Merge M10: run-record retention

Closes #32

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 review of M9 (mission capability policy) — then name the next 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. Run-record retention/pruning policy
+2. Second real adapter (parked — owner focused on Pi)
+3. Auto-apply policy for worker patches (capability policy now exists as its substrate)
+4. Session forking from a common ancestor (pi JSONL trees make this native)
 
 ## Rules
 
@@ -31,3 +32,8 @@ 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 — M9 mission capability policy (#30) — merged, least-privilege intersection, 41/36/14 + 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
+- 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 +526,53 @@ function retryRun(runId) {
       // Best-effort cleanup only.
     }
   });
-  runTask(tempTaskFile);
+  runTask(tempTaskFile, { retriedFrom: runId });
+}
+
+function pruneRuns(args) {
+  const resolved = loadConfig();
+  const root = runsRoot(resolved);
+  let keep = 50;
+  let apply = false;
+  for (const arg of args) {
+    if (arg === "--yes") apply = true;
+    else if (arg === "--keep") fail(4, "prune: --keep requires a value");
+    else if (arg.startsWith("--keep=")) {
+      keep = Number(arg.slice("--keep=".length));
+      if (!Number.isInteger(keep) || keep < 1) fail(4, `--keep must be a positive integer (got ${arg.slice(7)})`);
+    } else fail(4, `unknown prune option: ${arg}`);
+  }
+
+  let entries = [];
+  try {
+    entries = fs.readdirSync(root, { withFileTypes: true })
+      .filter((e) => e.name.startsWith("r-") && e.isDirectory() && !e.isSymbolicLink())
+      .map((e) => e.name)
+      .sort();
+  } catch {
+    // No runs yet.
+  }
+
+  if (entries.length <= keep) {
+    process.stdout.write(`prune: ${entries.length} run(s) present, keep=${keep} -> nothing to prune\n`);
+    process.exit(0);
+  }
+
+  const doomed = entries.slice(0, entries.length - keep); // oldest first
+  if (!apply) {
+    process.stdout.write(`prune (dry-run): would remove ${doomed.length} oldest run(s), keep ${entries.length - doomed.length}:\n`);
+    for (const id of doomed) process.stdout.write(`  would remove: ${id}\n`);
+    process.stdout.write("prune: re-run with --yes to apply\n");
+    process.exit(0);
+  }
+
+  const receipt = path.join(root, ".pruned.log");
+  for (const id of doomed) {
+    fs.rmSync(path.join(root, id), { recursive: true, force: true });
+    fs.appendFileSync(receipt, `${JSON.stringify({ at: new Date().toISOString(), event: "pruned", runId: id })}\n`);
+  }
+  process.stdout.write(`prune: removed ${doomed.length} run(s), kept ${keep}; receipt: ${receipt}\n`);
+  process.exit(0);
 }
 
 const operation = process.argv[2];
@@ -493,10 +598,13 @@ switch (operation) {
   case "list":
     listRuns();
     process.exit(0);
+  case "prune":
+    pruneRuns(process.argv.slice(3));
+    break;
   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)`);
+    fail(4, `unknown operation: ${JSON.stringify(operation ?? "")} (expected validate | run | show | list | retry | prune)`);
 }

scripts/test-config.sh

@@ -10,10 +10,16 @@ SANDBOX="$(mktemp -d)"
 trap 'rm -rf "$SANDBOX"' EXIT
 
 PASS=0
+# Status colors: terminal-only, NO_COLOR-respecting; plain when piped.
+if [ -t 1 ] && [ -z "${NO_COLOR:-}" ]; then
+    C_OK=$'\033[0;32m'; C_FAIL=$'\033[0;31m'; C_RESET=$'\033[0m'
+else
+    C_OK=""; C_FAIL=""; C_RESET=""
+fi
 FAIL=0
 
 check() {
-    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "ok   $1"; else FAIL=$((FAIL+1)); echo "FAIL $1"; fi
+    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    $1"; else FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} $1"; fi
 }
 
 # expect_exit NAME EXPECTED_RC -- command...
@@ -25,10 +31,10 @@ expect_exit() {
     rc=$?
     if [ "$rc" -eq "$expected" ]; then
         PASS=$((PASS + 1))
-        echo "ok   $name (exit $rc)"
+        echo "${C_OK}OK${C_RESET}    $name (exit $rc)"
     else
         FAIL=$((FAIL + 1))
-        echo "FAIL $name (exit $rc, expected $expected)"
+        echo "${C_FAIL}FAIL${C_RESET} $name (exit $rc, expected $expected)"
     fi
 }
 
