cool-workflow 0.1.81 → 0.1.83

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (102) hide show
  1. package/.claude-plugin/plugin.json +2 -2
  2. package/.codex-plugin/plugin.json +4 -4
  3. package/README.md +125 -121
  4. package/apps/architecture-review/app.json +1 -1
  5. package/apps/architecture-review-fast/app.json +1 -1
  6. package/apps/end-to-end-golden-path/app.json +1 -1
  7. package/apps/pr-review-fix-ci/app.json +1 -1
  8. package/apps/release-cut/app.json +1 -1
  9. package/apps/research-synthesis/app.json +1 -1
  10. package/dist/candidate-scoring.js +20 -26
  11. package/dist/capability-core.js +80 -93
  12. package/dist/capability-registry.js +30 -3
  13. package/dist/cli.js +12 -1
  14. package/dist/commit.js +217 -204
  15. package/dist/coordinator/util.js +6 -9
  16. package/dist/dispatch.js +11 -3
  17. package/dist/doctor.js +153 -0
  18. package/dist/evidence-reasoning.js +4 -1
  19. package/dist/execution-backend/agent.js +11 -48
  20. package/dist/execution-backend.js +11 -31
  21. package/dist/gates.js +48 -0
  22. package/dist/mcp-server.js +11 -0
  23. package/dist/multi-agent/helpers.js +6 -10
  24. package/dist/multi-agent/ids.js +20 -0
  25. package/dist/multi-agent-eval.js +27 -1
  26. package/dist/multi-agent-host.js +53 -21
  27. package/dist/multi-agent-operator-ux.js +2 -1
  28. package/dist/multi-agent-trust.js +5 -5
  29. package/dist/node-projection.js +59 -0
  30. package/dist/node-snapshot.js +8 -18
  31. package/dist/orchestrator/lifecycle-operations.js +22 -1
  32. package/dist/orchestrator.js +29 -2
  33. package/dist/reclamation/hash.js +72 -0
  34. package/dist/reclamation.js +31 -112
  35. package/dist/run-registry/queue.js +6 -7
  36. package/dist/run-registry.js +35 -24
  37. package/dist/scheduler.js +112 -57
  38. package/dist/topology.js +25 -4
  39. package/dist/trust-audit.js +70 -38
  40. package/dist/validation.js +328 -0
  41. package/dist/version.js +1 -1
  42. package/dist/worker-isolation.js +163 -58
  43. package/docs/agent-delegation-drive.7.md +90 -85
  44. package/docs/agent-framework.md +33 -32
  45. package/docs/candidate-scoring.7.md +26 -24
  46. package/docs/canonical-workflow-apps.7.md +40 -40
  47. package/docs/capability-topology-registry.7.md +24 -24
  48. package/docs/cli-mcp-parity.7.md +227 -154
  49. package/docs/contract-migration-tooling.7.md +49 -41
  50. package/docs/control-plane-scheduling.7.md +45 -40
  51. package/docs/coordinator-blackboard.7.md +30 -30
  52. package/docs/dogfood-one-real-repo.7.md +44 -44
  53. package/docs/durable-state-and-locking.7.md +35 -30
  54. package/docs/end-to-end-golden-path.7.md +29 -29
  55. package/docs/error-feedback.7.md +27 -27
  56. package/docs/evidence-adoption-reasoning-chain.7.md +63 -58
  57. package/docs/execution-backends.7.md +84 -79
  58. package/docs/getting-started.md +35 -18
  59. package/docs/index.md +3 -3
  60. package/docs/mcp-app-surface.7.md +64 -64
  61. package/docs/multi-agent-cli-mcp-surface.7.md +83 -77
  62. package/docs/multi-agent-eval-replay-harness.7.md +59 -54
  63. package/docs/multi-agent-operator-ux.7.md +70 -65
  64. package/docs/multi-agent-runtime-core.7.md +39 -39
  65. package/docs/multi-agent-topologies.7.md +24 -24
  66. package/docs/multi-agent-trust-policy-audit.7.md +38 -38
  67. package/docs/node-snapshot-diff-replay.7.md +26 -21
  68. package/docs/observability-cost-accounting.7.md +50 -45
  69. package/docs/operator-ux.7.md +30 -30
  70. package/docs/pipeline-runner.7.md +31 -31
  71. package/docs/project-index.md +15 -6
  72. package/docs/real-execution-backends.7.md +47 -42
