.retriedFrom}` : null,
         `timing:    ${result.startedAt} -> ${result.finishedAt} (${result.durationMs} ms)`,
         `exit:      ${result.exitCode}${result.signal ? ` signal=${result.signal}` : ""}`,
       ].filter((line) => line !== null).join("\n") + "\n",
@@ -455,6 +496,24 @@ function retryRun(runId) {
     fail(4, `run task snapshot unreadable: ${runId}`);
   }
 
+  // Relative mission paths in a snapshot resolve against the ORIGINAL task
+  // location, which no longer exists here — rewrite them to the run's own
+  // recorded mission.json so retries stay faithful.
+  let snapshotDoc;
+  try {
+    snapshotDoc = JSON.parse(snapshot);
+  } catch {
+    fail(4, `run task snapshot is not valid JSON: ${runId}`);
+  }
+  if (snapshotDoc.mission && !path.isAbsolute(snapshotDoc.mission)) {
+    const recordedMission = path.join(dir, "mission.json");
+    if (!fs.existsSync(recordedMission)) {
+      fail(4, `cannot retry ${runId}: relative mission path but no mission.json snapshot in run dir`);
+    }
+    snapshotDoc.mission = recordedMission;
+    snapshot = `${JSON.stringify(snapshotDoc, null, 2)}\n`;
+  }
+
   // A retry is a brand-new run: replay the recorded task snapshot through
   // the ordinary run path; existing run records stay untouched.
   const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), "mosaic-retry-"));
@@ -467,7 +526,7 @@ function retryRun(runId) {
       // Best-effort cleanup only.
     }
   });
-  runTask(tempTaskFile);
+  runTask(tempTaskFile, { retriedFrom: runId });
 }
 
 const operation = process.argv[2];

scripts/test-task.sh

@@ -142,6 +142,50 @@ EOF
         && grep -q 'Seam directive A.' "$SANDBOX/data/system-prompt.md" \
         && check "mission section injected into generated prompt" 0 \
         || check "mission section injected into generated prompt" 1
+
+    # retry lineage + relative mission path resolution
+    expect_exit "retry of mission run succeeds" 0 -- \
+        env MOSAIC_CONFIG="$SANDBOX/mock-adapters.json" MOSAIC_MOCK_RESPONSE=MOCKED \
+        node scripts/mosaic-task.mjs retry "$(ls -dt "$SANDBOX/data/runs"/r-* | head -1 | xargs basename)"
+    RT="$(ls -dt "$SANDBOX/data/runs"/r-* | head -1 | xargs basename)"
+    ORIG="$(ls -dt "$SANDBOX/data/runs"/r-* | sed -n 2p | xargs basename)"
+    node -e 'const r=JSON.parse(require("fs").readFileSync(process.argv[1],"utf8"));process.exit(r.retriedFrom===process.argv[2]?0:1)' \
+        "$SANDBOX/data/runs/$RT/result.json" "$ORIG" \
+        && check "retriedFrom lineage recorded" 0 || check "retriedFrom lineage recorded" 1
+    grep -q 'MISSION (runtime)' "$SANDBOX/data/system-prompt.md" \
+        && check "mission section present after retry (relative path resolved)" 0 \
+        || check "mission section present after retry (relative path resolved)" 1
+    expect_exit "retry of missing run exits 4" 4 -- \
+        env MOSAIC_CONFIG="$SANDBOX/mock-adapters.json" node scripts/mosaic-task.mjs retry r-missing
+
+    # capability policy (M9): least-privilege intersection
+    POL="$SANDBOX/data/workspaces"; mkdir -p "$POL"
+    pol_run() { # missionTools(ABSENT|json) taskTools(ABSENT|json) -> stderr MOSAIC_TOOLS value
+        if [ "$1" = "ABSENT" ]; then
+            printf '{"missionVersion":1,"id":"m-pol","objective":"o"}' > "$SANDBOX/pol-m.json"
+        else
+            printf '{"missionVersion":1,"id":"m-pol","objective":"o","capabilities":{"tools":[%s]}}' "$1" > "$SANDBOX/pol-m.json"
+        fi
+        if [ "$2" = "ABSENT" ]; then
+            printf '{"taskVersion":1,"id":"t-pol","prompt":"x","mission":"pol-m.json"}' > "$SANDBOX/pol-t.json"
+        else
+            printf '{"taskVersion":1,"id":"t-pol","prompt":"x","mission":"pol-m.json","capabilities":{"tools":[%s]}}' "$2" > "$SANDBOX/pol-t.json"
+        fi
+        env MOSAIC_CONFIG="$SANDBOX/mock-adapters.json" MOSAIC_MOCK_RESPONSE=MOCKED \
+            scripts/run-task.sh run "$SANDBOX/pol-t.json" >/dev/null 2>&1
+        grep -o '^MOSAIC_TOOLS=.*' "$(ls -dt "$SANDBOX/data/runs"/r-* | head -1)/stderr.txt" 2>/dev/null
+    }
+    [ "$(pol_run '"read","bash"' ABSENT)" = 'MOSAIC_TOOLS=read,bash' ] \
+        && check "policy: mission only -> mission tools" 0 || check "policy: mission only -> mission tools" 1
+    [ "$(pol_run ABSENT '"read","bash"')" = 'MOSAIC_TOOLS=read,bash' ] \
+        && check "policy: task only -> task tools" 0 || check "policy: task only -> task tools" 1
+    [ "$(pol_run '"read"' '"bash","read"')" = 'MOSAIC_TOOLS=read' ] \
+        && check "policy: both -> intersection (task narrowed)" 0 || check "policy: both -> intersection (task narrowed)" 1
+    [ "$(pol_run '"grep"' '"bash","read"')" = 'MOSAIC_TOOLS=' ] \
+        && check "policy: empty intersection -> tool-free" 0 || check "policy: empty intersection -> tool-free" 1
+    printf '{"missionVersion":1,"id":"m-pol","objective":"o","capabilities":{"tools":["sudo"]}}' > "$SANDBOX/pol-m.json"
+    expect_exit "invalid mission capabilities rejected" 2 -- \
+        env MOSAIC_CONFIG="$SANDBOX/mock-adapters.json" $TASK validate "$SANDBOX/pol-t.json"
 else
     echo "skip adapter seam cases (docker daemon unavailable)"
 fi