菜单
Prepare code for commit by checking for bugs, ensuring consistency, removing debug code, and running checks until green. Use when finishing a feature, before committing, or when asked to "make it green" or "prep for commit".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Multi-phase audit pipeline with independent review agents. Use when finishing tickets, validating work against acceptance criteria, checking architecture compliance, or reviewing code against skill standards. Use /audit to run the full pipeline, /audit ac for acceptance criteria only, /audit arch for architecture only, /audit skill for skill compliance only.
JS/TS monorepo with pnpm workspaces and Turborepo. Use when setting up a monorepo, adding packages, or auditing config.
Next.js App Router data layer: SSR hydration, React Query, auth patterns. Use when creating pages, adding tabs, or auditing data fetching.
Manage project changelog. Use /project log after completing work, /project release when ready to ship.
React component development with shadcn/ui, React Hook Form, and Zod. Use when creating, extracting, or auditing React UI components.
Extract learnings from conversations and update project knowledge. Use when you've learned something valuable, solved a tricky problem, or notice patterns emerging. Sub-commands: capture (save observations before compaction), finalize (process into skill updates), migrate (move legacy principles into skills).
基于 SOC 职业分类
| name | finalize |
| description | Prepare code for commit by checking for bugs, ensuring consistency, removing debug code, and running checks until green. Use when finishing a feature, before committing, or when asked to "make it green" or "prep for commit". |
Prepare code for commit by systematically checking for issues and running checks until everything passes.
Invoke /finalize when:
Note: This skill prepares code for commit but does NOT create the commit itself.
Identify scope - What files were changed in this session?
git diff --name-only HEAD
git diff --cached --name-only
Understand context - What was implemented? This informs what edge cases to check.
Review changed files for common bugs and edge cases.
Null/Undefined Handling:
Edge Cases:
Logic Errors:
Type Safety:
any typesas) are justifiedConfigure this section for your project's high-risk areas.
Common critical areas include:
- Authentication/Identity: Token handling, session management, privilege escalation
- Data pipelines: Idempotency, deduplication, ordering guarantees
- External integrations: Rate limits, retry logic, credential handling
- Infrastructure provisioning: State drift, rollback strategy, resource cleanup
Add your project-specific critical areas to CLAUDE.md.
If changes touch critical areas, do multiple passes:
Verify code follows project patterns.
Naming:
Imports:
Patterns:
Structure:
Remove debug artifacts and ensure code is production-ready.
Search for and remove:
# Find console.logs in changed files
git diff --name-only HEAD | xargs grep -l "console.log" 2>/dev/null
# Find debugger statements
git diff --name-only HEAD | xargs grep -l "debugger" 2>/dev/null
# Find commented-out code blocks
git diff HEAD | grep "^+" | grep -E "//.*[a-zA-Z]+.*\(|/\*"
Remove these:
console.log() - Remove unless it's intentional production loggingconsole.debug() - Remove allconsole.warn() - Keep only if warning is useful in productionconsole.error() - Keep if it's error reportingdebugger statements - Remove all// TODO from this session - Keep and warn user (see below)Production Logging Guidelines:
console.error() - OK for actual errors that should be visibleSearch for TODOs in changed files:
git diff --name-only HEAD | xargs grep -n "TODO\|FIXME\|XXX\|HACK" 2>/dev/null
Action:
Run the test suite to catch regressions:
# Use your project's test command
npm test
# or: yarn test, pnpm test, pytest, dotnet test, etc.
If tests fail:
Test Coverage:
Run the full check suite and fix issues iteratively.
# Use your project's check/lint command
npm run check
# or: npm run lint && npm run typecheck && npm run build
1. Run check command
2. If passes → Done!
3. If fails:
a. Identify the failing step (format/lint/typecheck/build)
b. Fix the issues
c. Go to step 1
Format issues:
Lint issues:
Type issues:
Build issues:
If after 5 iterations issues persist:
After all phases complete, present summary:
## Finalize Summary
### Changes Reviewed
- {N} files checked
### Issues Found & Fixed
- Removed {N} console.log statements
- Fixed {N} lint errors
- Fixed {N} type errors
### Warnings
- {N} TODOs remain in code:
- `path/to/file.ts:123` - TODO: description
### Status
✅ All checks passing - Ready for commit
### Files Changed
{list of files modified during finalize}
Before completing:
// ❌ Leaving debug code
console.log('data:', data) // Remove!
debugger // Remove!
// ✓ Production-appropriate logging
logger.error('Failed to process request', { id, error })
// ❌ Silently removing TODOs
// (just delete TODO comments without telling user)
// ✓ Warn about TODOs
// "Found 2 TODOs in changed files. Should I address these or leave them?"
// ❌ Giving up after first failure
// "Check failed, here are the errors"
// ✓ Iterating until green
// Fix errors → re-run → repeat until passing
// ❌ Creating a commit
git commit -m "..." // Don't do this!
// ✓ Just prep the code
// "All checks passing. Ready for you to commit."
CLAUDE.mdpackage.json (or project-specific config)