baldart 4.25.0 → 4.26.1

package/CHANGELOG.md

@@ -5,6 +5,31 @@ All notable changes to BALDART will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [4.26.1] - 2026-06-11
+
+**`new2`: specialization integrity — a full audit of every agent spawn; no role mixing anywhere.** User principle: code is written ONLY by `coder`, UI only by `ui-expert`, each agent does one thing. The audit of all 16 spawn sites found two genuine violations and three under-specified plumbing roles. **PATCH** (role-integrity fixes on the EXPERIMENTAL `new2` surface; no config key, no change to `/new`).
+
+### Fixed
+
+- **`new2.js` E2 — the ops pre-flight agent no longer repairs the baseline.** "baseline FAILS → fix once" had a general-purpose git agent editing source code. Now: the pre-flight returns `baseline:'fail'` + an actionable log (explicit role boundary: it never edits source/doc files), and the WORKFLOW spawns the **coder** specialist for one bounded repair attempt — verified by deterministically re-running the baseline gates (no claim to trust), `E2-baseline: FIXED-BY-CODER` ledger row; still failing → batch-fatal as before. Zero extra spawns in the happy path.
+- **`new2.js` — the Codex driver never reviews.** Its runtime fallback was "perform the review yourself with the code-reviewer lens" — a general-purpose agent doing code review. Now the driver returns `note:'codex-unavailable'` with empty findings, and the workflow spawns the REAL `code-reviewer` (same `stdReview` call as the matrix), ledgered as `review-codex: FALLBACK`.
+- **`new2-resolve.js` — judge map completed per domain.** `doc` fixes are now judged by **doc-reviewer** (code-reviewer judging prose was cross-domain) and `test` fixes by **qa-sentinel**; `security`/`migration` → security-reviewer and `perf` → api-perf-cost-auditor unchanged; `ui`/`code` stay with code-reviewer (the judge verifies a CODE change — its charter, incl. DS rule 8 for UI). Fixer and judge of the same type remain two independent adversarial instances.
+
+### Changed
+
+- **`new2.js` — explicit ROLE BOUNDARY lines on every plumbing agent** (pre-flight, commit, merge, production-readiness): they never edit source/doc files; commit/merge touch ONLY card YAML status fields + registry rows; a content-level merge conflict is leave+report, never hand-resolved; production-readiness executes stack-matched commands but reports (never edits) code/config changes. The full writer map is now: code/perf/migration/test fixes → `coder`; UI → `ui-expert`; security fixes → `security-reviewer`; docs → `doc-reviewer`; backlog YAML → `prd-card-writer`; bookkeeping (status/registry) → mechanical commit/merge agents.
+
+## [4.26.0] - 2026-06-11
+
+**`new2`: Phase 1 decomposed into specialist agents — the owner implements, it no longer explores.** The old per-card pipeline gave the owner agent one mega-prompt ("you ARE claim + architect + plan-auditor + owner"), so the owner absorbed the whole codebase exploration into its own context and reached the actual coding with a degraded window. Phase 1 now runs as dedicated specialists with file handoff (`/tmp`), per the "ognuno fa una cosa" principle. Verified premise: nested subagent spawning does NOT exist in Claude Code (official docs: "Subagents cannot spawn other subagents" + 2 empirical probes), so the decomposition lives at the WORKFLOW level — the JS is the orchestrator, exactly what dynamic workflows are for. **MINOR** (pipeline capability on the EXPERIMENTAL `new2` surface only; no config key, no change to `/new`).
+
+### Changed
+
+- **`framework/.claude/workflows/new2.js` — per-card `codebase-architect` specialist (B7).** Context retrieval is now a dedicated `codebase-architect` spawn that writes the COMPLETE untruncated baseline to `/tmp/arch-baseline-<CARD>.md` (the owner and reviewers Read the file; the structured return stays minimal). Bonus over the old pre-flight snapshot: the per-card architect sees prior in-batch commits — card N's baseline reflects what cards 1..N-1 changed. The pre-flight no longer writes baselines (generalist relieved of specialist work; `archBaselinePaths` removed from its contract). **Fail-safe**: architect crash/not-ok degrades to the old inline behavior (the owner explores itself) — never blocks the card; transient outage re-queues it.
+- **`new2.js` — `plan-auditor` demoted to a deterministic DRIFT gate.** `/prd` already validates every card at creation time, so an unconditional re-audit per card was duplicate work. plan-auditor now runs ONLY when execution-time drift is plausible: (a) a prior in-batch commit touched the card's declared surface (`filesLikelyTouched` ∩ prior `filesChanged`, computed in JS), (b) the architect found declared paths missing (stale card — factual `missingPaths` check, no judgement), or (c) `review_profile: deep` (Rule C escalation). Its corrections amend the implementation BRIEFING (never the backlog YAML). A fresh single-card batch with no drift evidence skips it at zero cost. Crash → non-blocking skip, ledgered.
+- **`new2.js` — epic guard moved to JS (zero spawns for trackers).** The pre-flight returns `cardGraph[].isEpic` (implement.md §6b rule); an epic card now short-circuits in JS before ANY spawn — the old path burned a full owner-agent spawn just to learn the card was a tracker. The impl agent's own epic flag stays as backstop.
+- **`new2.js` telemetry — `per_card[].phase1`** records `{architect: done|inline-fallback, audit: pass|fixes|skipped-no-drift|skipped-error|skipped-no-baseline}` plus `phase1-architect`/`phase1-audit` ledger rows, so the A/B accounting can price the decomposition.
+
 ## [4.25.0] - 2026-06-11
 
