@sublang/playbook 13.3.0 → 14.0.0

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 (32) hide show
  1. package/README.md +5 -5
  2. package/docs/configuration.md +10 -2
  3. package/docs/embedding.md +20 -1
  4. package/package.json +6 -1
  5. package/reference/sdlc/code.playbook/bin/launch-config.js +88 -5
  6. package/reference/sdlc/code.playbook/bin/playbook.js +4 -0
  7. package/reference/sdlc/code.playbook/bin/repository-effects.js +11 -3
  8. package/reference/sdlc/code.playbook/bin/run.js +4 -0
  9. package/reference/sdlc/code.playbook/bin/session-store.js +2 -2
  10. package/reference/sdlc/code.playbook/playbook-captain.js +17 -4
  11. package/reference/sdlc/code.playbook/playbook-captain.ts +19 -4
  12. package/reference/sdlc/dev.playbook/dev.playbook.d.ts +3 -10
  13. package/reference/sdlc/dev.playbook/dev.playbook.js +5 -5
  14. package/reference/sdlc/dev.playbook/dev.playbook.ts +7 -7
  15. package/reference/sdlc/pr.md +17 -4
  16. package/reference/sdlc/pr.playbook/pr.fsm.d.ts +17 -3
  17. package/reference/sdlc/pr.playbook/pr.fsm.js +49 -18
  18. package/reference/sdlc/pr.playbook/pr.fsm.ts +63 -20
  19. package/reference/sdlc/pr.playbook/pr.gears.md +20 -7
  20. package/reference/sdlc/pr.playbook/pr.playbook.js +2 -1
  21. package/reference/sdlc/pr.playbook/pr.playbook.ts +2 -1
  22. package/slc/experiments/fsm-scaffold-guidance.md +31 -0
  23. package/slc/gears2fsm.md +145 -58
  24. package/slc/link.md +158 -7
  25. package/slc/materialize-link.mjs +460 -0
  26. package/slc/optimize.md +46 -6
  27. package/slc/scaffold-fsm.mjs +225 -0
  28. package/slc/slc.pin-inputs.json +13 -0
  29. package/slc/text2gears.md +38 -3
  30. package/slc/workflow-contracts.json +371 -0
  31. package/src/xstate-playbook-runtime.js +35 -1
  32. package/src/xstate-playbook-runtime.ts +54 -1
package/README.md CHANGED
@@ -32,9 +32,10 @@ Run `playbook` for an interactive tmux UI powered by [cligent](https://github.co
32
32
  Out of the box, Playbook includes **CODE** for implementation, **REVIEW** for commit-based review and fixes, **DECIDE** for independently proposed and reviewed specification decisions, **DEV** for repository-aware planning of a development request, **BRANCH** for checking out a new branch for a GitHub issue, and **PR** for publishing, checking, and merging that branch through a pull request.
33
33
  CODE and DECIDE call REVIEW as a nested playbook; DEV calls CODE, or DECIDE and then CODE, and wraps them in BRANCH and PR when the request names a GitHub issue or asks for a pull request.
34
34
 
35
- The shared starter config uses Claude as both Captain and the `dev.coder`
36
- player, and Codex as `dev.reviewer`. CODE, REVIEW, DECIDE, BRANCH, and PR bind
37
- their local roles explicitly to those two stable players, so nested and later
35
+ The shared starter config selects one adapter for the Captain and all players
36
+ from locally visible credentials: Claude first, then Codex. If neither is
37
+ configured, it seeds Claude and prints a notice. CODE, REVIEW, DECIDE, BRANCH, and PR bind
38
+ their local roles explicitly to `dev.coder` and `dev.reviewer`, so nested and later
38
39
  engagements share a conversation only where their bindings name the same
39
40
  player ID; DEV's Analyst binds a distinct `dev.analyst` player.
40
41
 
@@ -49,8 +50,7 @@ If an SDK is missing or older than cligent supports, Playbook prints the pinned
49
50
  Prerequisites:
50
51
 
51
52
  - Node.js >= 20.6.0
52
- - Authenticated [Claude Code](https://docs.anthropic.com/en/docs/claude-code/overview) or `ANTHROPIC_API_KEY`
53
- - Authenticated [Codex CLI](https://github.com/openai/codex) or `OPENAI_API_KEY`
53
+ - Credentials for the configured adapter: authenticated [Claude Code](https://docs.anthropic.com/en/docs/claude-code/overview) or `ANTHROPIC_API_KEY` for Claude; authenticated [Codex CLI](https://github.com/openai/codex) or `OPENAI_API_KEY` for Codex
54
54
  - [GitHub CLI](https://cli.github.com/) (`gh`) authenticated for the repository's GitHub remote, for `/branch`, `/pr`, and `/dev` requests that name an issue or ask for a pull request
55
55
 
56
56
  Interactive `playbook` additionally needs tmux 3.3 or newer and [`glow`](https://github.com/charmbracelet/glow#installation) on `PATH`; headless `playbook run` does not.
@@ -55,8 +55,16 @@ key is that playbook's option slice. Every manifest role must be present
55
55
  exactly once. The launcher injects the rest — you do not write host wiring by
56
56
  hand.
