mandrel 2.57.0 → 2.59.0

package/.agents/workflows/mandrel-plan.md

@@ -11,7 +11,7 @@ description:
 
 ## Inputs
 
-Single planning path — there is no Epic/Story router, no scope-triage
+Single planning path — there is no Epic/Story router, no split-triage
 `epic|story` verdict (Gate #3's container groups, never routes). **Derive the
 mode from what the operator typed, announce it, act**:
 
@@ -57,7 +57,8 @@ and derives source ids from its `sourceTickets[]`; it also writes
 **`stories.template.json`**, step 2's skeleton.
 
 The envelope carries docs context, the story-author prompt (`systemPrompts.story`,
-plus `systemPrompts.storySplitRules` for an N>1 draft), `sourceTickets[]`,
+plus `systemPrompts.storySplitRules` for an N>1 draft and
+`systemPrompts.storyTicketsRules` in tickets mode), `sourceTickets[]`,
 `duplicates[]` (open **Stories**, never Epics), `epicCandidates[]` +
 `dependencyCandidates[]` (Gate #3; path collisions), `priorFeedback` and
 advisory `complexitySignals` (**no routing authority**). An envelope over the
@@ -92,12 +93,14 @@ included. One rescue: a **never-tracked** one normalises to `creates`. Fields:
 [ref](helpers/plan-reference.md).
 
 Artifacts under `temp/plan-<slug>/`: `stories.json` (**length 1 by default**;
-a Spec is as long as the work needs, inline, never under `docs/`); optional
-`techspec.md` (**N===1 only**, folded into `## Spec`) and
-`acceptance-manifest.json` (N>1 — `--plan-acceptance`). Use the envelope
+a Spec is as long as the work needs, inline, never under `docs/`) and optional
+`techspec.md` (**N===1 only**, folded into `## Spec`). Use the envelope
 `systemPrompts.story`; split only under the policy above, and when you do,
-read `systemPrompts.storySplitRules` too — it carries the schedule and
-partition rules the core omits.
+read `systemPrompts.storySplitRules` too — it carries the schedule rules and
+the same-wave collision refusal the core omits. In **tickets mode** also read
+`systemPrompts.storyTicketsRules`: the source ticket is evidence, not a
+template — re-derive `acceptance[]` rather than carrying its list, handles
+and tier suffixes forward.
 
 **Tickets mode:** every Story authors a top-level `supersedes[]`; persist
 refuses a partial map ([shape](helpers/plan-reference.md)).
@@ -108,15 +111,20 @@ The maker-blind **pre-mortem** critic is not a step of this spine: run
 
 ### 3. Persist
 
-**Gate #2** — STOP for approval before persist **only** when the operator asked
-to review (`--force-review`). Under `--yes`, auto-proceed.
+**Gate #2** — STOP for approval before persist when the draft carries **more
+than one Story** (a split always earns operator eyes, `--force-review` or
+not), or when the operator asked to review (`--force-review`).
+Under `--yes`, auto-proceed.
 
 **Gate #3 — adopt, else create.** Offer the top `epicCandidates[]` Epic at
 **any N** (`--epic <id>`); else, at **N>2**, a new container (`--epic-title` /
 `--epic-goal`). Never unasked ([ref](helpers/plan-reference.md)).
 
 Run persist `--dry-run` **first** — same command, writes suppressed; every gate
-runs before the first `createIssue`, and the run **lists its warnings**
+runs before the first `createIssue` — including the **same-wave collision
+refusal**, which rejects an N>1 draft whose siblings declare a common path
+(merge them, or order them with `depends_on`) — and the run **lists its
+warnings**
 (a `creates` / `refactors-existing` the base branch disagrees with, a goal or
 acceptance path absent at base, an open question in a body) and the
 `changes[]` repairs it applied ([list](helpers/plan-reference.md)). Read
@@ -126,7 +134,6 @@ them; they never stop the persist:
 node .agents/scripts/plan-persist.js \
   --stories temp/plan-<slug>/stories.json \
   --plan-dir temp/plan-<slug> \
-  [--plan-acceptance temp/plan-<slug>/acceptance-manifest.json] \
   [--tech-spec temp/plan-<slug>/techspec.md] \
   [--source-tickets 123,456] \
   [--epic <id> | --epic-title "<name>" --epic-goal "<one paragraph>"]
@@ -154,6 +161,4 @@ also comments on and closes each source id ([ref](helpers/plan-reference.md)).
 ## See also
 