 **`new2`: deterministic per-card review matrix — `balanced` no longer spawns every specialist on every card.** The old fan-out ran code-reviewer + doc-reviewer + qa-sentinel + api-perf-cost-auditor (+ security-reviewer) unconditionally at `balanced`: an api-perf pass on a doc-only card, a doc pass on a pure-code card — 1–3 wasted spawns per card. Each specialist now runs IFF its domain is evidenced by the card's actual surface (`scopeFiles ∪ MAY-EDIT`), computed deterministically in JS (no agent judgement) and audited via a `review-matrix` ledger row per card. **MINOR** (review-behavior capability on the EXPERIMENTAL `new2` surface only; no config key, no change to `/new` — its interactive profiles are untouched; the schema-change propagation rule does not apply).

package/VERSION

@@ -1 +1 @@
-4.25.0
+4.26.1

package/framework/.claude/workflows/new2-resolve.js

@@ -99,7 +99,17 @@ const FOLLOWUP_SCHEMA = {
 // F-024 — domain-specialized fixer + judge (full map; reviewer-owns-its-domain — a doc
 // finding is fixed by doc-reviewer, a security finding by security-reviewer, never coder).
 const fixerAgent = ({ doc: 'doc-reviewer', ui: 'ui-expert', security: 'security-reviewer' })[domain] || 'coder'
-const judgeAgent = (domain === 'security' || domain === 'migration') ? 'security-reviewer' : domain === 'perf' ? 'api-perf-cost-auditor' : 'code-reviewer'
+// Specialization integrity (v4.26.1) — the judge is the VERIFICATION specialist of the
+// finding's domain: doc fixes judged by doc-reviewer (code-reviewer judging prose was
+// cross-domain), test fixes by qa-sentinel (THE test specialist). `ui` and `code` stay with
+// code-reviewer: the judge verifies a CODE change, which is code-reviewer's charter
+// (including DS-coherence rule 8 for UI). Fixer and judge of the same type are still two
+// independent instances — the judge prompt is adversarial and greps the files itself.
+const judgeAgent = (domain === 'security' || domain === 'migration') ? 'security-reviewer'
+  : domain === 'perf' ? 'api-perf-cost-auditor'
+  : domain === 'doc' ? 'doc-reviewer'
+  : domain === 'test' ? 'qa-sentinel'
+  : 'code-reviewer'
 
 const findingsBlock = findings.map((f, i) => `  ${i + 1}. [${f.kind || kind}/${f.domain || domain}] ${f.evidence}`).join('\n')
 const brief = [

package/framework/.claude/workflows/new2.js

@@ -148,6 +148,10 @@ const PREFLIGHT_SCHEMA = {
           // spawn per card in runCard (always false on a fresh run).
           alreadyCommitted: { type: 'boolean', description: 'commit referencing the card exists in trunk..HEAD of the worktree AND validation re-runs green AND no open follow-up' },
           alreadyCommittedSha: { type: 'string' },
+          // B7 (v4.26.0) — epic guard moved to JS (an epic never reaches any spawn) + the
+          // card's declared surface, used by the deterministic plan-audit drift gate.
+          isEpic: { type: 'boolean', description: "epic/tracker card per implement.md §6b: id ends '-00' OR filename ends '-epic.yml' OR group.is_epic:true OR review_profile 'skip' with no requirements" },
+          filesLikelyTouched: { type: 'array', items: { type: 'string' }, description: "the card's files_likely_touched, verbatim" },
           // F-016 — ACs whose only implementation file is outside the card MAY-EDIT,
           // pre-classified deferred-by-policy (never routed to resolve).
           policyDeferredACs: { type: 'array', items: { type: 'object', additionalProperties: true } },
@@ -156,6 +160,9 @@ const PREFLIGHT_SCHEMA = {
     },
     excluded: { type: 'array', items: { type: 'object', additionalProperties: true } },
     ownershipMapPath: { type: 'string' },
+    // B7 — archBaselinePaths removed: the per-card baseline is now written by the
+    // codebase-architect SPECIALIST in runCard (it also sees prior in-batch commits,
+    // which a pre-flight snapshot cannot), not by this generalist agent.
     crossCard: { type: 'string' },
     // G2 — deterministic Codex availability (glob-resolved, NOT a self-reported judgement),
     // so a false negative on the cross-card check is visible in telemetry. true=companion found.
@@ -164,7 +171,6 @@ const PREFLIGHT_SCHEMA = {
     // `light` card can run Codex as its finder. Empty string when codexResolved is false.
     codexScriptPath: { type: 'string' },
     workspaceNote: { type: 'string' },
-    archBaselinePaths: { type: 'array', items: { type: 'string' } },
   },
 }
 