57
57
 
58
- The seeded config runs every stable player on Claude Opus 5 — the Coder at
59
- `high` effort, the Reviewer and Analyst at `xhigh`:
58
+ The launcher seeds one adapter for the Captain and all stable players from
59
+ locally visible credentials: Claude first, then Codex. It uses Claude Opus 5
60
+ for Claude or GPT-5.6 Sol for Codex, with Captain and Coder at `high` effort
61
+ and Reviewer and Analyst at `xhigh`. If neither adapter is configured, it
62
+ seeds Claude and prints a notice. Credentials do not prove SDK availability
63
+ or remaining quota; launch still checks the configured adapter. Existing
64
+ configs are left unchanged.
65
+
66
+ The Claude default is shown below. A Codex seed also adds
67
+ `permissions.writablePaths: ['.git']` to each agent:
60
68
 
61
69
  ```yaml
62
70
  captain:
package/docs/embedding.md CHANGED
@@ -433,7 +433,7 @@ defines the files, versions and compatibility rules.
433
433
 
434
434
  ## Reading the published spec contracts
435
435
 
436
- The authored compiler-phase specs ship in the package and are exposed
436
+ The compiler definitions and supporting assets ship in the package and are exposed
437
437
  as a public, semver-stable surface under `@sublang/playbook/slc/*`.
438
438
  Resolve and read one with `import.meta.resolve` plus `fs`:
439
439
 
@@ -450,3 +450,22 @@ The four specs are [`slc/text2gears.md`](../slc/text2gears.md),
450
450
  [`slc/link.md`](../slc/link.md) — the FSM-to-runtime contract that
451
451
  `@sublang/playbook/runtime` projects into TypeScript — and
452
452
  [`slc/optimize.md`](../slc/optimize.md).
453
+
454
+ Five supporting assets use the same resolution mechanism:
455
+
456
+ | Asset | Purpose |
457
+ | --- | --- |
458
+ | [`slc/workflow-contracts.json`](../slc/workflow-contracts.json) | Public output schemas and default literal target bindings for the builtin REVIEW, DECIDE, CODE, BRANCH, and PR workflows; compilers and embedders can read them without importing workflow implementations. |
459
+ | [`slc/materialize-link.mjs`](../slc/materialize-link.mjs) | Optional CLI that emits a thin linked module from a supported FSM and JSON descriptor; see the profiles and invocation in `slc/link.md`. |
460
+ | [`slc/scaffold-fsm.mjs`](../slc/scaffold-fsm.mjs) | Optional CLI that initializes an incomplete typed FSM from supported GEARS source. |
461
+ | [`slc/experiments/fsm-scaffold-guidance.md`](../slc/experiments/fsm-scaffold-guidance.md) | Opt-in compiler guidance for the initializer, outside ordinary phase discovery. |
462
+ | [`slc/slc.pin-inputs.json`](../slc/slc.pin-inputs.json) | SLC semantic-input closures for `gears2fsm` and `link`, including the workflow catalog and, for `link`, the materializer. Other phases use SLC's inline-input discovery. |
463
+
464
+ For example, resolve `@sublang/playbook/slc/workflow-contracts.json`, read it
465
+ with `readFile`, and parse the result as JSON to inspect the published workflow
466
+ interfaces. Execute the `.mjs` helpers with Node and their documented CLI
467
+ arguments: their supported public interfaces are their resolvable paths and CLI
468
+ contracts. Their JavaScript module exports are internal implementation details
469
+ and carry no SemVer compatibility guarantee.
470
+ The initializer requires further semantic compilation and verification
471
+ before its output is runnable.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@sublang/playbook",
3
- "version": "13.3.0",
3
+ "version": "14.0.0",
4
4
  "type": "module",
5
5
  "description": "Composable XState v5 playbook runtime with compiled Captain, CODE, REVIEW, DECIDE, DEV, BRANCH, and PR workflows driven by GEARS specs.",
6
6
  "license": "Apache-2.0",
@@ -38,6 +38,11 @@
38
38
  "src/xstate-playbook-runtime.js",
39
39
  "src/xstate-playbook-runtime.d.ts",
40
40
  "slc/link.md",
41
+ "slc/materialize-link.mjs",
42
+ "slc/scaffold-fsm.mjs",
43
+ "slc/experiments/fsm-scaffold-guidance.md",
44
+ "slc/slc.pin-inputs.json",
45
+ "slc/workflow-contracts.json",
41
46
  "slc/gears2fsm.md",
42
47
  "slc/text2gears.md",
43
48
  "slc/optimize.md",
@@ -180,7 +180,7 @@ export function resolveLaunchSessionsDir({
180
180
  if (sessionsDir !== undefined) return sessionsDir;
181
181
 
182
182
  if (preparePrimary) {
183
- seedUserConfigIfMissing(userConfigPath, templatePath, onNotice);
183
+ seedUserConfigIfMissing(userConfigPath, templatePath, onNotice, env, homeDir);
184
184
  migrateUserConfigIfRetired(userConfigPath, onNotice);
185
185
  }
186
186
 
@@ -248,8 +248,10 @@ export async function loadLaunchPlan({
248
248
  templatePath = DEFAULT_TEMPLATE_PATH,
249
249
  onNotice = () => {},
250
250
  selectedMembers,
251
+ env = process.env,
252
+ homeDir = env.HOME ?? homedir(),
251
253
  }) {
252
- seedUserConfigIfMissing(userConfigPath, templatePath, onNotice);
254
+ seedUserConfigIfMissing(userConfigPath, templatePath, onNotice, env, homeDir);
253
255
  // PBCLI-22/46: a reopen must not rewrite, validate, or otherwise inspect
254
256
  // config members outside the stored projection.