 [`/mandrel-deliver`](mandrel-deliver.md), [`/audit-to-stories`](audit-to-stories.md),
-[`helpers/plan-reference.md`](helpers/plan-reference.md) (on-demand detail),
-[`core/scope-triage`](../skills/core/scope-triage/SKILL.md) — optional
-split-advisory notes only (no routing verdict).
+[`helpers/plan-reference.md`](helpers/plan-reference.md) (on-demand detail).

package/README.md

@@ -86,9 +86,9 @@ time to confirm the install is healthy.
 >
 > Prefer a surgical alternative? Replace `shamefully-hoist` with a scoped
 > `public-hoist-pattern[]=` line per package listed in
-> `.agents/runtime-deps.json` (`ajv`, `ajv-formats`, `js-yaml`, `minimatch`,
-> `picomatch`, `typhonjs-escomplex`). If `mandrel doctor`
-> reports `runtime-deps missing: …`, this is the fix.
+> `.agents/runtime-deps.json` — read the file rather than copying a list from
+> here, since the complexity kernel's closure is several packages. If
+> `mandrel doctor` reports `runtime-deps missing: …`, this is the fix.
 
 `bootstrap.js` is interactive on a TTY and auto-accepts the
 owner/repo/base branch/operator handle it can infer from your local

package/docs/CHANGELOG.md

@@ -15,6 +15,27 @@ All notable changes to this project will be documented in this file.
 -->
 <!-- markdownlint-disable-file MD004 MD012 MD037 -->
 
+## [2.59.0](https://github.com/dsj1984/mandrel/compare/mandrel-v2.58.0...mandrel-v2.59.0) (2026-09-14)
+
+
+### Added
+
+* planning stops fragmenting cohesive work: the acceptance band goes, the dispatcher's own collision predicate becomes the split gate, and the audit seed stops pre-cutting the partition ([#5332](https://github.com/dsj1984/mandrel/issues/5332)) ([#5334](https://github.com/dsj1984/mandrel/issues/5334)) ([e6408e4](https://github.com/dsj1984/mandrel/commit/e6408e44d3f9a7fcbdf8c9a7df4a1b68fdffd88b))
+* replace the complexity kernel's parse and dispatch layers and declare the runtime closure it leaves behind, so the framework's dependency guards describe what actually loads ([#5336](https://github.com/dsj1984/mandrel/issues/5336)) ([#5337](https://github.com/dsj1984/mandrel/issues/5337)) ([821c4f5](https://github.com/dsj1984/mandrel/commit/821c4f55fe3fe9b91b83a4ef9fe9ddee5da9db8f))
+
+## [2.58.0](https://github.com/dsj1984/mandrel/compare/mandrel-v2.57.0...mandrel-v2.58.0) (2026-09-12)
+
+
+### Added
+
+* make the close test credit earnable on any test runner, and surface the gap at setup instead of mid-close ([#5324](https://github.com/dsj1984/mandrel/issues/5324)) ([#5329](https://github.com/dsj1984/mandrel/issues/5329)) ([04db145](https://github.com/dsj1984/mandrel/commit/04db145b8f1017121e129897f9207ab51db1c3f8))
+* tickets-mode planning re-derives the Story instead of carrying the source ticket's shape: acceptance handles, tier suffixes, generated-artifact footprints and pinned identifiers ([#5323](https://github.com/dsj1984/mandrel/issues/5323)) ([#5326](https://github.com/dsj1984/mandrel/issues/5326)) ([5563e7a](https://github.com/dsj1984/mandrel/commit/5563e7a264864f78001beb19ce58160a1b483cda))
+
+
+### Fixed
+
+* give base-sync and the Story-scope code review one base ref, so a stale local base branch cannot raise false blockers ([#5325](https://github.com/dsj1984/mandrel/issues/5325)) ([#5328](https://github.com/dsj1984/mandrel/issues/5328)) ([c4032b6](https://github.com/dsj1984/mandrel/commit/c4032b63a63c4cb8cc3f9c9956b6c646e5acd460))
+
 ## [2.57.0](https://github.com/dsj1984/mandrel/compare/mandrel-v2.56.0...mandrel-v2.57.0) (2026-09-12)
 