@@ -213,18 +219,19 @@ try {
     `You are the deterministic PRE-FLIGHT for an autonomous /new batch (variant new2). Follow ${REF}/setup.md (Phase 0 + Pre-flight) and ${REF}/implement.md (Phase 1 depends-on gate) for the SEMANTICS, but replace EVERY AskUserQuestion with the deterministic policy below. You run all git/bash yourself (the workflow cannot).\n\n` +
       `${projectBrief}\n\nCards in batch (Read each YAML):\n${cardPaths.join('\n')}\nCard IDs: ${cardIds.join(' ')}\n\n` +
       `Create/maintain the recovery tracker at /tmp/batch-tracker-${firstCard}.md (per setup.md § Context Tracking).\n\n` +
+      `ROLE BOUNDARY (specialization integrity): you are the OPS/GIT agent. You NEVER edit source or doc files — any needed content change belongs to the coder specialist; report it instead.\n\n` +
       `DETERMINISTIC GATE POLICIES (NO user prompts):\n` +
       `• G1 dirty-tree (main repo ${MAIN}): partition framework-managed noise exactly as setup.md step 3 ($METRICS=${METRICS}, .baldart/generated|state.json|skill-conflicts.json — NOT overlays/). Genuine user work → auto-stash 'baldart-new2-${firstCard}' (main checkout) and record the label. Never commit/abort/prompt.\n` +
-      `• Worktree (setup.md step 4): create ONE code worktree off ${TRUNK}; install deps; assign a port; run the baseline (tsc+lint+build). Copy ONLY the artifacts needed (env/.env.local/.env.example/supabase/.temp) — do NOT bulk-copy untracked files from the main repo (avoids stray backlog cards in the worktree). Use the git-authoritative idempotency pre-check. E2: baseline FAILS → fix once; still failing → baseline:'fail' + baselineLog (batch-fatal).\n` +
+      `• Worktree (setup.md step 4): create ONE code worktree off ${TRUNK}; install deps; assign a port; run the baseline (tsc+lint+build). Copy ONLY the artifacts needed (env/.env.local/.env.example/supabase/.temp) — do NOT bulk-copy untracked files from the main repo (avoids stray backlog cards in the worktree). Use the git-authoritative idempotency pre-check. E2: baseline FAILS → do NOT fix it yourself (role boundary — the coder specialist repairs it); return baseline:'fail' + a baselineLog precise enough for a coder to act (failing command, error excerpt, suspect files).\n` +
       codexResolveBullet +
       g3Bullet +
       `• G4 card-field validation (setup.md 1b/1c): card missing requirements/acceptance_criteria/files_likely_touched → EXCLUDE (excluded[] + reason). Never HALT for one bad card.\n` +
       `• G5 depends-on: a card whose depends_on names a non-DONE card NOT in this batch → EXCLUDE it AND every in-batch card that transitively depends on it.\n` +
-      `• cardGraph (REQUIRED, F-021): for every runnable card return { id, dependsOn:[IN-BATCH deps only], ownerAgent (the card's owner_agent; G25 unknown→'coder'), reviewProfile (the card's review_profile; default 'balanced'), policyDeferredACs, alreadyCommitted, alreadyCommittedSha }.\n` +
+      `• cardGraph (REQUIRED, F-021): for every runnable card return { id, dependsOn:[IN-BATCH deps only], ownerAgent (the card's owner_agent; G25 unknown→'coder'), reviewProfile (the card's review_profile; default 'balanced'), policyDeferredACs, alreadyCommitted, alreadyCommittedSha, isEpic (implement.md §6b epic guard: id ends '-00' OR filename ends '-epic.yml' OR group.is_epic:true OR review_profile 'skip' with no requirements), filesLikelyTouched (verbatim from the YAML) }.\n` +
       `• B1/F-026 idempotency (per card, AFTER the worktree exists): set alreadyCommitted:true (+ alreadyCommittedSha) IFF ALL hold: (a) a commit referencing the card id exists in ${TRUNK}..HEAD of the worktree; (b) the card's validation_commands re-run GREEN right now; (c) NO open follow-up card for it exists in ${paths.backlog_dir || 'backlog'}. On a FRESH worktree ${TRUNK}..HEAD is empty → all false, zero extra work.\n` +
       `• F-016 AC↔ownership consistency: for each acceptance_criterion, derive the file(s) it requires editing. If those files are NOT a subset of the card's MAY-EDIT/files_likely_touched → add the AC to policyDeferredACs:[{n,text,owningCard|owningFile,reason}] (it will become ONE follow-up, never a resolve). Do the same for any AC whose remedy is an owner-gated infra action (remote db push / deploy / secret / DNS).\n` +
       `• Ownership (setup.md 3c): build the file-ownership map → /tmp; return ownershipMapPath. F-040: each card's MAY-EDIT = files_likely_touched ∪ every path NAMED EXPLICITLY in that card's acceptance_criteria/definition_of_done (an ADR the DoD says to update, the data-model / ER doc for a schema-change, etc.) — so editing a DoD-mandated doc is NOT a file-diff violation. Do NOT add another card's files this way.\n` +
-      `• Persist per-card architecture baselines to /tmp/arch-baseline-<CARD>.md; return archBaselinePaths.\n\n` +
+      `• Do NOT write architecture baselines — the per-card codebase-architect specialist does that during the card pipeline (B7).\n\n` +
       `Return the structured PREFLIGHT object. ok:false ONLY if the workspace is unworkable.`,
     { label: 'preflight', phase: 'Pre-flight', agentType: 'general-purpose', schema: PREFLIGHT_SCHEMA }
   )
@@ -233,9 +240,28 @@ try {
   return finalReturn({ fatal: true, reason: 'pre-flight failed: ' + String(e && e.message) })
 }
 