  73. package/docs/release-and-migration.7.md +43 -38
  74. package/docs/release-tooling.7.md +74 -39
  75. package/docs/routines.md +16 -16
  76. package/docs/run-registry-control-plane.7.md +120 -115
  77. package/docs/run-retention-reclamation.7.md +46 -41
  78. package/docs/sandbox-profiles.7.md +32 -32
  79. package/docs/scheduled-tasks.md +14 -14
  80. package/docs/security-trust-hardening.7.md +29 -29
  81. package/docs/source-context-profiles.7.md +28 -28
  82. package/docs/state-explosion-management.7.md +64 -59
  83. package/docs/state-node.7.md +8 -8
  84. package/docs/team-collaboration.7.md +63 -58
  85. package/docs/trust-model.md +126 -126
  86. package/docs/unix-principles.md +80 -80
  87. package/docs/vendor-manifest-loadability.7.md +20 -20
  88. package/docs/verifier-gated-commit.7.md +16 -16
  89. package/docs/web-desktop-workbench.7.md +70 -65
  90. package/docs/worker-isolation.7.md +34 -37
  91. package/docs/workflow-app-framework.7.md +38 -38
  92. package/manifest/plugin.manifest.json +4 -4
  93. package/manifest/source-context-profiles.json +1 -1
  94. package/package.json +3 -2
  95. package/scripts/canonical-apps.js +4 -4
  96. package/scripts/children/batch-delegate-child.js +58 -0
  97. package/scripts/children/http-delegate-child.js +39 -0
  98. package/scripts/dogfood-release.js +1 -1
  99. package/scripts/gen-parity-doc.js +106 -0
  100. package/scripts/golden-path.js +4 -4
  101. package/scripts/release-flow.js +181 -5
  102. package/dist/verifier-registry.js +0 -46
@@ -3,7 +3,7 @@
3
3
  "id": "pr-review-fix-ci",
4
4
  "title": "PR Review Fix CI",
5
5
  "summary": "Review a pull request or branch, inspect CI failures, diagnose actionable issues, optionally patch, verify, and summarize with evidence.",
6
- "version": "0.1.81",
6
+ "version": "0.1.83",
7
7
  "author": "COOLWHITE LLC",
8
8
  "inputs": [
9
9
  {
@@ -3,7 +3,7 @@
3
3
  "id": "release-cut",
4
4
  "title": "Release Cut",
5
5
  "summary": "Prepare a release with checklist discipline: version checks, changelog, tests, packaging, release notes, and final verification.",
6
- "version": "0.1.81",
6
+ "version": "0.1.83",
7
7
  "author": "COOLWHITE LLC",
8
8
  "inputs": [
9
9
  {
@@ -3,7 +3,7 @@
3
3
  "id": "research-synthesis",
4
4
  "title": "Research Synthesis",
5
5
  "summary": "Split a research question into claims, investigate sources, cross-check evidence, verify claims, and synthesize a concise answer.",
6
- "version": "0.1.81",
6
+ "version": "0.1.83",
7
7
  "author": "COOLWHITE LLC",
8
8
  "inputs": [
9
9
  {
@@ -21,6 +21,8 @@ const state_node_1 = require("./state-node");
21
21
  const trust_audit_1 = require("./trust-audit");
22
22
  const collaboration_1 = require("./collaboration");
23
23
  const compare_1 = require("./compare");
24
+ const gates_1 = require("./gates");
25
+ const validation_1 = require("./validation");
24
26
  exports.CANDIDATE_SCHEMA_VERSION = 1;
25
27
  /** Verdict thresholds on a score's normalized value [0,1], declared once so the
26
28
  * numbers carry intent instead of being buried as literals in verdictFor(). A
@@ -110,7 +112,10 @@ function getCandidate(run, candidateId) {
110
112
  const file = candidateFile(run, candidateId);
111
113
  if (!node_fs_1.default.existsSync(file))
112
114
  return undefined;
113
- const candidate = JSON.parse(node_fs_1.default.readFileSync(file, "utf8"));
115
+ // Fail-closed integrity boundary (F4/F5): validate the parsed record against
116
+ // its type def BEFORE upserting it as a trusted CandidateRecord. A corrupt or
117
+ // forged candidate.json must throw here rather than flow into the run.