 

package/lib/cli/registry.js

@@ -4,8 +4,11 @@
  *
  * Exports an ordered array of check objects each shaped `{ name, run() }`.
  * `run()` returns `{ ok, detail, remedy? }` — `remedy` is present and
- * non-empty only when `ok` is false. The registry is the single source of
- * truth for which checks the doctor command runs and in what order.
+ * non-empty whenever `ok` is false, and on an `advisory` check that passes
+ * while still having something actionable to say (`test-credit-path`), which
+ * repeats it in `detail` because that is the field the doctor prints for a
+ * passing check. The registry is the single source of truth for which checks
+ * the doctor command runs and in what order.
  *
  * Checks run sequentially in the doctor runner (not in parallel) because
  * some checks are meaningless without a prerequisite having passed first
@@ -38,6 +41,8 @@ import {
 } from '../../.agents/scripts/lib/bootstrap/project-bootstrap.js';
 import { isCommandExcluded } from '../../.agents/scripts/lib/command-header.js';
 import { getDeliveryRouting } from '../../.agents/scripts/lib/config/delivery-routing.js';
+import { isResolvable } from '../../.agents/scripts/lib/runtime-deps/dep-resolution.js';
+import { describeParserMajorError } from '../../.agents/scripts/lib/runtime-deps/parser-major.js';
 import {
   defaultResolvePackageRoot,
   listFiles as listPayloadFiles,
@@ -516,6 +521,30 @@ function runAgentsInSync({
 // check: runtime-deps
 // ---------------------------------------------------------------------------
 
+/**
+ * The verdict when every required package is present.
+ *
+ * Presence is not the whole contract: `.agents/` resolves from the consumer's
+ * node_modules, so a declared range cannot enforce which `@babel/parser` major
+ * the complexity kernel actually gets, and the wrong major fails scoring
+ * mid-scan with an opaque plugin-list error. Reporting it here puts it where a
+ * consumer is already looking for what to fix.
+ *
+ * @param {() => string|null} parserMajorError
+ * @returns {{ ok: boolean, detail: string, remedy?: string }}
+ */
+function allPresentVerdict(parserMajorError) {
+  const parserProblem = parserMajorError();
+  if (parserProblem === null) {
+    return { ok: true, detail: 'all dependencies found' };
+  }
+  return {
+    ok: false,
+    detail: 'dependency version unsupported',
+    remedy: parserProblem,
+  };
+}
+
 /**
  * Verify that the framework's required runtime dependencies are resolvable
  * from the project's node_modules.
@@ -533,6 +562,8 @@ function runAgentsInSync({
  *   is missing.
  * - `manifestRequired` — array of required package names, skips the
  *   filesystem read of `runtime-deps.json`.
+ * - `parserMajorError()` — replaces the resolved-parser-major probe, so the
+ *   unsupported-major report is assertable without installing one.
  *
  * @param {{ projectRoot?: string, resolve?: (dep: string) => string, manifestRequired?: string[] }} [opts]
  * @returns {{ ok: boolean, detail: string, remedy?: string }}
@@ -541,6 +572,7 @@ function runRuntimeDeps({
   projectRoot,
   resolve: resolveSeam,
   manifestRequired,
+  parserMajorError = describeParserMajorError,
 } = {}) {
   // Anchor at process.cwd() (the consumer root), not resolveProjectRoot() (the
   // package root). Under pnpm isolated-mode the consumer's node_modules are not
@@ -566,33 +598,24 @@ function runRuntimeDeps({
 
   const missing = [];
 
