vietreport-admin/solution-erp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[CLAUDE] Workflow: adopt Harness-10 flat-refine + checklist-v2 + sleep-recovery-cmd (re-audit 3-workflow)
All checks were successful
Deploy SOLUTION_ERP / build-deploy (push) Successful in 5m16s
Details

Browse Source 
Re-audit Harness-8/9/10/10-refine + checklist v1/v2 + hmw vs canonical AI_INFRA
(mandate Harness-9 B1+B2): audit wf_13868efb-ea7 -> implement wf_ac43b5ff-7d1 ->
review wf_d482e10d-5dd. SE was BEHIND Harness-10 flat.

- run-trace SUBFOLDER->FLAT: hmw.js (:103 subMd + :52 schema + H4.5->H8 drift) +
  workflows/README full-rewrite + runs/README + session-start/end + agents/README
  Upgrade S72 + _ledger; 5 old S71 runs KEEP subfolder (C8 dual-accept). node --check OK.
- adopt 2 pending broadcasts (checklist-v2 + h10-flat-detector-refine).
- port /sleep-recovery-memory-l2 (A8, J2-tailored SE-only, floor intact, live skill).
- detector refine-b TAILORED-OUT (SE Workflow-tool no-CLI-bypass; containment
  git-diff+tracked+orphan-scan G-015).
- REVIEW (B2) caught 2 IMPLEMENT-self-assess missed (hmw.js:52 schema-stale +
  sleep-cmd auto-check un-wired overclaim) -> fixed (auto-check WIRED: budget.json
  +last_sleep_at + session-start/end INFORM). +3 minor fixed.
- em-main containment-check caught 1 reviewer residual-write (raw-Workflow no
  writeGuard) -> revert (reverse-finding #4). B3 self-correct: runs path 14->22.
- sleep-recovery-memory-l2 all = NO-OP (all periods already gisted). last_sleep_at set.

0 production code. State unchanged: Mig 53 / 88 tables / 306 test / 68 gotcha / menu 54
/ bundle BgNCjwsG/CBvh0vtf. adap-report + email ai_infra (5f511fe5c0f2).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
pqhuy1987
2026-06-18 14:04:03 +07:00
parent 4e4b5d47d1
commit 18fced6695
27 changed files with 668 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
docs/HANDOFF.md
View File

File diff suppressed because one or more lines are too long

15
docs/STATUS.md
View File

File diff suppressed because one or more lines are too long

41
docs/changelog/sessions/2026-06-18-S72-harness-10-flat-refine.md Normal file
View File