-if (!preflight || preflight.ok === false || preflight.baseline === 'fail') {
-  ledger(firstCard, 'E2-baseline', 'BATCH-FATAL', (preflight && preflight.baselineLog) || 'workspace unworkable')
-  return finalReturn({ fatal: true, reason: 'baseline build irrecoverable — see baselineLog' })
+if (!preflight || preflight.ok === false) {
+  ledger(firstCard, 'preflight', 'BATCH-FATAL', (preflight && preflight.workspaceNote) || 'workspace unworkable')
+  return finalReturn({ fatal: true, reason: 'workspace unworkable — see pre-flight' })
+}
+if (preflight.baseline === 'fail') {
+  // E2 (specialization integrity) — baseline repair is CODE work: it belongs to the coder
+  // specialist, not the ops pre-flight agent (which never edits source). ONE bounded attempt;
+  // the verification is the deterministic re-run of the baseline gates themselves (no claim
+  // to trust — and every card's G26 re-exercises them anyway). Still failing → batch-fatal.
+  let repair = null
+  try {
+    repair = await agentSafe(
+      `You are the coder. The batch worktree BASELINE is failing on trunk-derived code (this is NOT card work — no card has run yet). Worktree: ${preflight.worktreePath} (cd into it).\n\nFailure log:\n${preflight.baselineLog || '(missing — re-run tsc/lint/build to reproduce)'}\n\nApply the minimal correct fix so the baseline gates (tsc + lint + build) pass, RE-RUN them, and report honestly (fixed:true ONLY if they now pass). Return: { fixed, log }`,
+      { label: 'baseline-repair', phase: 'Pre-flight', agentType: 'coder',
+        schema: { type: 'object', required: ['fixed'], additionalProperties: true, properties: { fixed: { type: 'boolean' }, log: { type: 'string' } } } }
+    )
+  } catch (e) { if (e && e.transientExhausted) noteDegraded('outage'); repair = null }
+  if (repair && repair.fixed) ledger(firstCard, 'E2-baseline', 'FIXED-BY-CODER', String(repair.log || '').slice(0, 200))
+  else {
+    ledger(firstCard, 'E2-baseline', 'BATCH-FATAL', preflight.baselineLog || 'baseline irrecoverable')
+    return finalReturn({ fatal: true, reason: 'baseline build irrecoverable — see baselineLog' })
+  }
 }
 
 for (const ex of preflight.excluded || []) ledger(ex.card, 'preflight-exclude', 'EXCLUDED', ex.reason)
@@ -267,7 +293,6 @@ const sharedCtx = {
   worktreePath: preflight.worktreePath,
   branch: preflight.branch,
   ownershipMapPath: preflight.ownershipMapPath,
-  archBaselinePaths: preflight.archBaselinePaths || [],
   // v4.18.0 — per-card Codex-light finder needs the resolved companion path in runCard scope.
   codexResolved: !!preflight.codexResolved,
   codexScriptPath: preflight.codexScriptPath || '',
@@ -390,6 +415,11 @@ async function runCard(cardId, cardPath) {
   const deferredClasses = new Set(deferredOpen ? ['policy-deferred-ac'] : [])
   function g(name, decision, detail) { gates.push({ gate: name, decision, detail: detail || '' }); ledger(cardId, name, decision, detail) }
 
+  // B7 — epic guard in JS (pre-flight reads the YAML): an epic is a tracker, not implementable
+  // work — skip it BEFORE any spawn (the old path burned a full owner-agent spawn to learn it).
+  // The impl agent's own epic flag stays as backstop for a pre-flight false negative.
+  if (node.isEpic) { g('router', 'EPIC-SKIPPED', 'epic card (pre-flight, zero spawns)'); return { card: cardId, status: 'epic-skipped', gates, commit: '-' } }
+
   // F-026/B1 — skip-completed from the PRE-FLIGHT's git-authoritative probe (cardGraph[].
   // alreadyCommitted), not a per-card agent spawn: on a fresh run the old Haiku probe was N
   // guaranteed-false spawns, and on resume the journal cache already covers it. Keyed on the
@@ -399,14 +429,70 @@ async function runCard(cardId, cardPath) {
     return { card: cardId, status: 'committed', commit: node.alreadyCommittedSha || '-', filesChanged: [], scopeFiles: [], archBaselinePath: `/tmp/arch-baseline-${cardId}.md`, gates }
   }
 
-  const cardBrief = `${projectBrief}\n\nCard: ${cardId}\nCard YAML: ${cardPath}\nOwner agent: ${ownerAgent} · Review profile: ${reviewProfile}\nWorktree: ${sharedCtx.worktreePath} (cd into it)\nFile-ownership map: ${sharedCtx.ownershipMapPath}\nArch baseline (write to /tmp/arch-baseline-${cardId}.md): reuse if present.\nNOTE: ACs already pre-classified as policy-deferred MUST NOT be implemented or routed — they are tracked as follow-ups.`
+  const cardBrief = `${projectBrief}\n\nCard: ${cardId}\nCard YAML: ${cardPath}\nOwner agent: ${ownerAgent} · Review profile: ${reviewProfile}\nWorktree: ${sharedCtx.worktreePath} (cd into it)\nFile-ownership map: ${sharedCtx.ownershipMapPath}\nNOTE: ACs already pre-classified as policy-deferred MUST NOT be implemented or routed — they are tracked as follow-ups.`
 