-  if (resolveSeam) {
-    for (const dep of required) {
-      try {
-        resolveSeam(dep);
-      } catch {
-        missing.push(dep);
-      }
-    }
-  } else {
-    // Anchor resolution to the consumer project root so it mirrors the context
-    // in which the framework scripts run (they free-ride on the consumer's
-    // node_modules). Under pnpm isolated-mode the consumer's node_modules are
-    // not reachable from inside node_modules/mandrel/; anchoring at process.cwd()
-    // finds them correctly.
-    const req = createRequire(path.join(root, 'package.json'));
-    for (const dep of required) {
-      try {
-        req.resolve(dep);
-      } catch {
-        missing.push(dep);
-      }
-    }
+  // Anchor resolution to the consumer project root so it mirrors the context
+  // in which the framework scripts run (they free-ride on the consumer's
+  // node_modules). Under pnpm isolated-mode the consumer's node_modules are
+  // not reachable from inside node_modules/mandrel/; anchoring at process.cwd()
+  // finds them correctly.
+  //
+  // `isResolvable` is shared with the framework-side preflight rather than
+  // reimplemented here: it probes the bare specifier AND `<name>/package.json`,
+  // because a dependency with no `main` and no `exports` cannot be resolved by
+  // name at all. Probing only the bare name reports such a package missing
+  // while it sits installed, and this check gates `mandrel doctor`.
+  const resolve =
+    resolveSeam ?? createRequire(path.join(root, 'package.json')).resolve;
+  for (const dep of required) {
+    if (!isResolvable(dep, resolve)) missing.push(dep);
   }
 
-  if (missing.length === 0) {
-    return { ok: true, detail: 'all dependencies found' };
-  }
+  if (missing.length === 0) return allPresentVerdict(parserMajorError);
   return {
     ok: false,
     detail: `missing: ${missing.join(', ')}`,
@@ -1176,6 +1199,90 @@ function probeConfiguredDriver(command, runner, projectRoot) {
   };
 }
 
