SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsverify-before-commit

verify-before-commit

General & other

View repository

Updated 10-15 22:10

About this skill

Skill: verify-before-commit

WHAT

Purpose and context

  • Purpose: gate every commit — typecheck, lint, test, smoke before commit; after commit promote draft to wiki and close the task lifecycle.
  • Trigger (when to use): After any impl task, before git commit.
  • Non-goals: do not write new features, do not skip any step because of a "small change", do not change the task state-machine outside the command code-logger.py already declared.

Mental model

diff → typecheck → lint → test → qc-regression → task_lifecycle gate → smoke → commit (why) → promote draft (+ ## Origin) → log + index → task state=done.

Input and output contract

Field Required? Meaning
In implemented code changes yes gate target
In draft matching the feature in llmwiki/wiki/sources/draft/ no none (hotfix/refactor) → write "no draft — " to the log
In task: T-… in the draft frontmatter no none → skip step 8 (fail-open)
Out commit yes message states why, not what
Out promoted wiki page has ## Origin when there is a draft concepts/ or sources/
Out llmwiki/wiki/log.md + index.md yes update

Rules and capabilities

  • RULE-01 (MUST): All steps mandatory. No exceptions.
  • RULE-02 (MUST): Post-commit — mandatory, never skip.
  • Capabilities: run the repo's typecheck/lint/test/smoke (auto-detect via manifest); run the validator harness if present; write git commit + wiki + task lifecycle log.

Failure boundaries

  • Typecheck/lint/test red → blocked: fix then rerun, do not commit.
  • qc-regression.py --run red → blocked (bug already fixed has regressed or not yet fixed); no test qc-* → fail-open, do not block.
  • task_lifecycle.py blocks (task-ID wrong state-machine) → blocked; no file (old install) → skip.
  • No draft → do not promote, write the reason to the log (succeeded narrow). No task: id → skip step 8, not an error.

HOW

Main workflow

Step Type Inputs Action Outputs/exit Failure/next
W01 deterministic repo Type-check (detect command per repo) rc 0 error → fix, repeat
W02 deterministic repo Lint rc 0, warning note error → fix, repeat
W03 deterministic repo Test + qc-code reproduction test (3b, if harness exists) rc 0 fail → fix, repeat
W04 deterministic draft/task Pillar 3 gate task_lifecycle.py (if harness exists) rc 0 blocked → blocked
W05 judgment app Smoke check golden path end-to-end golden path runs broken → fix
W06 effect diff Commit, message why commit hash —
W07 effect draft Promote draft + ## Origin + log + index wiki page no draft → B01
W08 effect task: id Close task lifecycle via code-logger.py (best-effort) state=done no id → B02

Details of each step (source of truth for W01–W08; W01–W03 = steps 1–3b, W04 = 4, W05 = 5, W06 = 6 pre-commit, W07 = 6–7 post-commit, W08 = 8):

Pre-commit:

  1. RUN: <type-check-cmd> — detect from repo: npx tsc --noEmit (package.json) / go build ./... (go.mod). Fix all errors.
  2. RUN: <lint-cmd> — detect: npx eslint src/ / go vet ./.... Fix errors; note warnings.
  3. RUN: <test-cmd> — detect: npm test / go test ./.... Fix failures. 3b. qc-code reproduction test (if harness exists): RUN: [ -f harness/scripts/qc-regression.py ] && python3 harness/scripts/qc-regression.py --run — run the tests qc-* generated by /qc-code (reproduce found bugs). RED = a bug that was fixed has now regressed, or is not fixed yet. Deterministic, 0-token, does NOT call LLM. Fail-open if no test qc-* exists yet (project does not use qc-code yet) — do not block.
  4. Pillar 3 gate (if harness exists): RUN: [ -f harness/validators/task_lifecycle.py ] && python3 harness/validators/task_lifecycle.py --root . — hard block if a task-ID goes through the wrong state-machine (backward/jump/gate-bypass) or the draft points to a foreign task. No file (old install) → skip. This is the HARD version of step 8 best-effort below.
  5. Smoke check: verify golden path end-to-end.
  6. Commit with message: why, not what.