118
+ const candidate = (0, validation_1.validateCandidateRecord)(JSON.parse(node_fs_1.default.readFileSync(file, "utf8")));
114
119
  upsertCandidate(run, candidate);
115
120
  return candidate;
116
121
  }
@@ -245,10 +250,11 @@ function selectCandidate(run, candidateId, options = {}, scoringOptions = {}) {
245
250
  else if (!verifierNode.evidence.length) {
246
251
  failures.push(error("candidate-selection-missing-evidence", `Candidate ${candidateId} verifier node has no evidence`));
247
252
  }
248
- else if (emptyCaptureWarning(run, verifierNode)) {
249
- // HARD no-false-green gate (v0.1.43) — kept in SYNC with the commit gate
250
- // (commit.ts emptyCaptureWarning): a verifier node whose backing result was
251
- // an empty-capture must not be selectable, so selection + commit agree.
253
+ else if ((0, gates_1.emptyCaptureWarning)(run, verifierNode)) {
254
+ // HARD no-false-green gate (v0.1.43) — selection and the commit gate now
255
+ // share ONE emptyCaptureWarning (src/gates.ts), so they CANNOT drift: a
256
+ // verifier node whose backing result was an empty-capture is unselectable
257
+ // here for the same reason it is uncommittable, by construction.
252
258
  failures.push(error("candidate-selection-empty-capture", `Candidate ${candidateId} verifier node has no real evidence (empty-capture result)`));
253
259
  }
254
260
  }
@@ -312,7 +318,7 @@ function selectCandidate(run, candidateId, options = {}, scoringOptions = {}) {
312
318
  scoreCriteria: bestScore?.criteria,
313
319
  verifierNodeId: candidate.verifierNodeId,
314
320
  evidenceCount: mergeEvidence(candidate.evidence, verifierNode?.evidence || []).length,
315
- sandboxProfileId: sandboxProfileForCandidate(run, candidate),
321
+ sandboxProfileId: (0, gates_1.sandboxProfileForCandidate)(run, candidate),
316
322
  workerId: candidate.workerId,
317
323
  commitGateResult: "passed"
318
324
  }),
@@ -528,7 +534,10 @@ function loadCandidatesFromDisk(run) {
528
534
  .filter((entry) => entry.isDirectory() && entry.name !== "selections")
529
535
  .map((entry) => node_path_1.default.join(candidateRoot(run), entry.name, "candidate.json"))
530
536
  .filter((file) => node_fs_1.default.existsSync(file))
531
- .map((file) => JSON.parse(node_fs_1.default.readFileSync(file, "utf8")));
537
+ // Fail-closed integrity boundary (F4/F5): each candidate.json is validated
538
+ // against CandidateRecord before it merges into the run; a corrupt record
539
+ // throws rather than entering the candidate set as a trusted cast.
540
+ .map((file) => (0, validation_1.validateCandidateRecord)(JSON.parse(node_fs_1.default.readFileSync(file, "utf8"))));
532
541
  }
533
542
  function readScores(run, candidateId) {
534
543
  const dir = node_path_1.default.join(candidateDir(run, candidateId), "scores");
@@ -538,7 +547,10 @@ function readScores(run, candidateId) {
538
547
  .readdirSync(dir)
539
548
  .filter((file) => file.endsWith(".json"))
540
549
  .sort()
541
- .map((file) => JSON.parse(node_fs_1.default.readFileSync(node_path_1.default.join(dir, file), "utf8")));
550
+ // Fail-closed integrity boundary (F4/F5): a score file is validated against
551
+ // CandidateScore before it can feed ranking/selection. A corrupt score must
552
+ // throw, not silently widen the normalized/verdict surface the gate reads.