@@ -62,8 +68,8 @@ check "env exports adapter" $?
 rm -f "$SANDBOX/config.json"
 expect_exit "bootstrap creates default when absent" 0 -- \
     env MOSAIC_CONFIG="$SANDBOX/config.json" $CONFIG_OP bootstrap
-[ -f "$SANDBOX/config.json" ] && { PASS=$((PASS+1)); echo "ok   bootstrap wrote config file"; } \
-    || { FAIL=$((FAIL+1)); echo "FAIL bootstrap wrote config file"; }
+[ -f "$SANDBOX/config.json" ] && { PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    bootstrap wrote config file"; } \
+    || { FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} bootstrap wrote config file"; }
 
 SUM_BEFORE=$(sha256sum "$SANDBOX/config.json" | cut -d' ' -f1)
 MTIME_BEFORE=$(stat -c %Y "$SANDBOX/config.json")
@@ -73,9 +79,9 @@ expect_exit "bootstrap is idempotent on existing config" 0 -- \
 SUM_AFTER=$(sha256sum "$SANDBOX/config.json" | cut -d' ' -f1)
 MTIME_AFTER=$(stat -c %Y "$SANDBOX/config.json")
 if [ "$SUM_BEFORE" = "$SUM_AFTER" ] && [ "$MTIME_BEFORE" = "$MTIME_AFTER" ]; then
-    PASS=$((PASS+1)); echo "ok   bootstrap did not rewrite existing config"
+    PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    bootstrap did not rewrite existing config"
 else
-    FAIL=$((FAIL+1)); echo "FAIL bootstrap rewrote existing config"
+    FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} bootstrap rewrote existing config"
 fi
 
 # --- validate ---
@@ -140,9 +146,9 @@ cfg valid.json "$(valid_body "$DATA_ROOT")"
 EVAL_OUT="$(MOSAIC_CONFIG="$SANDBOX/valid.json" $CONFIG_OP env)" || true
 if eval "$EVAL_OUT" 2>/dev/null && [ "$MOSAIC_DATA_ROOT" = "$DATA_ROOT" ] \
     && [ "$MOSAIC_PROVIDER" = "zai" ] && [ "$MOSAIC_MODEL" = "glm-5.3-flash" ]; then
-    PASS=$((PASS+1)); echo "ok   env exports resolve correctly"
+    PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    env exports resolve correctly"
 else
-    FAIL=$((FAIL+1)); echo "FAIL env exports resolve correctly"
+    FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} env exports resolve correctly"
 fi
 
 # --- validation must not modify the file ---
@@ -150,9 +156,9 @@ SUM_INVALID_BEFORE=$(sha256sum "$SANDBOX/invalid.json" | cut -d' ' -f1)
 MOSAIC_CONFIG="$SANDBOX/invalid.json" $CONFIG_OP validate >/dev/null 2>&1
 SUM_INVALID_AFTER=$(sha256sum "$SANDBOX/invalid.json" | cut -d' ' -f1)
 if [ "$SUM_INVALID_BEFORE" = "$SUM_INVALID_AFTER" ]; then
-    PASS=$((PASS+1)); echo "ok   failed validation modified nothing"
+    PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    failed validation modified nothing"
 else
-    FAIL=$((FAIL+1)); echo "FAIL failed validation modified the file"
+    FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} failed validation modified the file"
 fi
 
 echo

scripts/test-release.sh

@@ -15,6 +15,12 @@ cp RELEASE "$RELEASE_BACKUP"
 trap 'cp "$RELEASE_BACKUP" RELEASE 2>/dev/null; rm -rf "$SANDBOX" "$RELEASE_BACKUP"' EXIT
 
 PASS=0
+# Status colors: terminal-only, NO_COLOR-respecting; plain when piped.
+if [ -t 1 ] && [ -z "${NO_COLOR:-}" ]; then
+    C_OK=$'\033[0;32m'; C_FAIL=$'\033[0;31m'; C_RESET=$'\033[0m'
+else
+    C_OK=""; C_FAIL=""; C_RESET=""
+fi
 FAIL=0
 
 expect_exit() {
@@ -24,14 +30,14 @@ expect_exit() {
     "$@" >/dev/null 2>&1
     rc=$?
     if [ "$rc" -eq "$expected" ]; then
-        PASS=$((PASS+1)); echo "ok   $name (exit $rc)"
+        PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    $name (exit $rc)"
     else
-        FAIL=$((FAIL+1)); echo "FAIL $name (exit $rc, expected $expected)"
+        FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} $name (exit $rc, expected $expected)"
     fi
 }
 
 check() {
-    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "ok   $1"; else FAIL=$((FAIL+1)); echo "FAIL $1"; fi
+    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    $1"; else FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} $1"; fi
 }
 