255
257
  if (selectedMembers === undefined) {
@@ -288,10 +290,12 @@ export async function loadSelectedLaunchPlanDataOnly({
288
290
  structuralProjection,
289
291
  templatePath = DEFAULT_TEMPLATE_PATH,
290
292
  onNotice = () => {},
293
+ env = process.env,
294
+ homeDir = env.HOME ?? homedir(),
291
295
  }) {
292
296
  const stored = validateStoredStructuralProjection(structuralProjection);
293
297
  const selectedMembers = selectedMembersFromStoredStructure(stored);
294
- seedUserConfigIfMissing(userConfigPath, templatePath, onNotice);
298
+ seedUserConfigIfMissing(userConfigPath, templatePath, onNotice, env, homeDir);
295
299
 
296
300
  let top = parseYaml(readFileSync(userConfigPath, "utf8")) ?? {};
297
301
  if (overlayPaths.length > 0 && !isObject(top)) {
@@ -1335,11 +1339,90 @@ function assertLegacyRelocationLocatorsSafe(
1335
1339
  );
1336
1340
  }
1337
1341
 
1338
- function seedUserConfigIfMissing(userConfigPath, templatePath, onNotice) {
1342
+ // PBCLI-11 / DR-053: the adapters a seeded lineup may take, highest
1343
+ // precedence first. A fixed order — not whichever the probe answers
1344
+ // first — is what makes one machine seed one file twice.
1345
+ const SEED_LINEUP = Object.freeze([
1346
+ Object.freeze({ adapter: "claude", model: "claude-opus-5" }),
1347
+ Object.freeze({ adapter: "codex", model: "gpt-5.6-sol" }),
1348
+ ]);
1349
+
1350
+ /**
1351
+ * PBCLI-11: the lineup a seed would write here, and whether anything
1352
+ * actually probed ready.
1353
+ *
1354
+ * Only the credential check participates. The SDK probe constructs an
1355
+ * adapter and awaits `isAvailable()`, which reads whatever happens to
1356
+ * be installed and spawns a subprocess for CLI-backed adapters: a seed
1357
+ * consulting it would differ between a developer's machine and a clean
1358
+ * one, and would make creating a file asynchronous.
1359
+ */
1360
+ export function selectSeedLineup(env = process.env, home = homedir()) {
1361
+ const ready = SEED_LINEUP.filter(
1362
+ (entry) =>
1363
+ checkReadiness([entry.adapter], env, home).failingAdapters.length === 0,
1364
+ );
1365
+ return { lineup: ready[0] ?? SEED_LINEUP[0], anyReady: ready.length > 0 };
1366
+ }
1367
+
1368
+ /**
1369
+ * The template is the documented default and the only source of the
1370
+ * starter's structure and comments, so the default choice is still the
1371
+ * byte copy it always was. A different adapter rewrites the agent
1372
+ * scalars in place, which the Document API carries comments through.
1373
+ */
1374
+ function seedTextForLineup(templatePath, lineup) {
1375
+ const template = readFileSync(templatePath, "utf8");
1376
+ if (lineup.adapter === SEED_LINEUP[0].adapter) return template;
1377
+ const doc = parseYamlDocument(template);
1378
+ const players = doc.getIn(["players"]);
1379
+ const agentPaths = [
1380
+ ["captain"],
1381
+ ...[...(players?.items ?? [])].map((pair) => [
1382
+ "players",
1383
+ String(pair.key),
1384
+ ]),
1385
+ ];
1386
+ for (const path of agentPaths) {
1387
+ doc.setIn([...path, "adapter"], lineup.adapter);
1388
+ doc.setIn([...path, "model"], lineup.model);
1389
+ // A codex agent writes git metadata under its sandbox; claude's
1390
+ // auto mode needs no grant, so the key exists for one adapter only.
1391
+ if (lineup.adapter === "codex") {
1392
+ doc.setIn([...path, "permissions", "writablePaths"], [".git"]);
1393
+ } else {
1394
+ doc.deleteIn([...path, "permissions", "writablePaths"]);
1395
+ }
1396
+ }
1397
+ return String(doc);
1398
+ }
1399
+
1400
+ function seedUserConfigIfMissing(
1401
+ userConfigPath,
1402
+ templatePath,
1403
+ onNotice,
1404
+ env = process.env,
1405
+ homeDir = env.HOME ?? homedir(),
1406
+ ) {
1339
1407
  if (existsSync(userConfigPath)) return;
1340
1408
  mkdirSync(dirname(userConfigPath), { recursive: true });
1341
- copyFileSync(templatePath, userConfigPath, constants.COPYFILE_EXCL);
1409
+ const { lineup, anyReady } = selectSeedLineup(env, homeDir);
1410
+ if (lineup.adapter === SEED_LINEUP[0].adapter && anyReady) {
1411
+ // Unchanged default path: one exclusive create of the template bytes.