553
+ .map((file) => (0, validation_1.validateCandidateScore)(JSON.parse(node_fs_1.default.readFileSync(node_path_1.default.join(dir, file), "utf8"))));
542
554
  }
543
555
  function candidateArtifacts(run, candidate) {
544
556
  return [
@@ -661,17 +673,6 @@ function error(code, message, options = {}) {
661
673
  details: options.details
662
674
  };
663
675
  }
664
- /** HARD no-false-green gate (v0.1.43) — kept in SYNC with commit.ts. Traces a
665
- * verifier node back to its source result node and returns the empty-capture
666
- * marker (set at ingest via isEmptyCapture) when present. Reads ONLY persisted
667
- * state, so selection replays deterministically. */
668
- function emptyCaptureWarning(run, verifierNode) {
669
- const resultNodeId = (typeof verifierNode.inputs?.inputNodeId === "string" ? verifierNode.inputs.inputNodeId : undefined) ||
670
- verifierNode.parents[0];
671
- const resultNode = resultNodeId ? run.nodes?.find((node) => node.id === resultNodeId) : undefined;
672
- const warning = resultNode?.metadata?.captureWarning;
673
- return typeof warning === "string" && warning ? warning : undefined;
674
- }
675
676
  function mergeCandidates(left, right) {
676
677
  const merged = [...left];
677
678
  for (const candidate of right) {
@@ -708,13 +709,6 @@ function mergeEvidence(left, right) {
708
709
  }
709
710
  return merged;
710
711
  }
711
- function sandboxProfileForCandidate(run, candidate) {
712
- const worker = candidate.workerId ? (run.workers || []).find((entry) => entry.id === candidate.workerId) : undefined;
713
- if (worker?.sandboxProfileId)
714
- return worker.sandboxProfileId;
715
- const task = candidate.taskId ? (run.tasks || []).find((entry) => entry.id === candidate.taskId) : undefined;
716
- return task?.sandboxProfileId;
717
- }
718
712
  function unique(values) {
719
713
  return Array.from(new Set(values.filter(Boolean)));
720
714
  }
@@ -198,18 +198,16 @@ function flag(value) {
198
198
  return false;
199
199
  return Boolean(value);
200
200
  }
201
- function withInvocationCwd(args, fn) {
202
- const cwd = optionalString(args.cwd);
203
- if (!cwd)
204
- return fn();
205
- const previous = process.cwd();
206
- process.chdir(cwd);
207
- try {
208
- return fn();
209
- }
210
- finally {
211
- process.chdir(previous);
212
- }
201
+ // F7: explicit invocation cwd — no more process.chdir bracket.
202
+ // The runner resolves a run from process.cwd() by default; to operate WITH a run's
203
+ // repo as base (run --drive --repo X from anywhere; cross-directory quickstart; run
204
+ // import/export against a target dir) we now pass that base EXPLICITLY —
205
+ // runner.withBaseDir(dir).loadRun(...) for run resolution (see
206
+ // CoolWorkflowRunner.withBaseDir), and invocationCwd(args) to anchor relative path
207
+ // args. Nothing mutates the global process.cwd(), so concurrent in-process callers
208
+ // can no longer corrupt each other's working directory (the former reentrancy hazard).
209
+ function invocationCwd(args) {
210
+ return node_path_1.default.resolve(optionalString(args.cwd) || process.cwd());
213
211
  }
214
212
  function runRegistryRefresh(reg, args) {
215
213
  return reg.refresh({ scope: scopeOf(args, "repo") });
@@ -277,36 +275,33 @@ function runRerun(reg, runId, args) {
277
275
  return reg.rerun(runId, { scope: scopeOf(args, "home"), reason: optionalString(args.reason) });
278
276
  }
279
277
  function runExportArchive(runner, runId, args) {
280
- return withInvocationCwd(args, () => {
281
- const output = optionalString(args.output || args.path || args.archive) || `${runId}.cwrun.json`;
282
- return (0, run_export_1.exportRun)(runner.loadRun(runId), node_path_1.default.resolve(output));
283
- });
278
+ const base = invocationCwd(args);
279
+ const output = optionalString(args.output || args.path || args.archive) || `${runId}.cwrun.json`;
280
+ return (0, run_export_1.exportRun)(runner.withBaseDir(optionalString(args.cwd)).loadRun(runId), node_path_1.default.resolve(base, output));
284
281
  }
285
282
  function runImportArchive(runner, args) {
286
- return withInvocationCwd(args, () => {
287
- const archive = optionalString(args.archive || args.path || args.file);
288
- if (!archive)
289
- throw new Error("run import requires an archive path (positional, --archive, --path, or --file)");