@ -0,0 +1,41 @@
# S72 (2026-06-18) — Harness-10 flat-refine + checklist-v2 adopt (re-audit 3-workflow) + sleep NO-OP
> **TL;DR:** Anh chỉ "sisters làm tốt mỗi SE thiếu" → re-audit Harness-8/9/10/10-refine + checklist + hmw vs canonical (3-workflow Harness-9 B1+B2) → SE BEHIND Harness-10 flat → migrate + adopt + report trung thực. **0 production code.** Full detail: [`adap-report`](../../governance/adap-reports/2026-06-18-Governance-harness-10-flat-refine-checklist-v2.md) + run-trace `.claude/workflows/runs/2026-06-18-harness-{audit-invest,fix-implement,fix-review}/`.
## Trigger
`/session-start` (bootstrap CLEAN — H1+H2 monitor 0-drift, cleanest since S59) → anh "trí nhớ có thay đổi lớn" → anh gõ `/sleep-recovery-memory-l2` (báo **Unknown** — SE chưa có) → em truy: AI_INFRA §J2-internal, **chưa broadcast** cho sisters → anh *"double check hết harness 8/9/10/10-refine + các checklist + hmw, lấy đầy đủ thông tin, điều chỉnh lại, report trung thực cho AI_INFRA"* → cuối: `/sleep-recovery-memory-l2 all` + `/session-end`.
## 3-stage workflow (mandate Harness-9 B1+B2)
- **INVESTIGATE** `wf_13868efb-ea7` — 4× investigator-codebase audit fidelity vs canonical. 1 fail-no-StructuredOutput + 2 truncated → em-main self-gate ground-truth (git/grep/glob) → `audit-synthesis.md` gap matrix.
- **IMPLEMENT** `wf_ac43b5ff-7d1` — 2× general-purpose file-disjoint (sleep-cmd port + runs/README flat) + em-main cluster (hmw.js + workflows/README + agents/README + session-cmds + _ledger).
- **REVIEW** `wf_d482e10d-5dd` — 3× reviewer adversarial. R3 PASS; R1/R2 PASS-with-concerns → 1 major + 4 minor ALL fixed.
## Gap (đo vs canonical) + fix
SE BEHIND Harness-10 flat: run-trace còn SUBFOLDER (`sub-md/`+`harvest/`), canonical=FLAT. Fixed:
- `hmw.js`: `:103` subMd `sub-md/`→`sub-<role>-<i>.md` + `:52` schema-desc + 5 stale-text + **H4.5→H8 doc-drift**. `node --check` PARSE-OK.
- `workflows/README.md` full-rewrite FLAT + roster 8→9 + detector-tailored-out + C8 · `runs/README.md` FLAT · session-start/end flat + dual-accept · agents/README Upgrade S72 · _ledger C8 note.
- **5 run cũ S71 GIỮ subfolder** (C8 no-history-rewrite). +2 broadcast 06-18 adopt (checklist-v2 + h10-flat-detector-refine).
- `/sleep-recovery-memory-l2` port (A8, 137 dòng, §J2-tailored SE-only, floor intact, live skill).
- **detector refine-b TAILORED-OUT** (anh chốt): SE Workflow-tool no-CLI-launcher-bypass; containment git-diff+tracked+orphan-scan (G-015). Documented, KHÔNG cargo-cult build.
## REVIEW caught (dogfood B2 value — review-workflow RIÊNG bắt cái implement-self-assess sót)
- `hmw.js:52` schema `subMdPath` desc stale `sub-md/` (operative `:103` đã flat, schema-desc lag) → fixed.
- sleep-cmd auto-check ≥7d **UN-WIRED overclaim** → **WIRED thật**: `budget.json` `+last_sleep_at` + `session-start:78` + `session-end:48` INFORM. grep-verified present (was zero).
- +3 minor (provenance cite → AI_INFRA-path · charter substring-anchor · `runs/README:31` dogfood-reword) fixed.
## em-main containment-check
1 reviewer agent (REVIEW via **raw Workflow tool** — KHÔNG hmw.js writeGuard) tự ghi `agent-memory/reviewer/MEMORY.md` (+2850B over-cap) → revert (record ở run-trace). → **reverse-finding #4** (raw-Workflow ≠ wrapper-governance; RECURRED vs `feedback_harness10_run_trace` lesson #2). `investigator-codebase/MEMORY.md` (M từ session-start = S71 audit-trail) commit-as-is + over-cap (28973) flagged → curate next.
## B3 self-correct (trung thực)
S71 report cite path tắt `runs/` + count "14" → thật `.claude/workflows/runs` = **22** (14 đúng tại `8c47bd0` → tăng). Đính chính trong adap-report mới.
## Sleep `/sleep-recovery-memory-l2 all` = NO-OP (dogfood live)
5/9 sub empty-archive (guard-rỗng) + 4/9 mọi period đã gisted (skip-if-gist + double-distill gen≥1, reviewer 2026-05.gist.md xác-nhận cover q1/q2) → **0 compressed**. `last_sleep_at` null→2026-06-18. Command port: executed-file → **executed-file+runtime** (guards fired đúng).
## State THẬT GIỮ NGUYÊN
Mig 53 · 88 bảng · 306 test (45D+261I) · 68 gotcha · menu 54 · bundle admin `BgNCjwsG`/user `CBvh0vtf`. **0 production code.**
## NEXT
- ⚠️ **restart CLI** (FLAT-runtime hmw.js + wired sleep-check + sleep-recovery skill + carry reviewer Cat-6 + H8-inherit).
- 🔴 **curate investigator-codebase L1 28973 OVER** + watch frontend-designer/test-specialist near-cap (chưa có archive). `budget.json` measured STALE → script re-run.
- Pending product/ops carry S69/S58/S59. Monthly audit 2026-07-01.