1412
+ copyFileSync(templatePath, userConfigPath, constants.COPYFILE_EXCL);
1413
+ } else {
1414
+ // Still one exclusive create, so two seeding calls cannot disagree.
1415
+ writeFileSync(userConfigPath, seedTextForLineup(templatePath, lineup), {
1416
+ encoding: "utf8",
1417
+ flag: "wx",
1418
+ });
1419
+ }
1342
1420
  onNotice(`playbook: created config at ${userConfigPath}\n`);
1421
+ if (!anyReady) {
1422
+ onNotice(
1423
+ `playbook: no adapter probed ready; seeded ${lineup.adapter}\n`,
1424
+ );
1425
+ }
1343
1426
  }
1344
1427
 
1345
1428
  // DR-021 §3: migrate once, keeping the original before any rewrite.
@@ -346,10 +346,14 @@ export async function runPlaybookCli(options = {}) {
346
346
  overlayPaths: withPaths,
347
347
  structuralProjection: selectedRecord.structuralProjection,
348
348
  onNotice: (line) => stderr.write(line),
349
+ env,
350
+ homeDir: home,
349
351
  })
350
352
  : await loadLaunchPlan({
351
353
  userConfigPath,
352
354
  overlayPaths: withPaths,
355
+ env,
356
+ homeDir: home,
353
357
  loadModule,
354
358
  prepareRegistryModule:
355
359
  options.prepareRegistryModule ??
@@ -659,10 +659,18 @@ export async function resolveCanonicalGitWorktree(cwd) {
659
659
  }
660
660
 
661
661
  async function sampleWorktree(worktree, env) {
662
- const head = await resolveHead(worktree, env);
663
- const indexVisibility = await rawIndexVisibility(worktree, env);
662
+ const [headResult, visibilityResult, statusResult] = await Promise.allSettled([
663
+ resolveHead(worktree, env),
664
+ rawIndexVisibility(worktree, env),
665
+ rawRepositoryStatus(worktree, env),
666
+ ]);
667
+ if (headResult.status === 'rejected') throw headResult.reason;
668
+ if (visibilityResult.status === 'rejected') throw visibilityResult.reason;
669
+ const head = headResult.value;
670
+ const indexVisibility = visibilityResult.value;
664
671
  assertIndexVisibility(indexVisibility);
665
- const status = await rawRepositoryStatus(worktree, env);
672
+ if (statusResult.status === 'rejected') throw statusResult.reason;
673
+ const status = statusResult.value;
666
674
  const projection = await projectionFromStatus(worktree, status);
667
675
  return { head, indexVisibility, status, projection };
668
676
  }
@@ -359,10 +359,14 @@ export async function runPlaybookRun(options = {}) {
359
359
  overlayPaths: args.withPaths,
360
360
  structuralProjection: priorRecord.structuralProjection,
361
361
  onNotice: (line) => configNotices.push(line),
362
+ env,
363
+ homeDir: home,
362
364
  })
363
365
  : await loadLaunchPlan({
364
366
  userConfigPath,
365
367
  overlayPaths: args.withPaths,
368
+ env,
369
+ homeDir: home,
366
370
  loadModule,
367
371
  prepareRegistryModule,
368
372
  onNotice: (line) => configNotices.push(line),
@@ -3163,7 +3163,7 @@ async function createLease({
3163
3163
  }
3164
3164
  if (
3165
3165
  snapshot.mode !== 'chat' ||
3166
- snapshot.lastAction !== 'runtime' ||
3166
+ (snapshot.lastAction !== 'runtime' && snapshot.lastAction !== 'dismiss') ||
3167
3167
  snapshot.lastSettlementStatus !== 'ok'
3168
3168
  ) {
3169
3169
  throw new Error(
@@ -4221,7 +4221,7 @@ function validateCanonicalCaptainSessionRecord(
4221
4221
  }
4222
4222
  if (
4223
4223
  snapshot.mode !== 'chat' ||
4224
- snapshot.lastAction !== 'runtime' ||
4224
+ (snapshot.lastAction !== 'runtime' && snapshot.lastAction !== 'dismiss') ||
4225
4225
  snapshot.lastSettlementStatus !== 'ok' ||
4226
4226
  Object.hasOwn(record.retainedGenerations ?? {}, rootPlaybookId)
4227
4227
  ) {
@@ -3321,6 +3321,7 @@ export function createPlaybookCaptainShell(options, deps = {}) {
3321
3321
  * may return an executed receipt to its result phase.