Post-commit — mandatory, never skip: 6. Find draft in llmwiki/wiki/sources/draft/ matching feature.

  • Promote to llmwiki/wiki/concepts/ or llmwiki/wiki/sources/ (permanent).
  • Add ## Origin section: Draft: <path> / Commit: <hash> — <msg> / Date: YYYY-MM-DD
  • CHECK: grep -l "## Origin" <promoted-file> — must return file.
  1. RUN: echo "promoted" >> llmwiki/wiki/log.md — then edit log properly; update llmwiki/wiki/index.md.
  2. Pillar 3 — close lifecycle (best-effort, fail-open): draft has frontmatter task: T-… → python3 harness/scripts/code-logger.py --task set <T-id> state=done note="<commit-hash>". Fail-open command (missing --task/id → skip, do not block commit). Confirm: --task show <T-id> in lifecycle proposed→approved→dispatched→done; --audit verify the transitions lie within the invariant chain.

No draft (hotfix/refactor)? Note "no draft — " in log.md, skip step 6. No task: id (old install / propose skipped mint)? Skip step 8 — fail-open, not an error.

All steps mandatory. No exceptions.

Branches

ID Kind Guard Behavior Skip / failure Rejoin
B01 conditional_required no draft (hotfix/refactor) write "no draft — " into log.md, skip step 6 post-commit — W08
B02 conditional_required draft has no task: id (old install / propose skipped mint) skip step 8 — fail-open — end
B03 capability_optional has harness/ (qc-regression, task_lifecycle, code-logger) run 3b, 4, 8 missing file → skip, fail-open next step

Validation and stopping

Deterministic: rc of typecheck/lint/test/qc-regression/task_lifecycle; CHECK: grep -l "## Origin" <promoted-file> must return files; --task show <T-id> + --audit confirm lifecycle. Needs review: smoke golden path. Stop when all pre + post steps are done; red at pre-commit means no commit.

Examples

  • Positive: repo has package.json → npx tsc --noEmit, npx eslint src/, npm test green, qc-regression.py --run green → commit "why" → promote draft DDMMYY-login-otp.md into wiki/concepts/ has ## Origin → code-logger.py --task set T-12 state=done note="<hash>".
  • Boundary/failure: go test ./... 1 test fails → stop at W03, fix and rerun, not yet committed. Hotfix has no draft → log "no draft — hotfix typo", skip promote.

Delivery — Output Report

After all main skill tasks complete, write a propose draft to the wiki.

Steps

1. Build the filename:

  • Format: DDMMYY-<ten>.md
  • DDMMYY = today (e.g., 020626 for 2 June 2026)
  • <ten> = 2–4 kebab-case words summarising what was done (e.g., landing-page-coteccons, brand-kit-fintech, ingest-auth-spec)

2. Write llmwiki/wiki/sources/draft/DDMMYY-<ten>.md:

---
type: draft
title: "DDMMYY-<ten>"
status: proposed
tags: [<skill-name>, output-report]
timestamp: YYYY-MM-DD
---

# DDMMYY-<ten>
**Type:** draft
**Status:** proposed
**Tags:** <skill-name>, output-report
**Proposed:** YYYY-MM-DD

## What
<One sentence — what this skill invocation produced or decided>

## Output
<Key artefacts, files created/modified, or decisions made>

## Files
| File | Action |
|------|--------|
| `path/to/file` | created / modified |

## Notes
- Invoked via: `/<skill-name>` skill

## Origin
- **Draft:** `wiki/sources/draft/DDMMYY-<ten>.md`
- **Commit:** _(filled by verify-before-commit)_
- **Date promoted:** _(filled by verify-before-commit)_

3. Update wiki index & log:

  • llmwiki/wiki/index.md — append one row: | [DDMMYY-<ten>](sources/draft/DDMMYY-<ten>.md) | draft | YYYY-MM-DD |
  • llmwiki/wiki/log.md — append: ## YYYY-MM-DD — <skill-name> — <ten>

Skip only when the skill produces zero artefacts and zero decisions (e.g., a pure display mode like /caveman-stats).