51
docs/governance/adap-reports/2026-06-18-Governance-harness-10-flat-refine-checklist-v2.md Normal file
View File

@ -0,0 +1,51 @@
# adap-report — Harness-10 flat-refine + checklist-v2 (re-audit + fix) — SE
- **id:** 2026-06-18-Governance-harness-10-flat-refine-checklist-v2
- **from:** se · **date:** 2026-06-18 (S72)
- **broadcasts adopted:** `2026-06-18-Governance-checklist-harness-9-10-v2` + `2026-06-18-Governance-h10-flat-detector-refine` (kèm re-audit fidelity Harness-8/9/10)
- **mandate:** Harness-9 PART-2 (2-workflow IMPLEMENT + REVIEW tách biệt + run-id) — anh giao *"double check hết Harness-8/9/10/10-refine + checklist + hmw, lấy đầy đủ thông tin, điều chỉnh lại, report trung thực"*.
- **run-ids (B3 evidence):** audit `wf_13868efb-ea7` · implement `wf_ac43b5ff-7d1` · review `wf_d482e10d-5dd`
- **reviewer_gate:** ✅ PASS sau-fix (R3 PASS + R1/R2 PASS-with-concerns → 1 major + 4 minor TẤT CẢ fixed)
## Bối cảnh (trung thực)
Anh chỉ ra SE chưa theo kịp Harness-10 flat-refine ("các project khác làm tốt, mỗi SE thiếu"). SE re-audit toàn bộ qua 3-stage workflow. **Baseline đo = canonical spec (ground-truth)** — `cross-project-status.md` đã stale (05-22, RAG Layer A), KHÔNG dùng để "chứng minh sisters ahead" (honest).
## Nấc thật per checklist-v2 (đo vs canonical)
**PART A — memory L2 (proposal/tailorable): substantially LANDED.** A1-A7 + A9 ok (executed-file / runtime / convention-met). **A8** (sleep-compress command) ĐÃ port — trước đó SE thiếu (chỉ nén ad-hoc). Watch: frontend-designer + test-specialist sát cap 25.6KB nhưng chưa có archive.
**PART B — adap 2-workflow (MANDATORY): LANDED.** B1/B2/B2.5/B3 = runtime (chính report này + 3 run-id = bằng chứng). B4 = convention-met (codified `adap-apply.md:38`, chưa có runtime instance của task-ngắn-cần-confirm).
**PART C + refine (MANDATORY): WAS BEHIND → NOW MIGRATED.** C1/C8/refine-a: run-trace SUBFOLDER→**FLAT** done (`hmw.js:103` + `workflows/README` + `runs/README` + `session-start/end` + `agents/README`; 5 run cũ S71 GIỮ subfolder, close-gate **dual-accept**). C2-C7 ok. **refine-b detector: TAILORED-OUT** (documented, anh chốt).
## Gap found → fixed
| gap (audit) | fix | nấc |
|---|---|---|
| run-trace subfolder (canonical=flat) | hmw.js + 5 docs → flat; 5 old runs keep subfolder | runtime (node --check OK + 2 new runs flat) |
| `/sleep-recovery-memory-l2` absent (A8) | ported tailored SE-only (137 dòng, floor intact, scope SE) | executed-file + live skill |
| 2 broadcast 06-18 unadopted | adopted (checklist-v2 + h10-refine) | runtime |
| detector refine-b | tailored-OUT + documented reasoning | convention (honest N/A) |
| hmw.js meta stale H4.5 two-tier | → H8 all-inherit (bonus doc-drift) | executed-file |
| B3 report imprecision (S71) | corrected (path + count, xem dưới) | runtime |
## Giá trị REVIEW tách biệt (dogfood B2 — mandate đúng)
Workflow REVIEW độc-lập bắt **2 lỗi mà workflow IMPLEMENT tự-chấm bỏ sót**:
1. **`hmw.js:52`** — schema `subMdPath` description còn ghi path cũ `sub-md/` (operative path :103 đã flat, nhưng bản-sao path trong schema-desc chưa đổi) → fixed.
2. **sleep-cmd auto-check ≥7d trigger UN-WIRED** — command tuyên bố session-start/end đọc `last_sleep_at` nhưng grep cả 2 file = 0 (overclaim) → **WIRED thật**: `memory-budget.json` `+last_sleep_at` + `session-start:78` + `session-end:48` INFORM step (grep-verified present sau fix).
