菜单
Five-lens review — security (playbook IDs), QA (adversarial), architect (scale), UX (a11y), code (simplicity+elegance). Challenges own work.
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。
Eight-lens planning with persona protocol. Writes to tasks/todo.md. STOP if sideways. Reviews gotchas first. Auto-loads skills. For 2+ step tasks.
Context recovery — gotchas review, task state, git state, health check. Use at session start or after /clear.
基于 SOC 职业分类
| name | review |
| description | Five-lens review — security (playbook IDs), QA (adversarial), architect (scale), UX (a11y), code (simplicity+elegance). Challenges own work. |
| allowed-tools | Bash(git *), Bash(npm run *), Bash(npx *), Read, Glob, Grep |
Challenge own work first. "Is there a cleaner way?" "Would a staff engineer approve?"
Security — AUTH/AUTHZ/INJ/PAY/API rules from playbook. RSC boundary violations. Secrets in client. Webhook sigs.
QA — worst+best case tests exist? Auth/payment edge cases? Error boundaries + loading + empty states? npm test passes?
Architect — waterfalls? Correct fetch pattern? TanStack only where needed? Separation of concerns?
UX — mobile-first? WCAG 2.1 AA? MiHCM patterns correct? Skeletons not spinners?
Code — dead code? duplicates? beginner-readable? <300 lines? magic values? Elegance: hacky → "implement the elegant solution"
Output per finding: Severity · Rule ID · File:line · Issue (1-2 sentences) · Fix
Update tasks/gotchas.md with new patterns discovered.