3322
3322
  */
3323
3323
  const settleUnresolvedEffectAbandonment = async (unresolvedLeaf) => {
3324
+ abandonmentSettlementUnsafe = true;
3324
3325
  if (leafFrame() !== unresolvedLeaf) {
3325
3326
  throw new Error('unresolved-effect abandonment requires the active leaf');
3326
3327
  }
@@ -3347,6 +3348,7 @@ export function createPlaybookCaptainShell(options, deps = {}) {
3347
3348
  rootPlaybookId,
3348
3349
  });
3349
3350
  await runEffect(() => unresolvedEffectSettlement.complete(settlement));
3351
+ abandonmentSettlementUnsafe = false;
3350
3352
  };
3351
3353
  const callResultFor = (frame, result) => {
3352
3354
  if (result.outcome === 'unresolved-effect') {
@@ -4693,6 +4695,12 @@ export function createPlaybookCaptainShell(options, deps = {}) {
4693
4695
  if (!root)
4694
4696
  return false;
4695
4697
  const label = frameLabel(root);
4698
+ if (activeTurn?.hostGiveUp === true &&
4699
+ freezeTurnUnresolvedEffects().length > 0) {
4700
+ await settleUnresolvedEffectAbandonment(leafFrame());
4701
+ facts.push(`Dismissed the ${label} engagement with unresolved repository effects reported.`);
4702
+ return false;
4703
+ }
4696
4704
  // Dismissal leaves the procedure unfinished. Persist the latest safe
4697
4705
  // generation captured for this turn before disposal erases the frames —
4698
4706
  // unless the Boss gave up on it (CAPTAIN-44), in which case retaining it
@@ -5066,7 +5074,8 @@ export function createPlaybookCaptainShell(options, deps = {}) {
5066
5074
  selection.actionId === UNRESOLVED_EFFECT_ABANDONMENT_ACTION_ID) &&
5067
5075
  fencedLeaf !== undefined;
5068
5076
  if (retainedEffectReconciliation !== undefined &&
5069
- !routesRetainedReconciliation) {
5077
+ !routesRetainedReconciliation &&
5078
+ !(selection.action === 'dismiss' && turn.hostGiveUp === true)) {
5070
5079
  return rejectSelection(selection, 'retained work must reconcile its repository-effect evidence before an ordinary action can run');
5071
5080
  }
5072
5081
  if (selection.action === 'resume') {
@@ -5359,7 +5368,6 @@ export function createPlaybookCaptainShell(options, deps = {}) {
5359
5368
  await settleUnresolvedEffectAbandonment(leaf);
5360
5369
  }
5361
5370
  catch (error) {
5362
- abandonmentSettlementUnsafe = true;
5363
5371
  const normalized = normalizeErrorCompact(error) ?? {
5364
5372
  name: 'Error',
5365
5373
  message: String(error),
@@ -5473,8 +5481,13 @@ export function createPlaybookCaptainShell(options, deps = {}) {
5473
5481
  const giveUpReplyText = () => {
5474
5482
  const report = activeTurn?.report;
5475
5483
  const facts = report?.bossFacts ?? report?.facts ?? [];
5476
- const opening = 'Stopped that workflow. I will not resume it.';
5477
- const closing = 'Send the command again to start fresh work.';
5484
+ const stopped = report?.status === 'ok';
5485
+ const opening = stopped
5486
+ ? 'Stopped that workflow. I will not resume it.'
5487
+ : 'I could not confirm that the workflow was stopped.';
5488
+ const closing = stopped
5489
+ ? 'Send the command again to start fresh work.'
5490
+ : 'The stop did not settle successfully; recover the session before continuing.';
5478
5491
  const composed = [
5479
5492
  opening,
5480
5493
  ...(facts.length === 0
@@ -5229,6 +5229,7 @@ export function createPlaybookCaptainShell(
5229
5229
  const settleUnresolvedEffectAbandonment = async (
5230
5230
  unresolvedLeaf: EngagementFrame,
5231
5231
  ): Promise<void> => {
5232
+ abandonmentSettlementUnsafe = true;
5232
5233
  if (leafFrame() !== unresolvedLeaf) {
5233
5234
  throw new Error(
5234
5235
  'unresolved-effect abandonment requires the active leaf',
@@ -5263,6 +5264,7 @@ export function createPlaybookCaptainShell(
5263
5264
  rootPlaybookId,
5264
5265
  });
5265
5266
  await runEffect(() => unresolvedEffectSettlement.complete(settlement));
5267
+ abandonmentSettlementUnsafe = false;
5266
5268
  };
5267
5269
 
5268
5270
  const callResultFor = (
@@ -6913,6 +6915,14 @@ export function createPlaybookCaptainShell(
6913
6915
  const root = rootFrame();
6914
6916
  if (!root) return false;
6915
6917
  const label = frameLabel(root);