→ Bằng chứng sống: 1-workflow-tự-chấm SẼ ship 2 lỗi này; review-workflow RIÊNG bắt TRƯỚC commit. PART-2 mandate hoạt động.
## Reverse-findings (B2.5) → AI_INFRA
1. **rename-migration audit phải grep runtime SCHEMA/contract-description strings**, không chỉ code-path + prose. Path hay bị duplicate ở (1) operative var, (2) comment, (3) schema/contract-desc — bản schema-desc dễ sót nhất vì là *data* không phải code engine đọc. (Bắt được hmw.js:52.)
2. **ported-command §-anchor phải grep/ls-verify TRONG sister repo** — port re-home path khác nguồn (SE budget ở `agent-memory/`, design-doc AI_INFRA-only). §-cite = claim "wiring tồn tại ở đây" → phải grep-prove như class 'wire BE' bug, áp cho governance-doc wiring.
3. **run-trace path-qualify:** SE `runs/` ở `.claude/workflows/runs/` không phải repo-root → bare `git ls-files runs` false-0 = kết-luận FALSE "không tracked"; luôn path-qualify + exit-branch `check-ignore` (negation `!.claude/**` last-match-wins vô-hình với text-read .gitignore). [chính là bẫy S71 report vấp]
4. **Raw Workflow tool KHÔNG mang writeGuard của `hmw.js`:** SE chạy REVIEW qua Workflow tool TRỰC-TIẾP (thay wrapper `hmw.js`) → KHÔNG inject anti-self-write guard → 1 sub `reviewer` theo charter "update MEMORY before return" tự ghi `agent-memory/reviewer/MEMORY.md` (+2850B → 27694 over-cap 25600; nếu N agent cùng-role = same-role-race trên 1 MEMORY). **Em-main git-status containment-check post-workflow BẮT** → revert (record đã nằm trong run-trace `sub-review-*.md`; B3 single-writer restored). Bài học: run-trace fan-out PHẢI qua `hmw.js` (mang writeGuard + return-delta) HOẶC replicate writeGuard trong prompt raw-Workflow. Đúng class mà `hmw.js` RUN-TRACE writeGuard (S71) sinh ra để chặn — **tái phát khi bypass wrapper**. (Đề-xuất infra: ghi rõ "raw Workflow tool ≠ project-wrapper governance" trong floor.)
## Honest caveats
- **Audit Part A auditor (1/4) failed no-StructuredOutput + Part C/H8 truncated** → em-main self-gate ground-truth từ đĩa (git/grep/glob), disclosed — valid per `feedback_agent_kill_recovery`.
- **Detector TAILORED-OUT (không build):** SE chạy qua Anthropic Workflow tool (KHÔNG có CLI-launcher bypass-surface như node-CLI của AI_INFRA hmw.js); containment = git-diff + run-folder tracked + ledger orphan-scan (G-015). refine nói detector = "chuyện nội bộ mỗi dự án tự quyết". SE **KHÔNG claim** đã adopt detector — đây là quyết định tailored có lý-lẽ, không phải né.
- **B3 self-correction (trung thực):** report S71 của SE cite path tắt `runs/` + count "14"; thật = `.claude/workflows/runs` (**22 file**, "14" đúng tại commit `8c47bd0` rồi tăng lên 22). Folder commit thật; chỉ path-string + count cũ sai → đã sửa.
- **Runtime:** hmw.js flat `node --check` PARSE-OK, nhưng RUN-TRACE runtime cần CLI restart để chạy thật (forward-looking — frontmatter/command no hot-reload).
## Net
SE **parity Harness-10 flat**. **0 production code** — state THẬT GIỮ NGUYÊN: Mig 53 · 88 bảng · 306 test · 68 gotcha · menu 54 · bundle admin `BgNCjwsG`/user `CBvh0vtf`. 3 run-trace folder (audit/implement/review) git-tracked = bằng chứng.