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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/InugamiDev/ultrathink-oss --skill pr-writer命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | pr-writer |
| description | Generate comprehensive pull request descriptions with summary, context, testing plan, and review guidance |
| layer | utility |
| category | developer-workflow |
| triggers | ["write PR","PR description","pull request","create PR","describe this PR"] |
| inputs | ["Branch diff (commits, changed files)","Related issue or ticket","Context about the change","Team conventions for PR format"] |
| outputs | ["Formatted PR title","Structured PR description with summary, context, changes, testing, and screenshots","Review checklist","Deployment notes if applicable"] |
| linksTo | ["commit-crafter","changelog-writer","code-review","git-workflow"] |
| linkedFrom | ["ship","git-workflow"] |
| preferredNextSkills | ["code-review","ship"] |
| fallbackSkills | ["docs-writer"] |
| riskLevel | low |
| memoryReadPolicy | selective |
| memoryWritePolicy | selective |
| sideEffects | [] |
Write pull request descriptions that enable efficient, thorough code review. A well-written PR saves reviewers time, reduces back-and-forth, and creates a permanent record of why changes were made. The PR description is not for you (you know what you did) — it is for everyone else.
A pull request communicates:
| Size | Files Changed | Lines Changed | Review Time |
|---|---|---|---|
| XS | 1-2 | < 50 | 5 min |
| S | 3-5 | 50-200 | 15 min |
| M | 5-10 | 200-500 | 30 min |
| L | 10-20 | 500-1000 | 1 hour |
| XL | 20+ | 1000+ | Split this PR |
Target: Small PRs (S-M). Large PRs get superficial reviews.
Gather information:
# What commits are in this branch
git log main..HEAD --oneline
# What files changed
git diff main..HEAD --stat
# Full diff for understanding changes
git diff main..HEAD
# Related issues
gh issue view <number>
Format: <type>: <concise description> (#issue)
Rules:
- Under 70 characters
- Imperative mood
- No period at end
- Include issue number if applicable
Good:
feat: add CSV export for transaction history (#312)
fix: prevent duplicate charges on double-click (#234)
refactor: extract auth middleware into separate package
Bad:
Updated the code
Fix
WIP: working on the new feature
Implementing the new user registration system with email verification
## Summary
[1-3 sentences explaining what this PR does and why]
## Context
[Link to issue/ticket. Background on why this change is needed.
What problem does it solve? What was the previous behavior?]
Closes #123
## Changes
- [Bullet point of each logical change]
- [Another change]
- [And another]
## Implementation Notes
[Optional: Explain non-obvious design decisions, trade-offs,
or alternative approaches you considered and rejected]
## Screenshots / Recordings
[For UI changes: before/after screenshots or screen recordings]
| Before | After |
|--------|-------|
|  |  |
## Testing
- [ ] Unit tests added/updated
- [ ] Integration tests pass
- [ ] Manual testing steps:
1. Go to [page/feature]
2. Do [action]
3. Verify [expected result]
## Deployment Notes
[Optional: Any special deployment steps, feature flags,
migrations, or rollback considerations]
## Checklist
- [ ] Code follows project conventions
- [ ] Self-reviewed the diff
- [ ] No console.log or debug statements
- [ ] No hardcoded secrets or credentials
- [ ] Documentation updated (if applicable)
- [ ] Accessibility checked (if UI change)
## Review Guide
**Start here:** `src/services/payment-service.ts` — This is the core change.
The rest of the files are supporting changes.
**Key decisions to review:**
1. I used an idempotency key approach instead of client-side debouncing
because [reason]. See `processPayment()` at line 45.
2. The retry logic uses exponential backoff with jitter. I considered
circuit breaker but it was overkill for this use case.
**What I am NOT confident about:**
- The error handling in `handleWebhook()` — I am not sure if we should
retry or dead-letter failed webhook events. Would appreciate input.
**Out of scope (intentionally):**
- Refactoring the webhook handler — tracked in #456
- Adding monitoring — tracked in #457
## Migration Details
**Type:** Database schema migration
**Reversible:** Yes (down migration included)
**Downtime:** None — backward compatible
**Data backfill:** Required, runs async after deploy
### Migration Steps
1. Deploy this PR (adds new column, nullable)
2. Run backfill job: `npm run migrate:backfill-user-roles`
3. Verify: `SELECT COUNT(*) FROM users WHERE role IS NULL` should be 0
4. Deploy follow-up PR (#460) that makes column NOT NULL
### Rollback Plan
1. Revert this PR
2. Run: `npm run migrate:down -- --to 20250615`
3. Verify application works without the new column
## Summary
Fix duplicate Stripe charges when users double-click the checkout button.