290
- const target = optionalString(args.target || args.repo || args.cwd) || process.cwd();
291
- const imported = (0, run_export_1.importRun)(node_path_1.default.resolve(archive), node_path_1.default.resolve(target));
292
- const registry = new run_registry_1.RunRegistry(node_path_1.default.resolve(target), runner);
293
- const registryReport = registry.refresh({ scope: "repo" });
294
- return { ...imported, registry: registryReport };
295
- });
283
+ const base = invocationCwd(args);
284
+ const archive = optionalString(args.archive || args.path || args.file);
285
+ if (!archive)
286
+ throw new Error("run import requires an archive path (positional, --archive, --path, or --file)");
287
+ const target = node_path_1.default.resolve(base, optionalString(args.target || args.repo || args.cwd) || base);
288
+ const imported = (0, run_export_1.importRun)(node_path_1.default.resolve(base, archive), target);
289
+ const registry = new run_registry_1.RunRegistry(target, runner.withBaseDir(target));
290
+ const registryReport = registry.refresh({ scope: "repo" });
291
+ return { ...imported, registry: registryReport };
296
292
  }
297
293
  // Read-only: inspect a portable archive's integrity WITHOUT importing it. Routes
298
294
  // both surfaces through one shared core entry. The runner is unused (no registry
299
295
  // touch — inspection writes nothing) but kept for dispatch-signature symmetry.
300
296
  function runInspectArchive(_runner, args) {
301
- return withInvocationCwd(args, () => {
302
- const archive = optionalString(args.archive || args.path || args.file);
303
- if (!archive)
304
- throw new Error("run inspect-archive requires an archive path (positional, --archive, --path, or --file)");
305
- return (0, run_export_1.inspectArchive)(node_path_1.default.resolve(archive));
306
- });
297
+ const base = invocationCwd(args);
298
+ const archive = optionalString(args.archive || args.path || args.file);
299
+ if (!archive)
300
+ throw new Error("run inspect-archive requires an archive path (positional, --archive, --path, or --file)");
301
+ return (0, run_export_1.inspectArchive)(node_path_1.default.resolve(base, archive));
307
302
  }
308
303
  function runVerifyImport(runner, runId, args) {
309
- return withInvocationCwd(args, () => (0, run_export_1.verifyImportedRun)(runner.loadRun(runId)));
304
+ return (0, run_export_1.verifyImportedRun)(runner.withBaseDir(optionalString(args.cwd)).loadRun(runId));
310
305
  }
311
306
  function queueAdd(reg, args) {
312
307
  return reg.queueAdd({
@@ -338,13 +333,12 @@ function queueShow(reg, id) {
338
333
  // ---- control-plane scheduling (v0.1.37) -----------------------------------
339
334
  function loadSchedulingPolicy(reg) {
340
335
  const file = reg.schedulingPolicyPath();
336
+ // Absent policy => conservative DEFAULT (an unconfigured backend, which §4
337
+ // permits). But a PRESENT-but-corrupt policy must fail closed: silently
338
+ // substituting defaults would schedule under settings the operator never
339
+ // chose while their broken file sits on disk. Let readJson's throw surface it.
341
340
  if (node_fs_1.default.existsSync(file)) {
342
- try {
343
- return { policy: (0, scheduling_1.normalizeSchedulingPolicy)((0, state_1.readJson)(file)), source: "file" };
344
- }
345
- catch {
346
- /* fall through to default */
347
- }
341
+ return { policy: (0, scheduling_1.normalizeSchedulingPolicy)((0, state_1.readJson)(file)), source: "file" };
348
342
  }
349
343
  return { policy: scheduling_1.DEFAULT_SCHEDULING_POLICY, source: "default" };
350
344
  }
@@ -405,8 +399,17 @@ function schedPolicySet(reg, args) {
405
399
  const current = loadSchedulingPolicy(reg).policy;
406
400
  const patch = {};
407
401
  for (const key of ["maxConcurrent", "maxAttempts", "leaseTtlMs", "backoffBaseMs", "backoffFactor", "backoffCapMs"]) {
408
- if (args[key] !== undefined)
409
- patch[key] = Number(args[key]);
402
+ if (args[key] === undefined)
403
+ continue;
404
+ // Fail closed on a non-numeric flag instead of letting normalizeSchedulingPolicy
405
+ // silently substitute the DEFAULT (which would report source:"file" + exit 0,
406
+ // so the operator believes they set a value they didn't). Matches the
407
+ // Number.isFinite guard the sibling reclaimPolicyFrom already uses.