+// ---------------------------------------------------------------------------
+// check: test-credit-path
+// ---------------------------------------------------------------------------
+
+/**
+ * The runner whose own green full run deposits the close `test` credit as a
+ * side effect (`.agents/scripts/run-tests.js` → `lib/test-run-credit.js`).
+ */
+const MANDREL_TEST_RUNNER = 'run-tests.js';
+
+/**
+ * The deposit that works whatever `npm test` resolves to: it spawns the
+ * project's own suite and stamps the result, so it is honest by construction.
+ */
+const TEST_CREDIT_DEPOSIT_COMMAND =
+  'node .agents/scripts/evidence-gate.js --standalone --scope-id <storyId> --gate test --worktree <workCwd> -- npm test';
+
+/** Remedy shared by every shape that does not earn the credit on its own. */
+const TEST_CREDIT_REMEDY = `run the suite through the depositor instead of bare \`npm test\`: ${TEST_CREDIT_DEPOSIT_COMMAND}`;
+
+/**
+ * The project's `test` script — what `npm test` runs, and therefore what the
+ * close `test` gate spawns. Deliberately read from `package.json` rather than
+ * `project.commands.test`: the close gate's argv is the literal `npm test`, so
+ * the npm script is the command whose shape decides whether a bare run
+ * deposits anything.
+ *
+ * @param {string} projectRoot
+ * @param {typeof fs.readFileSync} readFileImpl
+ * @returns {string|null} The trimmed script, or null when there is none.
+ */
+function readProjectTestScript(projectRoot, readFileImpl) {
+  try {
+    const raw = readFileImpl(path.join(projectRoot, 'package.json'), 'utf8');
+    const script = JSON.parse(raw)?.scripts?.test;
+    return typeof script === 'string' && script.trim().length > 0
+      ? script.trim()
+      : null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Report whether this project's test command reaches mandrel's own runner,
+ * and therefore whether a bare `npm test` deposits the close `test` credit
+ * by itself (Story #5324).
+ *
+ * **Always `ok: true`.** A project on `vitest`, `jest` or any other runner is
+ * a supported setup, not a broken install — the only thing it lacks is the
+ * runner-side bonus deposit, and the remedy command covers it. So the check
+ * informs rather than gates, and (unlike every other entry here) carries its
+ * `remedy` alongside a passing verdict; the same guidance also rides in
+ * `detail`, which is the field the doctor prints for a passing check.
+ *
+ * The runner is detected by name, never by executing anything: a script that
+ * reaches `run-tests.js` indirectly reads as "own runner", which errs toward
+ * naming the deposit command — correct on either shape.
+ *
+ * @param {{ projectRoot?: string, readFile?: typeof fs.readFileSync }} [opts]
+ * @returns {{ ok: boolean, detail: string, remedy?: string }}
+ */
+function runTestCreditPath({ projectRoot, readFile = fs.readFileSync } = {}) {
+  const script = readProjectTestScript(projectRoot ?? process.cwd(), readFile);
+  if (script === null) {
+    return {
+      ok: true,
+      detail: `no \`test\` script in package.json — close still spawns \`npm test\`, so ${TEST_CREDIT_REMEDY}`,
+      remedy: TEST_CREDIT_REMEDY,
+    };
+  }
+  if (script.includes(MANDREL_TEST_RUNNER)) {
+    return {
+      ok: true,
+      detail: `\`npm test\` → \`${script}\` reaches mandrel's runner — a green full run on a story branch deposits the close \`test\` credit itself`,
+    };
+  }
+  return {
+    ok: true,
+    detail: `\`npm test\` → \`${script}\` is this project's own runner and never reaches \`${MANDREL_TEST_RUNNER}\`, so it deposits no close \`test\` credit and prints nothing — ${TEST_CREDIT_REMEDY}`,
+    remedy: TEST_CREDIT_REMEDY,
+  };
+}
+
 /**
  * Ordered array of doctor checks. Each entry follows the
  * `{ name: string, run(opts?): { ok: boolean, detail: string, remedy?: string } }` contract.
@@ -1243,6 +1350,15 @@ export const registry = [
     advisory: true,
     run: (opts) => runVersionCurrent(opts),
   },
+  {
+    name: 'test-credit-path',
+    // Non-fatal: `run()` always returns ok:true. A project on its own test
+    // runner is a supported setup that simply has to deposit the close
+    // `test` credit explicitly, so this reports a condition rather than
+    // failing the install (Story #5324).
+    advisory: true,
+    run: (opts) => runTestCreditPath(opts),
+  },
 ];
 
 export default registry;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mandrel",
-  "version": "2.57.0",
+  "version": "2.59.0",
   "description": "Claude Code-first opinionated workflow framework: instructions, skills, rules, and SDLC workflows that govern AI coding assistants.",
   "files": [
     ".agents/",
@@ -109,12 +109,17 @@
     "typescript": "^6.0.3"
   },
   "dependencies": {
+    "@babel/parser": "^7.29.3",
     "ajv": "^8.20.0",
     "ajv-formats": "^3.0.1",
+    "babel-runtime": "^6.26.0",
+    "escomplex-plugin-metrics-module": "^0.1.0",
+    "escomplex-plugin-syntax-babylon": "^0.1.0",
     "js-yaml": "^4.3.2",
     "minimatch": "^10.0.0",
     "picomatch": "^4.0.4",
-    "typhonjs-escomplex": "^0.1.0"
+    "typhonjs-ast-walker": "^0.2.1",
+    "typhonjs-escomplex-commons": "^0.1.1"
   },
   "peerDependencies": {
     "@cucumber/gherkin": ">=32.0.0",