-  // --- Phase 1+2: dispatch the card's OWNER_AGENT (F-024), not general-purpose. ---
+  // --- Phase 1 (B7, v4.26.0) — SPECIALIST decomposition: each Phase-1 duty runs as its own
+  // agent ("ognuno fa una cosa"), with handoff via /tmp files so the owner's context stays
+  // clean of exploration noise. The old single mega-prompt ("you ARE claim+architect+
+  // plan-auditor+owner") made the owner absorb the whole exploration into its context.
+  const baselinePath = `/tmp/arch-baseline-${cardId}.md`
+  const phase1 = { architect: 'inline-fallback', audit: 'skipped-no-drift' }
+  let arch = null
+  try {
+    arch = await agentSafe(
+      `You are the Phase-1 context retriever for card ${cardId} (per ${REF}/implement.md Phase 1 step 3 / 5b). cd into the worktree ${sharedCtx.worktreePath}.\n\n${cardBrief}\n\n` +
+        `Explore the codebase exactly as your system prompt mandates for this card's scope (requirements + files_likely_touched: ${JSON.stringify(node.filesLikelyTouched || [])}). Write your COMPLETE untruncated findings (file paths, type signatures, patterns, high-risk paths) to ${baselinePath} — refresh it if a stale copy exists. The owner agent and the per-card reviewers will Read that file; keep your structured return MINIMAL.\n\n` +
+        `Also report missingPaths: every path in files_likely_touched that does NOT exist in the worktree (factual ls/test check — no judgement). Return: { ok, missingPaths:[...], note }`,
+      { label: `architect:${cardId}`, phase: 'Implement', agentType: 'codebase-architect',
+        schema: { type: 'object', required: ['ok'], additionalProperties: true, properties: { ok: { type: 'boolean' }, missingPaths: { type: 'array', items: { type: 'string' } }, note: { type: 'string' } } } }
+    )
+    if (arch && arch.ok) { phase1.architect = 'done'; g('phase1-architect', 'DONE', `baseline → ${baselinePath}${(arch.missingPaths || []).length ? ' · missing: ' + arch.missingPaths.join(', ') : ''}`) }
+    else g('phase1-architect', 'FALLBACK-INLINE', (arch && arch.note) || 'architect returned not-ok — owner explores inline')
+  } catch (e) {
+    if (e && e.transientExhausted) { noteDegraded('outage'); return { card: cardId, status: 'pending', gates } }
+    g('phase1-architect', 'FALLBACK-INLINE', `architect crashed (${String(e && e.message)}) — owner explores inline`)
+  }
+  const architectOk = phase1.architect === 'done'
+
+  // B7 — plan-auditor as a DETERMINISTIC DRIFT GATE, not an unconditional duplicate of the
+  // validation /prd already ran at card-creation time. It runs ONLY when execution-time drift
+  // is plausible: (a) a prior in-batch commit touched this card's declared surface (the
+  // strongest signal — the codebase moved under the card), (b) the architect found declared
+  // paths missing (stale card), or (c) review_profile 'deep' (Rule C escalation). Fresh
+  // single-card batches with no drift evidence skip it — /prd's validation stands.
+  const flt = (node.filesLikelyTouched || []).map(String)
+  const priorTouched = perCardResults.filter((r) => r.status === 'committed').flatMap((r) => r.filesChanged || [])
+  const inBatchOverlap = flt.length && priorTouched.some((f) => flt.some((c) => String(f).includes(c) || c.includes(String(f))))
+  const needAudit = reviewProfile === 'deep' || inBatchOverlap || (architectOk && (arch.missingPaths || []).length > 0)
+  let auditCorrections = []
+  if (needAudit && architectOk) {
+    try {
+      const audit = await agentSafe(
+        `Audit card ${cardId} for EXECUTION-TIME DRIFT per ${REF}/implement.md Phase 1 step 4 (you are plan-auditor; the card was already validated by /prd at creation time — your job is what changed SINCE). cd into ${sharedCtx.worktreePath}.\n\n${cardBrief}\nArchitecture baseline (Read it): ${baselinePath}\nDrift signals: ${inBatchOverlap ? 'prior in-batch commits touched this card surface; ' : ''}${(arch.missingPaths || []).length ? 'missing declared paths: ' + arch.missingPaths.join(', ') : ''}\n\n` +
+          `Check ONLY: (1) paths in files_likely_touched still exist; (2) type/field references in the requirements still correct per the baseline; (3) [ASSUMED] items now answerable from the code. Return PASS or the exact corrections — corrections amend the IMPLEMENTATION BRIEFING, never the backlog YAML.\n\nReturn: { verdict, corrections:[strings] }`,
+        { label: `plan-audit:${cardId}`, phase: 'Implement', agentType: 'plan-auditor',
+          schema: { type: 'object', required: ['verdict'], additionalProperties: true, properties: { verdict: { enum: ['PASS', 'FIXES_NEEDED'] }, corrections: { type: 'array', items: { type: 'string' } } } } }
+      )
+      auditCorrections = (audit && audit.corrections) || []