6918
+ if (
6919
+ activeTurn?.hostGiveUp === true &&
6920
+ freezeTurnUnresolvedEffects().length > 0
6921
+ ) {
6922
+ await settleUnresolvedEffectAbandonment(leafFrame()!);
6923
+ facts.push(`Dismissed the ${label} engagement with unresolved repository effects reported.`);
6924
+ return false;
6925
+ }
6916
6926
  // Dismissal leaves the procedure unfinished. Persist the latest safe
6917
6927
  // generation captured for this turn before disposal erases the frames —
6918
6928
  // unless the Boss gave up on it (CAPTAIN-44), in which case retaining it
@@ -7368,7 +7378,8 @@ export function createPlaybookCaptainShell(
7368
7378
  fencedLeaf !== undefined;
7369
7379
  if (
7370
7380
  retainedEffectReconciliation !== undefined &&
7371
- !routesRetainedReconciliation
7381
+ !routesRetainedReconciliation &&
7382
+ !(selection.action === 'dismiss' && turn.hostGiveUp === true)
7372
7383
  ) {
7373
7384
  return rejectSelection(
7374
7385
  selection,
@@ -7726,7 +7737,6 @@ export function createPlaybookCaptainShell(
7726
7737
  try {
7727
7738
  await settleUnresolvedEffectAbandonment(leaf);
7728
7739
  } catch (error) {
7729
- abandonmentSettlementUnsafe = true;
7730
7740
  const normalized = normalizeErrorCompact(error) ?? {
7731
7741
  name: 'Error',
7732
7742
  message: String(error),
@@ -7852,8 +7862,13 @@ export function createPlaybookCaptainShell(
7852
7862
  const giveUpReplyText = (): string => {
7853
7863
  const report = activeTurn?.report;
7854
7864
  const facts = report?.bossFacts ?? report?.facts ?? [];
7855
- const opening = 'Stopped that workflow. I will not resume it.';
7856
- const closing = 'Send the command again to start fresh work.';
7865
+ const stopped = report?.status === 'ok';
7866
+ const opening = stopped
7867
+ ? 'Stopped that workflow. I will not resume it.'
7868
+ : 'I could not confirm that the workflow was stopped.';
7869
+ const closing = stopped
7870
+ ? 'Send the command again to start fresh work.'
7871
+ : 'The stop did not settle successfully; recover the session before continuing.';
7857
7872
  const composed = [
7858
7873
  opening,
7859
7874
  ...(facts.length === 0
@@ -1,19 +1,12 @@
1
- import { type XStatePlaybookRuntimeFactory, type XStatePlaybookRuntimeConstruction } from '@sublang/playbook/xstate-runtime';
2
- import { type DevInput, type PlayerInput } from './dev.fsm.js';
1
+ import { type PlaybookPlayerInput, type XStatePlaybookRuntimeFactory, type XStatePlaybookRuntimeConstruction } from '@sublang/playbook/xstate-runtime';
2
+ import { type DevInput } from './dev.fsm.js';
3
3
  import type { PlaybookHostConstructionCapabilities } from '../code.playbook/playbook-captain.js';
4
4
  import type { CaptainCallOptions, CaptainResult, JsonValue, NormalizedError, PlayerCallOptions, PlaybookCallRequest, PlaybookCallResult, PlaybookCallStart, PlaybookControlReceipt, PlaybookControlView, PlaybookPendingCall, PlaybookPorts, PlaybookRunResult, PlaybookRuntime, PlaybookRuntimeFactory, PlaybookRuntimeSnapshot, PlaybookSession, PlaybookState, PlaybookStateValue, PlaybookTraceEvent, PlaybookTraceType, PlayerResult, PlayerSessionStore } from '@sublang/playbook/runtime';
5
5
  export type { CaptainCallOptions, CaptainResult, JsonValue, NormalizedError, PlayerCallOptions, PlaybookCallRequest, PlaybookCallResult, PlaybookCallStart, PlaybookControlReceipt, PlaybookControlView, PlaybookPendingCall, PlaybookPorts, PlaybookRunResult, PlaybookRuntime, PlaybookRuntimeFactory, PlaybookRuntimeSnapshot, PlaybookSession, PlaybookState, PlaybookStateValue, PlaybookTraceEvent, PlaybookTraceType, PlayerResult, PlayerSessionStore, };
6
6
  export type DevPlaybookOptions = DevInput;
7
7
  export type DevPlaybookHostCapabilities = PlaybookHostConstructionCapabilities & XStatePlaybookRuntimeConstruction<DevPlaybookOptions, object>['hostCapabilities'];
8
- /**
9
- * Preserve authored Markdown quote markers around every line of relayed
10
- * runtime text. The generic composer preserves the marker itself; DEV's
11
- * override additionally keeps a multiline value inside that quote and drops
12
- * the optional relays that have no value yet.
