메뉴
性能优化全流程。先 benchmark 立基线 → profile 找瓶颈 → 单一变量优化 → 再 benchmark 对比。禁止"凭感觉优化"，每个改动必须带数据证明。 Use when optimizing for performance, page speed, web vitals, or backend latency. Voice triggers: "性能优化", "/perf-flow", "提速".
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use whenever an AI agent executes a multi-step task to save tokens, parallelize tool calls, cache prompts, and keep replies under 2000 chars
Use when creating or developing, before writing code or implementation plans - refines rough ideas into fully-formed designs through collaborative questioning, alternative exploration, and incremental validation. Don't use during clear 'mechanical' processes
Bug 修复全流程。强制 /investigate 找根因，禁止"看似修好就交付"。/freeze 锁定文件 防止顺手"清理"无关代码；先写复现失败测试再写修复。 Use when reporting bugs, errors, 500s, stack traces, or "it stopped working" type issues. Voice triggers: "修 bug", "/bugfix-flow", "bugfix".
开发组合拳总入口。根据输入形式（需求文档 / 原型项目 / 一句话）× 开发范围（前端 / 后端 / 全栈） × 修饰符（带设计稿 / 数据库迁移 / 只联调 等）自动路由到正确的子 skill。同时覆盖 4 类非需求任务： bugfix / 重构 / 性能优化 / 第三方接入。 Use when asked to "接需求", "开始开发", "/dev-flow", "组合拳", "dev flow", or whenever the user hands over a development task and you need to decide which workflow skill to invoke. Voice triggers: "接需求", "开始开发", "组合拳". Note: 阶段切片型 skills（/dev-understand、/dev-implement、/dev-ship-retro）按阶段拆分， 与本路由按场景类型拆分的子 skill 互补；当用户明确说"只做某阶段"时优先匹配那些 skill。
Use when partner provides a complete implementation plan to execute in controlled batches with review checkpoints - loads plan, reviews critically, executes tasks in batches, reports for review between batches
代码重构 / 整理。任何重构必须先有覆盖待重构区域的测试网；分步原子 commit；每步 绿测，不允许"大爆炸式重构"。性能/行为变更属于 perf-flow / dev-flow，不在此范围。 Use when refactoring code, cleaning up tech debt, or restructuring without behavior change. Voice triggers: "重构", "整理代码", "/refactor-flow".
| name | perf-flow |
| description | 性能优化全流程。先 benchmark 立基线 → profile 找瓶颈 → 单一变量优化 → 再 benchmark 对比。禁止"凭感觉优化"，每个改动必须带数据证明。 Use when optimizing for performance, page speed, web vitals, or backend latency. Voice triggers: "性能优化", "/perf-flow", "提速". |
| version | 0.1.0 |
| applies_to | ["all"] |
| priority | P1 |
| usage_frequency | weekly |
| depends_on | [] |
| related | ["verification-before-completion"] |
页面慢、接口慢、内存高、CPU 高、bundle 大等性能问题。
确认 _playbook.md Part B.3 启动参数；额外 perf_target（待优化的页面/接口/作业）+ perf_metric（耗时？吞吐？内存？）+ perf_goal（目标值）。
move_agent_to_root；建知识库目录（极简 META.md）。
/benchmark（前端 web vitals）或 cursor-ide-browser 的 browser_profile_start/stop（性能录制）建立基线ab/wrk/k6 跑基准PERF_BASELINE.md：当前数值 + 测试条件 + 数据样本browser_profile_*、/benchmark 的 web vitals 拆解BOTTLENECK.md：top 3 热点 + 数据证据prompt: "瓶颈定位完成（BOTTLENECK.md）。优化方案："
options:
- "方案 A（首选）"
- "方案 B（更激进）"
- "需要更多 profile"
perf(scope): 优化 [瓶颈点]PERF_DIFF_[step].md/benchmarkPERF_FINAL.md：基线 vs 终态对比perf_goal → 回阶段 2 再 profile/qa（确保功能未退化）+ /review。
/ship → /land-and-deploy → /canary（重点监控同指标）。
append _lessons.md：瓶颈类型 + 优化手法 + 收益数据。
META.md、PERF_BASELINE.md、BOTTLENECK.md、若干 PERF_DIFF_*.md、PERF_FINAL.md、PR 链接。
无标准修饰符；可叠加 M4（如优化包含建索引/分表）→ 必出 MIGRATION_PLAN.md。
/dev-flow 路由到（也可被用户直接触发）/benchmark 与 cursor-ide-browser profile 工具/qa、/review/ship、/land-and-deploy、/canary