+      phase1.audit = audit && audit.verdict === 'FIXES_NEEDED' ? 'fixes' : 'pass'
+      g('phase1-audit', phase1.audit === 'fixes' ? 'FIXES-APPLIED-TO-BRIEF' : 'PASS', `drift gate (${reviewProfile === 'deep' ? 'deep' : inBatchOverlap ? 'in-batch overlap' : 'missing paths'})${auditCorrections.length ? ' · ' + auditCorrections.length + ' corrections' : ''}`)
+    } catch (e) {
+      if (e && e.transientExhausted) { noteDegraded('outage'); return { card: cardId, status: 'pending', gates } }
+      phase1.audit = 'skipped-error'
+      g('phase1-audit', 'SKIPPED', `auditor crashed (${String(e && e.message)}) — proceeding without audit (non-blocking)`)
+    }
+  } else if (needAudit && !architectOk) { phase1.audit = 'skipped-no-baseline'; g('phase1-audit', 'SKIPPED', 'drift signaled but no baseline — owner re-derives context inline') }
+  else g('phase1-audit', 'SKIPPED', 'no drift evidence — /prd creation-time validation stands')
+
+  // --- Phase 2: dispatch the card's OWNER_AGENT (F-024), not general-purpose. ---
+  const phase1Brief = architectOk
+    ? `Phase 1 already ran as specialist agents. READ the architecture baseline at ${baselinePath} BEFORE writing any code — it is your codebase context; do NOT redo the exploration.${auditCorrections.length ? `\nPlan-audit corrections (apply them as amendments to this briefing — do NOT modify the backlog YAML):\n${auditCorrections.map((c, i) => `  ${i + 1}. ${c}`).join('\n')}` : ''}`
+    : `Phase 1 fallback: the architect specialist was unavailable — do the Phase 1 claim+architect exploration yourself per ${REF}/implement.md and persist the baseline to ${baselinePath} before coding.`
   let impl
   try {
     impl = await agentSafe(
-      `Implement card ${cardId} per ${REF}/implement.md (Phase 1 claim+architect+plan-auditor, Phase 2 you ARE the owner_agent '${ownerAgent}') and ${REF}/completeness.md (Phase 2.5 + 2.5b AC-closure ledger). Run all gates/bash yourself.\n\n${cardBrief}\n\n` +
-        `POLICIES: G26 Phase-2 lint/tsc/test/build failing after the module's retry cap → buildBlocked:true + blockedGate. Build the AC Closure Ledger (one row per AC: implemented|unmet|policy-deferred). DO NOT silently defer; report unmet rows (excluding policy-deferred). Persist arch baseline to /tmp/arch-baseline-${cardId}.md and the diff to /tmp/diff-${cardId}.txt.\n\n` +
+      `Implement card ${cardId} per ${REF}/implement.md Phase 2 — you ARE the owner_agent '${ownerAgent}' — and ${REF}/completeness.md (Phase 2.5 + 2.5b AC-closure ledger). Run all gates/bash yourself.\n\n${phase1Brief}\n\n${cardBrief}\n\n` +
+        `POLICIES: G26 Phase-2 lint/tsc/test/build failing after the module's retry cap → buildBlocked:true + blockedGate. Build the AC Closure Ledger (one row per AC: implemented|unmet|policy-deferred). DO NOT silently defer; report unmet rows (excluding policy-deferred). Persist the diff to /tmp/diff-${cardId}.txt.\n\n` +
         `E4 OWNERSHIP RECONCILE (implement.md §11b — do this BEFORE returning): the card's MAY-EDIT includes files_likely_touched ∪ paths NAMED EXPLICITLY in this card's acceptance_criteria/definition_of_done (e.g. an ADR the DoD says to update, the data-model / ER doc for a schema change). Editing THOSE is in-scope. For any OTHER dirty file outside MAY-EDIT (another card's file, or unrelated): \`git checkout -- <file>\` to revert it (NEVER leave it orphaned), list it in revertedOutOfOwnership. Set fileDiffViolation:true ONLY if such an edit genuinely could not be reverted (then say why in note) — it is no longer a silent label.\n\n` +
         `Return: { epic, buildBlocked, blockedGate, unmetACs:[{n,text}], scopeFiles, mayEditPaths, revertedOutOfOwnership:[paths], fileDiffViolation, note }`,
       { label: `implement:${cardId}`, phase: 'Implement', agentType: ownerAgent,
@@ -497,28 +583,41 @@ async function runCard(cardId, cardPath) {
   const reviewSchema = { type: 'object', required: ['blocks', 'scopeExpansion'], additionalProperties: true,
     properties: { blocks: { type: 'array', items: { type: 'object', additionalProperties: true } }, scopeExpansion: { type: 'array', items: { type: 'object', additionalProperties: true } }, note: { type: 'string' } } }
   let reviewResults = []
+  const onErr = (e) => { if (e && e.transientExhausted) noteDegraded('outage'); return null }
+  // The standard SPECIALIST reviewer spawn — also reused as the JS-level fallback when the
+  // Codex companion dies at runtime (specialization integrity: the driver never reviews).
+  const stdReview = (ra) => agentSafe(
+    `You are ${ra}. Review card ${cardId} per ${REF}/review-cycle.md + ${REF}/codex-gate.md (your domain only). Run your gates on the COMMITTED-or-working state.\n\n${cardBrief}\nDiff: /tmp/diff-${cardId}.txt\n\n` +