 # ---------- fast: release identity ----------

scripts/test-task.sh

@@ -11,6 +11,12 @@ SANDBOX="$(mktemp -d)"
 trap 'rm -rf "$SANDBOX"' EXIT
 
 PASS=0
+# Status colors: terminal-only, NO_COLOR-respecting; plain when piped.
+if [ -t 1 ] && [ -z "${NO_COLOR:-}" ]; then
+    C_OK=$'\033[0;32m'; C_FAIL=$'\033[0;31m'; C_RESET=$'\033[0m'
+else
+    C_OK=""; C_FAIL=""; C_RESET=""
+fi
 FAIL=0
 
 expect_exit() {
@@ -20,14 +26,14 @@ expect_exit() {
     "$@" >/dev/null 2>&1
     rc=$?
     if [ "$rc" -eq "$expected" ]; then
-        PASS=$((PASS+1)); echo "ok   $name (exit $rc)"
+        PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    $name (exit $rc)"
     else
-        FAIL=$((FAIL+1)); echo "FAIL $name (exit $rc, expected $expected)"
+        FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} $name (exit $rc, expected $expected)"
     fi
 }
 
 check() {
-    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "ok   $1"; else FAIL=$((FAIL+1)); echo "FAIL $1"; fi
+    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    $1"; else FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} $1"; fi
 }
 
 latest_reason() {
@@ -36,10 +42,6 @@ latest_reason() {
     [ -n "$latest" ] && node -e 'try{const r=JSON.parse(require("fs").readFileSync(process.argv[1],"utf8"));console.log(r.reason??"")}catch{console.log("")}' "$latest/result.json" 2>/dev/null
 }
 
-check() {
-    if [ "$2" = "0" ]; then PASS=$((PASS+1)); echo "ok   $1"; else FAIL=$((FAIL+1)); echo "FAIL $1"; fi
-}
-
 CONFIG="$SANDBOX/config.json"
 DATA_ROOT="$SANDBOX/data"
 mkdir -p "$DATA_ROOT"
@@ -102,6 +104,37 @@ $TASK validate "$SANDBOX/ok.json" >/dev/null 2>&1
 M2=$(stat -c %Y "$SANDBOX/ok.json")
 [ "$M1" = "$M2" ] && check "validation does not modify the task file" 0 || check "validation does not modify the task file" 1
 
+# ---------- retention: prune (deterministic, no Docker) ----------
+mkdir -p "$SANDBOX/data"
+cat > "$SANDBOX/prune-config.json" <<EOF
+{"configVersion":1,"environment":"development","dataRoot":"$SANDBOX/data","execution":{"backend":"docker","provider":"zai","model":"m"}}
+EOF
+for i in 1 2 3 4 5; do
+    D="$SANDBOX/data/runs/r-20260903T0100_0${i}Z-suite00$i"
+    mkdir -p "$D"
+    printf '{"runVersion":1,"runId":"r-20260903T0100_0%sZ-suite00%s","taskId":"t","status":"succeeded"}' "$i" "$i" > "$D/result.json"
+done
+mkdir -p "$SANDBOX/data/sessions/sentinel" "$SANDBOX/data/workspaces/sentinel"
+
+expect_exit "prune dry-run exits 0" 0 -- env MOSAIC_CONFIG="$SANDBOX/prune-config.json" node scripts/mosaic-task.mjs prune
+[ "$(ls "$SANDBOX/data/runs" | grep -c '^r-')" -eq 5 ] \
+    && check "dry-run deleted nothing" 0 || check "dry-run deleted nothing" 1
+
+expect_exit "prune --keep=2 --yes removes oldest" 0 -- env MOSAIC_CONFIG="$SANDBOX/prune-config.json" node scripts/mosaic-task.mjs prune --keep=2 --yes
+[ "$(ls "$SANDBOX/data/runs" | grep -c '^r-')" -eq 2 ] \
+    && check "kept exactly 2 newest runs" 0 || check "kept exactly 2 newest runs" 1
+NEWEST="r-20260903T0100_05Z-suite005"
+[ -d "$SANDBOX/data/runs/$NEWEST" ] \
+    && check "newest run kept, oldest pruned" 0 || check "newest run kept, oldest pruned" 1
+[ -f "$SANDBOX/data/runs/.pruned.log" ] \
+    && [ "$(grep -c 'pruned' "$SANDBOX/data/runs/.pruned.log")" -eq 3 ] \
+    && check "append-only receipt written (3 entries)" 0 \
+    || check "append-only receipt written (3 entries)" 1
+[ -d "$SANDBOX/data/sessions/sentinel" ] && [ -d "$SANDBOX/data/workspaces/sentinel" ] \
+    && check "sessions/workspaces untouched by prune" 0 \
+    || check "sessions/workspaces untouched by prune" 1
+expect_exit "prune with invalid keep exits 4" 4 -- env MOSAIC_CONFIG="$SANDBOX/prune-config.json" node scripts/mosaic-task.mjs prune --keep=0 --yes
+
 # ---------- adapter seam: deterministic mock cases (Docker, no provider) ----------
 if docker info >/dev/null 2>&1; then
     good_task "$SANDBOX/ok.json"
@@ -142,6 +175,50 @@ EOF
         && grep -q 'Seam directive A.' "$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
@@ -194,18 +271,12 @@ dump_latest_run() {
     fi
 }
 