408
+ const value = Number(args[key]);
409
+ if (!Number.isFinite(value)) {
410
+ throw new Error(`Invalid --${key} "${String(args[key])}": expected a number (e.g. --${key} 4)`);
411
+ }
412
+ patch[key] = value;
410
413
  }
411
414
  const policy = (0, scheduling_1.normalizeSchedulingPolicy)({ ...current, ...patch });
412
415
  (0, state_1.writeJson)(reg.schedulingPolicyPath(), policy);
@@ -458,26 +461,25 @@ function planInputsFor(args) {
458
461
  * the golden path runs each verb from the repo dir) — then restores cwd. This lets
459
462
  * `cw run <app> --drive --repo X` work when invoked from anywhere. */
460
463
  function runDrive(runner, args) {
461
- const cwd0 = process.cwd();
462
- try {
463
- let runId = optionalString(args.runId || args.run);
464
- let repoCwd = optionalString(args.repo);
465
- if (!runId) {
466
- const appId = String(args.appId || args.workflowId || args.app || "");
467
- if (!appId)
468
- throw new Error("run --drive requires an app id (or --run <run-id> to continue)");
469
- const run = runner.plan(appId, planInputsFor(args));
470
- runId = run.id;
471
- repoCwd = run.cwd;
472
- }
473
- if (repoCwd && repoCwd !== process.cwd() && node_fs_1.default.existsSync(repoCwd))
474
- process.chdir(repoCwd);
475
- return (0, drive_1.drive)(runner, runId, { once: isTrue(args.once), now: optionalString(args.now), args });
476
- }
477
- finally {
478
- if (process.cwd() !== cwd0)
479
- process.chdir(cwd0);
464
+ let runId = optionalString(args.runId || args.run);
465
+ let repoCwd = optionalString(args.repo);
466
+ if (!runId) {
467
+ const appId = String(args.appId || args.workflowId || args.app || "");
468
+ if (!appId)
469
+ throw new Error("run --drive requires an app id (or --run <run-id> to continue)");
470
+ const run = runner.plan(appId, planInputsFor(args));
471
+ runId = run.id;
472
+ repoCwd = run.cwd;
480
473
  }
474
+ // The runner resolves the run from its baseDir, so the drive must run WITH the
475
+ // run's repo as base. Pass it explicitly via withBaseDir (F7 — no process.chdir).
476
+ const target = repoCwd && node_fs_1.default.existsSync(repoCwd) ? repoCwd : undefined;
477
+ const driveRunId = runId;
478
+ return (0, drive_1.drive)(runner.withBaseDir(target), driveRunId, {
479
+ once: isTrue(args.once),
480
+ now: optionalString(args.now),
481
+ args
482
+ });
481
483
  }
482
484
  /** The app the one-command quickstart plans when none is named. */
483
485
  exports.QUICKSTART_DEFAULT_APP = "architecture-review";
@@ -503,23 +505,16 @@ function quickstart(runner, args) {
503
505
  // `--preview`: read-only, deterministic next-step projection (no spawn, no commit).
504
506
  // Plan a fresh run (the read-only first verb) then project the next drive step.