package/.agents/scripts/lib/orchestration/split-policy-validator.js

@@ -1,188 +0,0 @@
-/**
- * v2 split-policy validator — the plan-time "one-owner-AC" split rejector.
- *
- * Under the v2 default-single split policy (`docs/roadmap.md` § v2.0.0), a
- * plan authors **one Story by default**; it splits into N>1 Stories only when
- * the pieces have near-zero overlap or sit across an architectural seam. This
- * validator is the deterministic guardrail on that policy: **every acceptance
- * criterion must belong to exactly one Story.** An identical AC appearing in
- * two Stories is evidence the split coupled what should have stayed one Story,
- * so the plan is refused rather than reconciled at delivery time (there is no
- * epic-level acceptance reconcile in v2).
- *
- * Scope of the deterministic check:
- *   - **Cross-Story duplication** (always): the same normalized AC text must
- *     not appear in more than one Story.
- *   - **Full coverage** (optional, when a plan-level `acceptance` manifest is
- *     supplied): every manifest AC is claimed by exactly one Story, and no
- *     Story claims an AC absent from the manifest.
- *
- * Semantic overlap between differently-worded ACs is **not** caught here — it
- * is a gate-#2 review call. This validator only sees identical (normalized)
- * text, keeping it deterministic and false-positive-free.
- *
- * Normalization for comparison: trim, collapse internal whitespace, and
- * lower-case. Reporting always uses the first-seen original text.
- */
-
-/**
- * Normalize an acceptance string for equality comparison. Trims, collapses
- * runs of whitespace to a single space, and lower-cases. Non-strings and
- * empty/whitespace-only strings normalize to `null` (ignored).
- *
- * @param {unknown} ac
- * @returns {string | null}
- */
-export function normalizeAcceptance(ac) {
-  if (typeof ac !== 'string') return null;
-  const norm = ac.trim().replace(/\s+/g, ' ').toLowerCase();
-  return norm === '' ? null : norm;
-}
-
-/**
- * @typedef {object} StorySlice
- * @property {string} [id]        Story identifier for reporting (slug or #id).
- * @property {string} [slug]      Alternate identifier (used when `id` absent).
- * @property {string[]} acceptance Acceptance criteria this Story claims.
- */
-
-/**
- * @typedef {object} SplitPolicyViolation
- * @property {'cross-story-duplicate'|'orphan-ac'|'unclaimed-manifest-ac'} kind
- * @property {string} acceptance The original (first-seen) AC text.
- * @property {string[]} [stories] Story ids sharing a duplicated AC (`cross-story-duplicate`).
- * @property {string}   [story]   Story id owning an AC absent from the manifest (`orphan-ac`).
- */
-
-/**
- * Resolve a Story's reporting id.
- *
- * @param {StorySlice} story
- * @param {number} index
- * @returns {string}
- */
-function storyId(story, index) {
-  if (story && typeof story.id === 'string' && story.id.trim() !== '') {
-    return story.id;
-  }
-  if (story && typeof story.slug === 'string' && story.slug.trim() !== '') {
-    return story.slug;
-  }
-  return `story[${index}]`;
-}
-
-/**
- * Validate that acceptance criteria partition cleanly across Stories.
- *
- * @param {StorySlice[]} stories The plan's Stories, each with `acceptance[]`.
- * @param {object} [opts]
- * @param {string[]} [opts.planAcceptance] Optional plan-level acceptance
- *   manifest. When supplied, coverage is enforced (every manifest AC claimed
- *   exactly once; no Story claims an off-manifest AC).
- * @returns {{ ok: boolean, violations: SplitPolicyViolation[] }}
- */