-latest_reason() {
-    local latest
-    latest="$(ls -dt "$SANDBOX/data/runs"/r-* 2>/dev/null | head -1)"
-    [ -n "$latest" ] && node -e 'try{const r=JSON.parse(require("fs").readFileSync(process.argv[1],"utf8"));console.log(r.reason??"")}catch{console.log("")}' "$latest/result.json" 2>/dev/null
-}
-
 if docker info >/dev/null 2>&1; then
     RUNS1=$(ls "$DATA_ROOT/runs" 2>/dev/null | wc -l)
     if scripts/run-task.sh run "$SANDBOX/ok.json" >/dev/null 2>&1; then
-        PASS=$((PASS+1)); echo "ok   live hello task succeeds with exact marker"
+        PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    live hello task succeeds with exact marker"
     else
-        FAIL=$((FAIL+1)); echo "FAIL live hello task succeeds with exact marker" >&2
+        FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} live hello task succeeds with exact marker" >&2
         dump_latest_run
     fi
 
@@ -222,9 +293,9 @@ process.exit(r.status === "succeeded" && r.response === "MOSAIC_HELLO_OK" && r.e
     RC=$?
     WRONG_REASON="$(latest_reason)"
     if [ "$RC" -eq 1 ] && [ "$WRONG_REASON" = "expect-mismatch" ]; then
-        PASS=$((PASS+1)); echo "ok   wrong expectExact fails with exit 1 (reason: expect-mismatch)"
+        PASS=$((PASS+1)); echo "${C_OK}OK${C_RESET}    wrong expectExact fails with exit 1 (reason: expect-mismatch)"
     else
-        FAIL=$((FAIL+1)); echo "FAIL wrong expectExact (exit $RC, reason: '${WRONG_REASON:-none}')" >&2
+        FAIL=$((FAIL+1)); echo "${C_FAIL}FAIL${C_RESET} wrong expectExact (exit $RC, reason: '${WRONG_REASON:-none}')" >&2
         dump_latest_run
     fi
 

scripts/verify.sh

@@ -16,6 +16,13 @@ source scripts/common.sh
 
 EXPECTED="${EXPECTED_MARKER:-MOSAIC_HELLO_OK}"
 
+# Status colors: terminal-only, NO_COLOR-respecting; plain when piped.
+if [ -t 1 ] && [ -z "${NO_COLOR:-}" ]; then
+    C_OK=$'\033[0;32m'; C_FAIL=$'\033[0;31m'; C_RESET=$'\033[0m'
+else
+    C_OK=""; C_FAIL=""; C_RESET=""
+fi
+
 load_config
 load_release
 IMAGE="$MOSAIC_IMAGE_TAG"
@@ -51,11 +58,11 @@ TRIMMED="$(printf '%s' "$RESPONSE" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]
 
 # 4-6. Exact comparison gate.
 if [ "$TRIMMED" = "$EXPECTED" ]; then
-    echo "PASS: response matches expected marker"
+    echo "${C_OK}PASS${C_RESET}: response matches expected marker"
     exit 0
 fi
 
-echo "FAIL: response does not match expected marker" >&2
+echo "${C_FAIL}FAIL${C_RESET}: response does not match expected marker" >&2
 printf 'expected: %s\n' "$EXPECTED" >&2
 printf 'actual  : %s\n' "$TRIMMED" >&2
 exit 1