505
507
  if (isTrue(args.preview)) {
506
- const cwd0 = process.cwd();
507
- try {
508
- let runId = optionalString(args.runId || args.run);
509
- let repoCwd = optionalString(args.repo);
510
- if (!runId) {
511
- const run = runner.plan(appId, planInputsFor(args));
512
- runId = run.id;
513
- repoCwd = run.cwd;
514
- }
515
- if (repoCwd && repoCwd !== process.cwd() && node_fs_1.default.existsSync(repoCwd))
516
- process.chdir(repoCwd);
517
- return (0, drive_1.drivePreview)(runner, runId, args);
518
- }
519
- finally {
520
- if (process.cwd() !== cwd0)
521
- process.chdir(cwd0);
508
+ let runId = optionalString(args.runId || args.run);
509
+ let repoCwd = optionalString(args.repo);
510
+ if (!runId) {
511
+ const run = runner.plan(appId, planInputsFor(args));
512
+ runId = run.id;
513
+ repoCwd = run.cwd;
522
514
  }
515
+ const previewRunId = runId;
516
+ const target = repoCwd && node_fs_1.default.existsSync(repoCwd) ? repoCwd : undefined;
517
+ return (0, drive_1.drivePreview)(runner.withBaseDir(target), previewRunId, args);
523
518
  }
524
519
  // Drive end-to-end (or one `--once` step). runDrive plans the run, delegates each
525
520
  // worker to the agent backend, and commits — we add only the report write + a
@@ -530,23 +525,15 @@ function quickstart(runner, args) {
530
525
  const result = runDrive(runner, { ...args, appId, ...(resume && !resumeRunId ? { once: true } : {}) });
531
526
  // Always (re)write the report so the one command yields a report.md on disk, even
532
527
  // when the drive blocked/parked (a partial report is still useful triage).
533
- const cwd0 = process.cwd();
534
- let reportPath = result.reportPath;
535
- try {
536
- // runDrive restored cwd, so the runs root would resolve against the CALLER's
537
- // cwd here orphaning the run when quickstart is invoked cross-directory
538
- // (cwd = plugin dir, --repo elsewhere: the README's headline command). The
539
- // run's statePath (<repo>/.cw/runs/<id>/state.json) is authoritative however
540
- // the run was planned or continued; chdir to ITS repo BEFORE any run read.
541
- const runRepoCwd = node_path_1.default.resolve(node_path_1.default.dirname(result.statePath), "..", "..", "..");
542
- if (runRepoCwd !== process.cwd() && node_fs_1.default.existsSync(runRepoCwd))
543
- process.chdir(runRepoCwd);
544
- reportPath = runner.report(result.runId).path;
545
- }
546
- finally {
547
- if (process.cwd() !== cwd0)
548
- process.chdir(cwd0);
549
- }
528
+ //
529
+ // runDrive restored cwd, so the runs root would resolve against the CALLER's cwd
530
+ // here — orphaning the run when quickstart is invoked cross-directory (cwd =
531
+ // plugin dir, --repo elsewhere: the README's headline command). The run's
532
+ // statePath (<repo>/.cw/runs/<id>/state.json) is authoritative however the run
533
+ // was planned or continued; resolve to ITS repo BEFORE any run read, reentrant-safe.
534
+ const runRepoCwd = node_path_1.default.resolve(node_path_1.default.dirname(result.statePath), "..", "..", "..");
535
+ const reportTarget = node_fs_1.default.existsSync(runRepoCwd) ? runRepoCwd : undefined;
536
+ const reportPath = runner.withBaseDir(reportTarget).report(result.runId).path;
550
537
  let hint;
551
538
  if (!agentConfigured) {
552
539
  hint =
@@ -28,6 +28,7 @@ exports.CAPABILITY_REGISTRY = void 0;
28
28
  exports.declaredMcpTools = declaredMcpTools;
29
29
  exports.declaredCliTokens = declaredCliTokens;
30
30
  exports.requiresReason = requiresReason;
31
+ exports.isPayloadProbeOptOut = isPayloadProbeOptOut;
31
32
  exports.payloadIdenticalCapabilities = payloadIdenticalCapabilities;
32
33
  exports.buildParityReport = buildParityReport;
33
34
  // ---------------------------------------------------------------------------
@@ -51,6 +52,14 @@ const BUILTIN_CAPABILITIES = [
51
52
  cli: { path: ["list"], jsonMode: "default" },
52
53
  mcp: { tool: "cw_list" }
53
54
  },
55
+ {
56
+ capability: "doctor",
57
+ summary: "Diagnose the host for setup problems (Node version, agent backend, agent binary on PATH, git, writable home/repo state) and print an actionable fix per check.",
58
+ entry: "runDoctor",
59
+ surface: "cli-only",
60
+ cli: { path: ["doctor"], jsonMode: "flag" },
61
+ reason: "Environment diagnostics are inherently local to the CLI host — Node version, $PATH, $CW_HOME/cwd writability. An MCP client diagnosing the server process's environment is not meaningful; agents already receive the same readiness facts in their typed results (e.g. status: blocked, agentConfigured). Inspired by `brew doctor`."