-export function validateAcceptancePartition(stories, opts = {}) {
-  const violations = [];
-  const list = Array.isArray(stories) ? stories : [];
-
-  // normalized AC → { original, owners: Set<storyId> }
-  const owners = new Map();
-  list.forEach((story, index) => {
-    const id = storyId(story, index);
-    const acceptance = Array.isArray(story?.acceptance) ? story.acceptance : [];
-    for (const ac of acceptance) {
-      const norm = normalizeAcceptance(ac);
-      if (norm === null) continue;
-      const existing = owners.get(norm);
-      if (existing) {
-        existing.owners.add(id);
-      } else {
-        owners.set(norm, { original: ac.trim(), owners: new Set([id]) });
-      }
-    }
-  });
-
-  // Cross-Story duplication: any AC owned by more than one Story.
-  for (const { original, owners: set } of owners.values()) {
-    if (set.size > 1) {
-      violations.push({
-        kind: 'cross-story-duplicate',
-        acceptance: original,
-        stories: [...set],
-      });
-    }
-  }
-
-  // Optional coverage check against a plan-level manifest.
-  const manifest = Array.isArray(opts.planAcceptance)
-    ? opts.planAcceptance
-    : null;
-  if (manifest !== null) {
-    const manifestNorms = new Map();
-    for (const ac of manifest) {
-      const norm = normalizeAcceptance(ac);
-      if (norm !== null && !manifestNorms.has(norm)) {
-        manifestNorms.set(norm, ac.trim());
-      }
-    }
-    // Every manifest AC must be claimed by exactly one Story.
-    for (const [norm, original] of manifestNorms) {
-      if (!owners.has(norm)) {
-        violations.push({
-          kind: 'unclaimed-manifest-ac',
-          acceptance: original,
-        });
-      }
-    }
-    // No Story may claim an AC absent from the manifest.
-    for (const [norm, { original, owners: set }] of owners) {
-      if (!manifestNorms.has(norm)) {
-        violations.push({
-          kind: 'orphan-ac',
-          acceptance: original,
-          story: [...set][0],
-        });
-      }
-    }
-  }
-
-  return { ok: violations.length === 0, violations };
-}
-
-/**
- * Render a single violation as a human-readable line.
- *
- * @param {SplitPolicyViolation} v
- * @returns {string}
- */
-function formatViolation(v) {
-  switch (v.kind) {
-    case 'cross-story-duplicate':
-      return `acceptance criterion appears in ${v.stories.length} Stories (${v.stories.join(', ')}) — a coupled split; keep it one Story: "${v.acceptance}"`;
-    case 'unclaimed-manifest-ac':
-      return `plan acceptance criterion is claimed by no Story: "${v.acceptance}"`;
-    case 'orphan-ac':
-      return `Story ${v.story} claims an acceptance criterion absent from the plan manifest: "${v.acceptance}"`;
-    default:
-      return `unknown split-policy violation: "${v.acceptance}"`;
-  }
-}
-
-/**
- * Throwing wrapper for the persist path: throws a single batched error when
- * the acceptance criteria do not partition cleanly, otherwise returns
- * `stories` unchanged. Wired into `plan-persist` in Stage 3.
- *
- * @param {StorySlice[]} stories
- * @param {object} [opts] See {@link validateAcceptancePartition}.
- * @returns {StorySlice[]}
- */
-export function assertAcceptancePartition(stories, opts = {}) {
-  const { ok, violations } = validateAcceptancePartition(stories, opts);
-  if (ok) return stories;
-  throw new Error(
-    `[split-policy] ${violations.length} acceptance-partition violation(s) — the plan splits coupled work; refuse:\n${violations
-      .map((v) => `  - ${formatViolation(v)}`)
-      .join('\n')}`,
-  );
-}