## Context
Users reported being charged twice for single purchases. The root cause
was that rapid button clicks could trigger multiple API calls before the
UI disabled the button.
Closes #234
Reported by: 3 users in support tickets SW-891, SW-895, SW-903
## Changes
- Add idempotency key generation to checkout session creation
- Pass idempotency key to Stripe charge API
- Add client-side request deduplication as defense-in-depth
- Add regression test for concurrent checkout attempts
## Testing
- [ ] Unit test: `checkout-service.test.ts` — concurrent payment test
- [ ] Integration test: verify Stripe idempotency with test mode
- [ ] Manual test:
1. Add item to cart, go to checkout
2. Rapidly click "Pay Now" 5 times
3. Verify only one charge appears in Stripe dashboard
4. Verify order confirmation shows single charge
## Summary
Add CSV and JSON export functionality for transaction history,
respecting current filters and visible columns.
## Context
Users with accounting needs requested the ability to export their
transaction data for import into QuickBooks and other tools.
Closes #312
Design: [Figma link]
RFC: [link to design doc if applicable]
## Changes
- New `ExportButton` component with format dropdown (CSV, JSON)
- `GET /api/transactions/export` endpoint with format query param
- Export respects active filters (date range, type, status)
- Rate limited to 10 exports per hour per user
- Streaming response for large datasets (no memory issues)
## Screenshots
| Export button | Format selection |
|--------------|-----------------|
|  |  |
## Testing
- [ ] Unit: ExportButton renders and triggers download
- [ ] Integration: API returns correct CSV/JSON format
- [ ] Integration: Rate limiting works (11th request returns 429)
- [ ] Manual: Export 10,000+ transactions without timeout
- [ ] Manual: Verify CSV opens correctly in Excel and Google Sheets
- [ ] Manual: Verify JSON is valid and parseable
## Deployment Notes
- Feature flag: `ENABLE_EXPORT` — deploy disabled, enable after QA
- No database migration needed
- No environment variable changes
## Summary
Extract authentication logic from route handlers into dedicated
middleware, reducing duplication across 23 endpoints.
## Context
Auth checks were copy-pasted across route handlers with slight
variations, leading to inconsistencies (some endpoints checked
token expiry, others did not). This refactor centralizes auth
into composable middleware.
Related: #445 (auth audit findings)
## Changes
- New `requireAuth` middleware — validates JWT, attaches user to request
- New `requireRole(role)` middleware — checks user role after auth
- New `optionalAuth` middleware — attaches user if token present, continues if not
- Updated all 23 route handlers to use middleware instead of inline auth
- Removed 450 lines of duplicated auth code
- No behavior changes — all existing tests pass unmodified
## Review Guide
**Start with:** `src/middleware/auth.ts` — the three new middleware functions.
Then spot-check 2-3 route handler changes to verify the pattern.
The route handler changes are mechanical (remove inline auth, add middleware).
I verified each one but they are not individually interesting to review.
## Testing
- [ ] All existing auth tests pass (no behavior change)
- [ ] New middleware unit tests for edge cases
- [ ] Integration: verified all 23 endpoints still work with valid/invalid/expired tokens
# Generate PR from commits using gh CLI
gh pr create \
--title "feat: add CSV export for transactions (#312)" \
--body "$(cat <<'EOF'
## Summary
Add CSV and JSON export for transaction history.
## Changes
- ExportButton component
- Export API endpoint
- Rate limiting
## Test plan
- [ ] Unit tests
- [ ] Manual export test
Generated with Claude Code
EOF
)"
Unified design foundations — design system architecture, tokens, component specs, visual principles, creative vision, figma integration, plus brand design system loader (66 real brands via DESIGN.md). Absorbs design, design-system, design-systems, design-principles, design-router, creative-vision, figma, design-md.
Render, summarize, and present markdown documents and structured content in multiple output modes
Ultra UI skill - combines Google's DESIGN.md spec (machine-readable design tokens) with the ui-ux-pro-max knowledge base (91 styles, 161 palettes, 73 font pairings, 161 products, 104 UX guidelines, 25 chart types). Generates lint-clean DESIGN.md files, validates token references and WCAG contrast, exports Tailwind/DTCG tokens, and diffs design systems version-over-version.
基于 SOC 职业分类