13
- */
14
- declare function composePlayerPrompt(input: PlayerInput, resuming?: boolean): string;
15
8
  export declare const _internal: {
16
- composePlayerPrompt: typeof composePlayerPrompt;
9
+ composePlayerPrompt: (input: PlaybookPlayerInput, _identity: import("@sublang/playbook/xstate-runtime").XStatePromptIdentity, resuming?: boolean) => string;
17
10
  VERBATIM_PAYLOAD_FIELDS: ReadonlySet<string>;
18
11
  UNFINISHED_FINAL_STATE_IDS: ReadonlySet<string>;
19
12
  };
@@ -73,11 +73,6 @@ function composePlayerPrompt(input, resuming = false) {
73
73
  });
74
74
  return composePlayerContinuation(input, body, resuming);
75
75
  }
76
- export const _internal = {
77
- composePlayerPrompt,
78
- VERBATIM_PAYLOAD_FIELDS,
79
- UNFINISHED_FINAL_STATE_IDS,
80
- };
81
76
  const runtimeSpec = {
82
77
  label: 'DEV',
83
78
  // DR-022 / slc/link.md: the declaration carries the value current at link
@@ -141,5 +136,10 @@ const runtimeSpec = {
141
136
  unfinishedFinalStateIds: UNFINISHED_FINAL_STATE_IDS,
142
137
  transitionEventFields: ['developmentRequest', 'answer', 'questionId'],
143
138
  };
139
+ export const _internal = {
140
+ composePlayerPrompt: runtimeSpec.composePlayerPrompt,
141
+ VERBATIM_PAYLOAD_FIELDS,
142
+ UNFINISHED_FINAL_STATE_IDS,
143
+ };
144
144
  const createPlaybookRuntime = createXStatePlaybookRuntime(devMachine, runtimeSpec);
145
145
  export default createPlaybookRuntime;
@@ -160,12 +160,6 @@ function composePlayerPrompt(input: PlayerInput, resuming = false): string {
160
160
  return composePlayerContinuation(input, body, resuming);
161
161
  }
162
162
 
163
- export const _internal = {
164
- composePlayerPrompt,
165
- VERBATIM_PAYLOAD_FIELDS,
166
- UNFINISHED_FINAL_STATE_IDS,
167
- };
168
-
169
163
  const runtimeSpec = {
170
164
  label: 'DEV',
171
165
  // DR-022 / slc/link.md: the declaration carries the value current at link
@@ -224,7 +218,7 @@ const runtimeSpec = {
224
218
  },
225
219
  },
226
220
  },
227
- composePlayerPrompt: (input: PlaybookPlayerInput, _identity, resuming) =>
221
+ composePlayerPrompt: (input: PlaybookPlayerInput, _identity, resuming?: boolean) =>
228
222
  composePlayerPrompt(input as PlayerInput, resuming),
229
223
  verbatimPayloadFields: VERBATIM_PAYLOAD_FIELDS,
230
224
  controlContextFields: [],
@@ -232,6 +226,12 @@ const runtimeSpec = {
232
226
  transitionEventFields: ['developmentRequest', 'answer', 'questionId'],
233
227
  } satisfies XStatePlaybookRuntimeSpecV3<DevPlaybookOptions>;
234
228
 
229
+ export const _internal = {
230
+ composePlayerPrompt: runtimeSpec.composePlayerPrompt,
231
+ VERBATIM_PAYLOAD_FIELDS,
232
+ UNFINISHED_FINAL_STATE_IDS,
233
+ };
234
+
235
235
  const createPlaybookRuntime: XStatePlaybookRuntimeFactory<
236
236
  XStatePlaybookRuntimeConstruction<
237
237
  DevPlaybookOptions,
@@ -19,7 +19,9 @@ The caller supplies:
19
19
  `pr` delivers a reviewed branch into the repository default branch through a GitHub pull request: it publishes the branch, waits for the pull request's checks, fixes red checks through playbook `code` no more than once, and merges.
20
20
  It changes no files and owns no repository commit; the one fix it may request is owned by `code`.
21
21
  The check waits, the fix publication, the merge, and the local update are mechanical steps: each runs one fixed command whose exit status alone decides its two outcomes, reads no conversation, and produces no prose.
22
- `gh` infers the pull request from the checked-out branch, so no mechanical step needs a runtime value.
22
+ `gh` infers the pull request from the checked-out branch, which the run does not own: the nested `code` call suspends across Boss turns, so the checkout can change before the fix is published or the merge runs.
23
+ The two steps that act on the pull request itself therefore carry one runtime value — the pull request `pr` published — and refuse unless the checkout still infers exactly it.
24
+ That identity is reported text, so it binds as a single-quoted shell literal: the command compares it as data and never executes it as shell syntax.
23
25
 
24
26
  When the caller gives its input, Captain shall relay the complete caller input in quotes (`>`) to Coder, along with the following instruction:
25
27
 
@@ -66,6 +68,7 @@ When the nested `code` call fails outside that authored result contract, `pr` sh
66
68
 
67
69
  When `code` succeeds, Captain shall publish the fix to the pull request by running exactly the following command in the repository, with no other action and without reading its output:
68
70
 
71
+ > [ "$(gh pr view --json url --jq .url)" = '<pull-request-url>' ] || exit 1
69
72
  > git push || exit 1
70
73
  > n=0
71
74
  > until [ "$(gh pr view --json headRefOid --jq .headRefOid 2>/dev/null)" = "$(git rev-parse HEAD)" ]; do
@@ -75,6 +78,7 @@ When `code` succeeds, Captain shall publish the fix to the pull request by runni
75
78
  > done
76
79
 
77
80
  The publication has exactly two outcomes decided by the exit status alone: fix published on status zero, once the pull request's head is the pushed commit, and fix not published otherwise.
81
+ The command pushes nothing until the checked-out branch infers the pull request `pr` published, so a checkout that changed while `code` ran publishes to no other branch.
