@dzhechkov/skills-feature-adr 1.3.34 → 1.3.36

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dzhechkov/skills-feature-adr",
-  "version": "1.3.34",
+  "version": "1.3.36",
   "description": "Adaptive Feature Development skill pack for Claude Code — 11-step pipeline with Complexity Router (S/M/L/XL), ADR-driven architecture, 15 agentic-qe skills, multi-agent fleet QE. Supports --full-qe, --full-qe-extended, --with-learning, and --knowledge-extractor modes.",
   "bin": {
     "skills-feature-adr": "./bin/cli.js"

package/templates/.claude/rules/feature-adr-ultracode.md

@@ -60,6 +60,12 @@ Without this, the Code agent returns null and the run stalls. With `coder: 'code
 detects the null, logs *"Claude unavailable (limit?) — falling back to Codex auto"*, and finishes
 the code + tests on Codex — no restart, no lost work.
 
+**Codex-landed barrier (Step 7.5):** Codex applies edits OUT-OF-BAND via its own runtime, so a naive
+pipeline runs Step-8 QE before the async write flushes and false-grades *"Step 7 never ran"* (grade D on
+real, landed code — observed on the goap-ed25519 fix). When Codex was the coder, feature-adr now polls
+`git status` (excluding pipeline artifacts) up to ~30s until the code changes appear and hands the
+confirmed file list to QE. Claude-coded runs are synchronous, so the barrier is skipped (zero cost).
+
 **Example invocations:**
 ```js
 // Plan on Codex, code+QE fall back to Codex only if Claude runs out, auto:

package/templates/.claude/skills/feature-adr/SKILL.md

@@ -366,7 +366,7 @@ to Codex — always ASKING first, always with a Claude fallback (never blocks):
 - **Model choice** → `args.codexModel` (`auto` default (Codex self-selects) · or an id your account exposes (e.g. `gpt-5.5`)).
 
 *Scenario:* an L/XL run hits the session limit during coding → with `coder: 'codex-fallback'` feature-adr
-logs *"Claude unavailable (limit?) — falling back to Codex auto"* and finishes on Codex, no restart.
+logs *"Claude unavailable (limit?) — falling back to Codex auto"* and finishes on Codex, no restart. Codex writes out-of-band, so feature-adr confirms the edits LANDED (`git status`, ~30s poll) before QE — no false "Step 7 never ran" grade.
 
 *Example:* `Workflow({ scriptPath: '.claude/workflows/feature-adr.js', args: { slug, description, tier:
 'M', planner: 'codex', coder: 'codex-fallback', qeReviewer: 'codex-fallback', codexModel: 'auto' } })  // 'auto' = Codex picks top; or pin e.g. 'gpt-5.5'`.

package/templates/.claude/workflows/feature-adr.js

@@ -113,9 +113,21 @@ if (code === null && (CODER === 'codex' || CODER === 'codex-fallback')) {
   if (codexCode) { code = { wrote: [FDIR + '/07_code_changes/change_manifest.md'], summary: String(codexCode).slice(0, 500) }; coderUsed = CODER === 'codex' ? 'codex' : 'codex-fallback' }
 }
 
+// Step 7.5: Codex-landed barrier. Codex applies edits OUT-OF-BAND via its own runtime; without this,
+// Step-8 QE reads the tree before the async write flushes and false-grades "Step 7 never ran" (grade D
+// on real, landed code — observed on the goap-ed25519 crypto fix). Poll git status (excluding pipeline
+// artifacts) up to ~30s until real code changes appear, and hand the confirmed file list to QE so it
+// reviews the ACTUAL landed changes. Claude-coded runs are synchronous → this barrier is skipped.
+let landedNote = ''
+if (coderUsed === 'codex' || coderUsed === 'codex-fallback') {
+  const barrierCmd = 'for i in 1 2 3 4 5 6; do n=$(git -C ' + REPO + ' status --porcelain 2>/dev/null | grep -vE "features/|[.]dz/|[.]agentic-qe/|roam/" | wc -l); [ "$n" -gt 0 ] && break; sleep 5; done; echo "changed=$n"; git -C ' + REPO + ' status --porcelain 2>/dev/null | grep -vE "features/|[.]dz/|[.]agentic-qe/|roam/" | head -40'
+  const probe = await agent('Confirm the Codex Step-7 edits have LANDED in the working tree BEFORE QE runs (Codex writes out-of-band). Run EXACTLY this via Bash and return its stdout verbatim, nothing else:\n' + barrierCmd, { label: 'code:confirm-landed', phase: 'Code' })
+  landedNote = '\n\nCODEX-CODED (out-of-band): review the CONFIRMED landed working-tree changes below — do NOT report "Step 7 never ran" if files are listed. If changed=0 the implementation genuinely did not land, then grade accordingly.\n' + String(probe || '(landed-probe failed)').slice(0, 1500)
+}
+
 // Step 8: QE (brutal-honesty, agentic-qe) + MANDATORY teach
 phase('QE')
-const qePrompt = 'Step 8 (QE - brutal-honesty review, agentic-qe) of /feature-adr for "' + DESC + '" (' + SLUG + '). Adversarially review the SHIPPED code (read it): correctness, edge cases, error handling, and the LOAD-BEARING property the ADR named (ASSERT it has a test - the recurring lesson). Grade A/B/C/D honestly. Assess code-test adequacy + doc-test presence. List CONFIRMED gaps with severity. Write ' + FDIR + '/08_qe_report.md. MANDATORY SELF-LEARNING STORE (close the loop, never skip): via Bash run `dz teach "<a durable reusable lesson from this feature - a rule/pattern/pitfall, NOT a checkpoint echo>" --reward <0.7-0.95> --domain <area>` for each genuine lesson (1-3 max, high-signal), then run `' + DZ + ' statusline --fa-record --slug ' + SLUG + ' --step "Step 8 QE" --recalled 3 --stored <count taught> --mode ' + MODE + ' --project ' + REPO + '` (run it verbatim via Bash, do not skip). Do NOT teach trivia or invent gaps. Return {grade, gaps, codeTestsAdequate, docTestsPresent}.'
+const qePrompt = 'Step 8 (QE - brutal-honesty review, agentic-qe) of /feature-adr for "' + DESC + '" (' + SLUG + '). Adversarially review the SHIPPED code (read it): correctness, edge cases, error handling, and the LOAD-BEARING property the ADR named (ASSERT it has a test - the recurring lesson). Grade A/B/C/D honestly. Assess code-test adequacy + doc-test presence. List CONFIRMED gaps with severity. Write ' + FDIR + '/08_qe_report.md. MANDATORY SELF-LEARNING STORE (close the loop, never skip): via Bash run `dz teach "<a durable reusable lesson from this feature - a rule/pattern/pitfall, NOT a checkpoint echo>" --reward <0.7-0.95> --domain <area>` for each genuine lesson (1-3 max, high-signal), then run `' + DZ + ' statusline --fa-record --slug ' + SLUG + ' --step "Step 8 QE" --recalled 3 --stored <count taught> --mode ' + MODE + ' --project ' + REPO + '` (run it verbatim via Bash, do not skip). Do NOT teach trivia or invent gaps. Return {grade, gaps, codeTestsAdequate, docTestsPresent}.' + landedNote
 let qe = null
 let qeReviewerUsed = 'claude'
 if (QE_REVIEWER !== 'codex') {