+      `Report ONLY blocking failures that survive your retry cap as blocks:[{gate,domain,evidence}] (each MUST have non-empty gate AND evidence — F-014). Report legitimate findings BEYOND this card's AC as scopeExpansion:[{evidence,domain,withinOwnership,newAC}].\n\n` +
+      `Return: { blocks:[...], scopeExpansion:[...], note }`,
+    { label: `review:${cardId}:${ra}`, phase: 'Implement', agentType: ra, schema: reviewSchema }
+  ).catch(onErr)
   try {
     reviewResults = (await parallel(reviewers.map((ra) => () => {
-      const onErr = (e) => { if (e && e.transientExhausted) noteDegraded('outage'); return null }
       if (ra === 'codex') {
-        // Codex-light finder: a general-purpose agent drives the resolved companion (Bash, --wait).
+        // Codex-light finder: a general-purpose agent DRIVES the resolved companion (Bash,
+        // --wait). Driver role only — it never reviews; runtime failure → note flag, and the
+        // workflow spawns the real code-reviewer below.
         return agentSafe(
-          `You are the Codex review driver for card ${cardId} (review_profile=light — Codex is the SOLE finder since v4.18.0; you are NOT code-reviewer). Run the OpenAI Codex companion as a REVIEW-ONLY adversarial pass over this card's diff, then return its material findings in the schema below.\n\n${cardBrief}\nDiff: /tmp/diff-${cardId}.txt\nMAY-EDIT: ${JSON.stringify(mayEdit)}\n\n` +
+          `You are the Codex review DRIVER for card ${cardId} (review_profile=light — Codex is the SOLE finder since v4.18.0; you are a driver, NOT a reviewer). Run the OpenAI Codex companion as a REVIEW-ONLY adversarial pass over this card's diff, then return its material findings in the schema below.\n\n${cardBrief}\nDiff: /tmp/diff-${cardId}.txt\nMAY-EDIT: ${JSON.stringify(mayEdit)}\n\n` +
             `Run it in the FOREGROUND (it blocks; do NOT pass run_in_background):\n  node "${sharedCtx.codexScriptPath}" task "Review-only — DO NOT make edits, no --write flag. Adversarial review of card ${cardId} using the diff at /tmp/diff-${cardId}.txt. Focus: auth/permission boundaries, data-loss paths, race conditions, rollback safety, schema drift, invariant violations. Report ONLY material findings with file+line evidence." --wait\n` +
-            `Read the Codex output ONLY through a [codex]-trace-stripping filter. **Fallback**: if it exits non-zero / prints CODEX_NOT_FOUND / stays empty, set note='codex-unavailable' and perform the review yourself with the code-reviewer lens (your domain).\n\n` +
+            `Read the Codex output ONLY through a [codex]-trace-stripping filter. **Fallback**: if it exits non-zero / prints CODEX_NOT_FOUND / stays empty, return note:'codex-unavailable' with EMPTY blocks/scopeExpansion — do NOT review yourself (role boundary); the workflow spawns the real code-reviewer.\n\n` +
             `Map Codex BLOCKER/HIGH findings to blocks:[{gate:'codex-light',domain,evidence}] (each non-empty gate AND evidence — F-014). Map legitimate findings BEYOND this card's AC to scopeExpansion:[{evidence,domain,withinOwnership,newAC}].\n\n` +
             `Return: { blocks:[...], scopeExpansion:[...], note }`,
           { label: `review:${cardId}:codex`, phase: 'Implement', agentType: 'general-purpose', schema: reviewSchema }
         ).catch(onErr)
       }
-      return agentSafe(
-        `You are ${ra}. Review card ${cardId} per ${REF}/review-cycle.md + ${REF}/codex-gate.md (your domain only). Run your gates on the COMMITTED-or-working state.\n\n${cardBrief}\nDiff: /tmp/diff-${cardId}.txt\n\n` +
-          `Report ONLY blocking failures that survive your retry cap as blocks:[{gate,domain,evidence}] (each MUST have non-empty gate AND evidence — F-014). Report legitimate findings BEYOND this card's AC as scopeExpansion:[{evidence,domain,withinOwnership,newAC}].\n\n` +
-          `Return: { blocks:[...], scopeExpansion:[...], note }`,
-        { label: `review:${cardId}:${ra}`, phase: 'Implement', agentType: ra, schema: reviewSchema }
-      ).catch(onErr)
+      return stdReview(ra)
     }))).filter(Boolean)
   } catch (_) { /* parallel never rejects; nulls filtered */ }
+  // Specialization integrity — companion died at runtime: replace the empty driver result
+  // with a REAL code-reviewer pass (the same fallback the JS-level codexAvail gate uses).
+  if (reviewResults.some((r) => r && r.note === 'codex-unavailable')) {
+    reviewResults = reviewResults.filter((r) => !(r && r.note === 'codex-unavailable'))
+    g('review-codex', 'FALLBACK', 'companion failed at runtime → code-reviewer spawned (driver never reviews)')
+    const fb = await stdReview('code-reviewer')
+    if (fb) reviewResults.push(fb)
+  }
 