62
+ },
54
63
  {
55
64
  capability: "init",
56
65
  summary: "Scaffold a new workflow definition.",
@@ -428,10 +437,28 @@ function requiresReason(cap) {
428
437
  return true;
429
438
  return false;
430
439
  }
431
- /** Descriptors for the payload-identity probe: both-surface, identical payloads,
432
- * read-only (safe to call on a planned run with just runId-style args). */
440
+ /**
441
+ * Whether a `surface:"both"` capability is DOCUMENTED out of the payload-identity
442
+ * probe. The probe defaults capabilities IN (every both-surface, dual-bound verb —
443
+ * including write/complex-arg verbs) and requires an EXPLICIT, REASONED opt-out to
444
+ * fall out of scope. A capability escapes the probe only when it carries BOTH
445
+ * `payloadIdentical: false` AND a non-empty `reason`. A bare `payloadIdentical:
446
+ * false` with no recorded reason does NOT silently escape — it stays in the probe
447
+ * set so the undocumented divergence trips the gate (FAIL CLOSED). This mirrors the
448
+ * `requiresReason`/`reasonlessExceptions` gate, but enforces it at the probe-scope
449
+ * boundary too, so the marshalling-drift surface cannot be narrowed without a paper
450
+ * trail.
451
+ */
452
+ function isPayloadProbeOptOut(cap) {
453
+ return cap.payloadIdentical === false && !!(cap.reason && cap.reason.trim());
454
+ }
455
+ /** Descriptors for the payload-identity probe. Defaults to EVERY both-surface,
456
+ * dual-bound capability (read OR write); a capability is excluded only by a
457
+ * documented opt-out (`payloadIdentical: false` + a non-empty `reason`) — see
458
+ * `isPayloadProbeOptOut`. Fail-closed: an undocumented `payloadIdentical: false`
459
+ * stays in scope so its divergence is caught, not silently excused. */
433
460
  function payloadIdenticalCapabilities() {
434
- return exports.CAPABILITY_REGISTRY.filter((cap) => cap.surface === "both" && cap.payloadIdentical !== false && cap.cli && cap.mcp);
461
+ return exports.CAPABILITY_REGISTRY.filter((cap) => cap.surface === "both" && cap.cli && cap.mcp && !isPayloadProbeOptOut(cap));
435
462
  }
436
463
  function lintRegistry() {
437
464
  const issues = [];
package/dist/cli.js CHANGED
@@ -21,6 +21,7 @@ const multi_agent_operator_ux_1 = require("./multi-agent-operator-ux");
21
21
  const multi_agent_eval_1 = require("./multi-agent-eval");
22
22
  const state_explosion_1 = require("./state-explosion");
23
23
  const evidence_reasoning_1 = require("./evidence-reasoning");
24
+ const doctor_1 = require("./doctor");
24
25
  async function main() {
25
26
  const args = (0, orchestrator_1.parseArgv)(process.argv.slice(2));
26
27
  const runner = new orchestrator_1.CoolWorkflowRunner({
@@ -36,6 +37,16 @@ async function main() {
36
37
  case "list":
37
38
  printJson(runner.listWorkflows());
38
39
  return;
40
+ case "doctor": {
41
+ const report = (0, doctor_1.runDoctor)(args.options, process.env, String(args.options.cwd || process.cwd()));
42
+ if (wantsJson(args.options))
43
+ printJson(report);
44
+ else
45
+ process.stdout.write(`${(0, doctor_1.formatDoctorReport)(report)}\n`);
46
+ if (!report.ok)
47
+ process.exitCode = 1;
48
+ return;
49
+ }
39
50
  case "init": {
40
51
  const [workflowId] = args.positionals;
41
52
  if (!workflowId)
@@ -1281,7 +1292,7 @@ async function main() {
1281
1292
  }
1282
1293
  function required(value, label) {
1283
1294
  if (!value)
1284
- throw new Error(`Missing ${label}`);
1295
+ throw new Error(`Missing ${label}. Run "cw.js help" for usage.`);
1285
1296
  return value;
1286
1297
  }
1287
1298
  function optionalArg(value) {