package/.agents/scripts/lib/templates/spec-author-prompts.js

@@ -1,76 +0,0 @@
-/**
- * spec-author-prompts.js — Story-scoped Spec / Acceptance authoring prompts.
- *
- * v2 keeps a single executable document per Story. These prompts used to
- * author a separate Epic Tech Spec + Acceptance Spec that were folded into
- * the Epic body and then restated on Stories — a duplication source. They
- * now author only Story `## Spec` approach prose and remind authors that
- * acceptance lives once on the Story (`acceptance[]` / `## Acceptance`).
- */
-
-/**
- * @returns {string}
- */
-export function renderTechSpecSystemPrompt() {
-  return TECH_SPEC_SYSTEM_PROMPT;
-}
-
-/**
- * @returns {string}
- */
-export function renderAcceptanceSpecSystemPrompt() {
-  return ACCEPTANCE_SPEC_SYSTEM_PROMPT;
-}
-
-const TECH_SPEC_SYSTEM_PROMPT = `You are an expert Engineering Architect.
-Your job is to author the Story's \`## Spec\` approach section — the technical
-how for one cohesive, executable Story. There is no separate Epic Tech Spec
-document and no spill-to-docs path.
-
-The Spec should outline only what an implementing agent needs beyond Goal /
-Changes / Acceptance:
-1. Architecture & Design (approach, seams, reuse)
-2. Data Models (if any)
-3. API Changes (if any)
-4. Core Components
-5. Security & Privacy Considerations
-
-CRITICAL REQUIREMENTS:
-- Respond ONLY with valid Markdown suitable to paste into a Story \`## Spec\`.
-- Do not use top-level <h1> (# ) tags. Prefer \`##\` / \`###\` under Spec.
-- Do NOT restate the Story's Goal, Acceptance, Verify, Changes, or Non-Goals —
-  those sections already live on the same Story body. Restatement is
-  duplication and a drift risk. If a brief orientation helps, keep a
-  \`## Technical Overview\` to 2–3 sentences naming the technical approach only.
-- Do NOT author a Delivery Slicing / fan-out table. If the work needs multiple
-  independent Stories, say so in one short note and stop — oversized Specs
-  mean the Story should be split, not documented elsewhere.
-- Format architectural decisions clearly with bullet points.
-- Keep the Spec lean enough to stay inline on the Story (persist rejects
-  over-budget Specs; they are never written under docs/).`;
-
-const ACCEPTANCE_SPEC_SYSTEM_PROMPT = `You are an expert Acceptance Engineer.
-Your job is to help author the Story's binding acceptance contract — not a
-separate Acceptance Spec document.
-
-v2 rule: acceptance lives **once**, on the Story:
-- Machine contract: top-level \`acceptance[]\` (and \`verify[]\`) on the ticket JSON
-- Human document: the same items rendered under \`## Acceptance\` / \`## Verify\`
-  on the Story body (persist syncs top-level into the body)
-
-Do **not** author an Epic Acceptance Table, PRD restatement, or a second
-criteria list inside \`## Spec\`.
-
-CRITICAL REQUIREMENTS:
-- Respond ONLY with guidance or a draft \`acceptance[]\` / \`verify[]\` list for
-  one Story — never a parallel "Acceptance Spec" markdown artifact.
-- Every acceptance item MUST be observable from outside the agent (command
-  exits 0, file exists, selector resolves, fixture count matches). Reject
-  vague "matches the spec" / "looks good" items.
-- Every verify entry MUST name a tier in parentheses: unit | contract | e2e |
-  validate (or \`manual:<reason>\` when genuinely unverifiable in isolation).
-- Do NOT re-elaborate Goal or Spec prose inside acceptance items — bind the
-  outcome, not the approach.
-- Acceptance Outcomes MUST NOT prescribe a commit subject that begins with a
-  non-Conventional-Commits prefix (allowed leading types: feat|fix|chore|
-  refactor|perf|docs|style|test|build|ci|revert).`;