   // F-014 — only route well-formed blocks (non-empty gate+evidence).
   const blocks = reviewResults.flatMap((r) => (r.blocks || [])).filter((b) => b && b.gate && b.evidence)
@@ -576,7 +675,7 @@ async function runCard(cardId, cardPath) {
   let commitRes
   try {
     commitRes = await agentSafe(
-      `Commit card ${cardId} in worktree ${sharedCtx.worktreePath}. MECHANICAL — do NOT re-read reference modules.\n` +
+      `Commit card ${cardId} in worktree ${sharedCtx.worktreePath}. MECHANICAL — do NOT re-read reference modules. ROLE BOUNDARY: you NEVER modify file contents except the card YAML status/note fields and the ssot-registry row — source/doc changes are not yours.\n` +
         `Steps: (1) \`git status --porcelain\`; (2) stage = MAY-EDIT (${JSON.stringify(mayEdit)}) ∩ dirty — NEVER \`git add -A\`, NEVER \`git stash\`; if dirty has files OUTSIDE MAY-EDIT, do NOT stage them and set reconcileNote; (3) commit message \`[${cardId}] <concise>\`; ${doneStep} (5) 'nothing to commit' = already committed (record HEAD).\n` +
         `On COMMIT_LOCK: clear stale lock + retry once. Still locked → committed:false.\n\n` +
         `Return: { committed, commit, filesChanged, reconcileNote }`,
@@ -609,6 +708,8 @@ async function runCard(cardId, cardPath) {
     // B6 — which reviewers ACTUALLY ran (the gated matrix); Phase Final keys its slim/skip
     // decision on this, and per_card telemetry records it for the A/B spawn accounting.
     reviewersRun: reviewers,
+    // B7 — Phase-1 specialist record (architect: done|inline-fallback; audit: pass|fixes|skipped-*).
+    phase1,
     scopeFiles, archBaselinePath: `/tmp/arch-baseline-${cardId}.md`, gates,
   }
 }
@@ -795,6 +896,7 @@ if (!committed.length) {
     mergeResult = await agentSafe(
       `Auto-merge the batch worktree to ${TRUNK} per ${REF}/merge-cleanup.md (Phase 6 via /mw programmatic checksAlreadyPassed:true, Phase 6b status reconciliation, Phase 6c hygiene). Run git yourself.\n\n${projectBrief}\nWorktree: ${sharedCtx.worktreePath}\nBranch: ${sharedCtx.branch}\nmerge_strategy: ${mergeStrategy}\nCommitted cards: ${committed.map((r) => r.card).join(' ')}\nPhase-0 stash to restore (if any): see /tmp/batch-tracker-${firstCard}.md.\n\n` +
         `DETERMINISTIC POLICIES (NO prompts):\n` +
+        `• ROLE BOUNDARY: you are the OPS/GIT agent — you NEVER edit source or doc files. Reconciliation touches ONLY card YAML status fields + registry rows. A merge conflict on content is leave+report, never hand-resolved here.\n` +
         `• G24 → auto-merge via merge_strategy.\n` +
         `• F-030 HARD RULE: NEVER \`git add\`/commit code that did not pass the per-card gates. If the worktree is dirty with uncommitted code → DO NOT commit it; leave it, set uncommittedLeft:true, and report. NO "safety commit". Security/migration code is NEVER swept in.\n` +
         `• F-029 HARD RULE: Phase 6b reconciliation marks a card DONE ONLY if it has a real commit in ${TRUNK}..HEAD AND its gates are green. NEVER force a non-implemented card to DONE. Return forcedDone:[] (must be empty).\n` +
@@ -826,7 +928,7 @@ phase('Production')
 if (mergeResult && mergeResult.merged) {
   try {
     prodReadiness = await agentSafe(
-      `Run the post-merge Production Readiness checklist per ${REF}/production-readiness.md (Phase 7) over the batch's changed files. Auto-EXECUTE only stack-matched index/access-rule/cron deploys; REPORT (do not execute) env vars, feature flags, DB migrations, secrets, DNS. NON-BLOCKING.\n\n${projectBrief}\nChanged files: ${dedupe(committed.flatMap((r) => r.filesChanged || [])).join(', ') || '(derive from git)'}\n\nReturn: { autoExecuted:[...], manualItems:[...], note }`,
+      `Run the post-merge Production Readiness checklist per ${REF}/production-readiness.md (Phase 7) over the batch's changed files. Auto-EXECUTE only stack-matched index/access-rule/cron deploys; REPORT (do not execute) env vars, feature flags, DB migrations, secrets, DNS. NON-BLOCKING. ROLE BOUNDARY: you EXECUTE commands, you never edit repository files — a needed code/config change is reported as a manual item.\n\n${projectBrief}\nChanged files: ${dedupe(committed.flatMap((r) => r.filesChanged || [])).join(', ') || '(derive from git)'}\n\nReturn: { autoExecuted:[...], manualItems:[...], note }`,
       { label: 'production-readiness', phase: 'Production', agentType: 'general-purpose',
         schema: { type: 'object', required: ['manualItems'], additionalProperties: true, properties: { autoExecuted: { type: 'array', items: { type: 'string' } }, manualItems: { type: 'array', items: { type: 'string' } }, note: { type: 'string' } } } }
     )
@@ -877,7 +979,7 @@ function buildTelemetry() {
     // cost — total_tokens via budget.spent() delta; agent_count via counter; wall_clock_s stamped by the SKILL.
     total_tokens: totalTokens,
     agent_count: agentCount,
-    per_card: perCardResults.map((r) => ({ card: r.card, status: r.status, deferred: !!r.deferred, deferredClasses: r.deferredClasses || [], reviewers: r.reviewersRun || [], gates: (r.gates || []).length })),
+    per_card: perCardResults.map((r) => ({ card: r.card, status: r.status, deferred: !!r.deferred, deferredClasses: r.deferredClasses || [], reviewers: r.reviewersRun || [], phase1: r.phase1 || null, gates: (r.gates || []).length })),
     stats_requested: !!FLAGS.stats,
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "baldart",
-  "version": "4.25.0",
+  "version": "4.26.1",
   "description": "Claude Agent Framework - Reusable framework for coordinating AI agents and humans in software projects",
   "bin": {
     "baldart": "./bin/baldart.js"