78
82
  Fix not published is an authored failure that leaves the pull request open.
79
83
 
80
84
  When the fix is published, Captain shall wait for the pull request's checks again by running exactly the same command as the first wait in the repository, with no other action and without reading its output:
@@ -92,11 +96,20 @@ Checks still failing is an authored failure that leaves the pull request open; t
92
96
 
93
97
  When the checks pass, before or after the one fix attempt, Captain shall merge the pull request by running exactly the following command in the repository, with no other action and without reading its output:
94
98
 
95
- > gh pr merge --merge --delete-branch --match-head-commit "$(git rev-parse HEAD)"
99
+ > pr=$(gh pr view --json url --jq .url) || exit 1
100
+ > [ "$pr" = '<pull-request-url>' ] || exit 1
101
+ > base=$(gh repo view --json defaultBranchRef --jq .defaultBranchRef.name) || exit 1
102
+ > [ -n "$base" ] || exit 1
103
+ > [ "$(gh pr view "$pr" --json baseRefName --jq .baseRefName)" = "$base" ] || exit 1
104
+ > gh pr merge --merge --delete-branch --match-head-commit "$(git rev-parse HEAD)" || exit 1
105
+ > [ "$(gh pr view "$pr" --json state --jq .state)" = MERGED ] || exit 1
106
+ > [ "$(git branch --show-current)" = "$base" ]
96
107
 
97
- The merge creates a merge commit on the repository default branch, deletes the remote and local branch, and checks out the local default branch; GitHub closes the linked issue on merge.
108
+ The merge creates a merge commit on the repository default branch, requests deletion of the remote and local branch, and checks out the local default branch; GitHub closes the linked issue on merge.
98
109
  The merge has exactly two outcomes decided by the exit status alone: merged on status zero and merge refused otherwise.
99
- Merge refused is an authored failure that leaves the pull request in the state GitHub reports: GitHub refused the merge for a conflict, a branch protection, a forbidden merge method, or a head that moved, or the merge landed but the local switch to the default branch or the branch deletion failed.
110
+ The command requires the inferred pull request to be the one `pr` published and to target the repository default branch before the irreversible merge, and confirms that same pull request's merged state and the default-branch checkout after the merge command succeeds; a queued pull request is not a merged result.
111
+ Merge refused is an authored failure that leaves the pull request in the state GitHub reports: the checkout no longer infers the published pull request, the pull request targets another branch, GitHub refused the merge for a conflict, a branch protection, a forbidden merge method, or a head that moved, or the merge may have landed while its confirmation, the local switch to the default branch, or the branch deletion failed.
112
+ A refused merge does not establish that the pull request is unmerged, so `pr` reports it as an unconfirmed merge rather than as not merged, and no outcome claims a branch was deleted, because `gh` skips the remote deletion for a pull request from another repository or one already merged and exits zero anyway.
100
113
 
101
114
  When the pull request is merged, Captain shall bring the local default branch to the merged head by running exactly the following command in the repository, with no other action and without reading its output:
102
115
 
@@ -113,11 +113,21 @@ export type PrPlaybookOutput = {
113
113
  readonly childResult: CompletedChildResult;
114
114
  } | {
115
115
  readonly status: 'not-merged';
116
- readonly reason: 'fix-not-published' | 'checks-failed' | 'merge-refused';
116
+ readonly reason: 'fix-not-published' | 'checks-failed';
117
+ readonly pullRequest: string;
118
+ readonly pullRequestUrl: string;
119
+ }
120
+ /**
121
+ * The merge command exited nonzero. It ran, so whether GitHub merged the
122
+ * pull request is unknown — script output never enters context — and the
123
+ * result claims neither a merge nor its absence.
124
+ */
125
+ | {
126
+ readonly status: 'merge-unconfirmed';
117
127
  readonly pullRequest: string;
118
128
  readonly pullRequestUrl: string;
119
129
  };
120
- export type PrCompletion = 'merged' | 'not-published' | 'fix-failed' | 'fix-not-published' | 'checks-failed' | 'merge-refused';
130
+ export type PrCompletion = 'merged' | 'not-published' | 'fix-failed' | 'fix-not-published' | 'checks-failed' | 'merge-unconfirmed';
121
131
  /** The machine reads no input: the caller input arrives on `START_PR`. */
122
132
  export type PrInput = Readonly<Record<never, never>>;
123
133
  export type PrContext = {
@@ -290,7 +300,11 @@ export declare const prMachine: import("xstate").StateMachine<PrContext, {
290
300
  readonly childResult: CompletedChildResult;
291
301
  } | {
292
302
  readonly status: "not-merged";
293
- readonly reason: "fix-not-published" | "checks-failed" | "merge-refused";
303
+ readonly reason: "fix-not-published" | "checks-failed";
304
+ readonly pullRequest: string;
305
+ readonly pullRequestUrl: string;
306
+ } | {
307
+ readonly status: "merge-unconfirmed";
294
308
  readonly pullRequest: string;
295
309
  readonly pullRequestUrl: string;
296
310
  }, import("xstate").EventObject, import("xstate").MetaObject, {