使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/software-mansion/argent --skill argent-react-native-optimization开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Rules
- Do not apply shotgun optimizations. Measure first, define what "good enough" looks like (target metric + threshold), fix the top offender, re-measure honestly.
- Profile for discovery, not only verification. Use the profiler to find issues static analysis missed, not only to confirm fixes.
- Verify before fixing: recollect the evidence and conclude whether the fix is worth undertaking. Validate a fix with
debugger-evaluatebefore committing it. - One fix per cycle for architectural changes. Mechanical batch fixes (inline styles, index keys) can be grouped; re-profile once after the batch. When the measurement involves device interaction, record it as an e2e test (
argent-create-e2eskill) before the first run so every cycle replays identical steps. - React Compiler: if the project uses it (
babel-plugin-react-compilerin the Babel config, orexperiments.reactCompiler: truein the Expo config), skip memoization fixes (React.memo,useMemo,useCallback, hoisting inline objects and functions) in every phase; the compiler does them. Keep them for components the compiler skips (compilationMode: "annotation","use no memo", asourcesfilter). Where the profiler report marks a component ⚠️ (bailout suspected), fix what stops the compiler. - Sub-agents: Phases 1–2 dispatch sub-agents: one per file for lint results, one per checklist item for semantic. Sub-agents CANNOT touch the device - all profiling and E2E verification must happen in the main agent.
Pipeline
Lint and semantic sweeps catch deterministic issues cheaply. Profiling finds runtime bottlenecks that static analysis misses. Do both.
Copy this checklist into your TODO list:
Optimization Progress:
- [ ] Phase 1: Lint sweep (deterministic: catch mechanical issues without a running app)
- [ ] Phase 2: Semantic sweep (judgment: memoization, lists, animations, etc.)
- [ ] Phase 3: Baseline profile (find real bottlenecks, fix top offenders)
- [ ] Phase 4: Verify no regressions (crashes, errors, red screens)
- [ ] Re-run lint to confirm the fixes added no new issues and none were missedPhase 1: Lint sweep
Run ESLint once at the project root with a comprehensive RN performance ruleset. Dispatch sub-agents to fix results: one per file, in parallel. See references/lint-rules.md for ruleset and procedure.
Phase 2: Semantic sweep
Review each area requiring judgment: memoization, list rendering, animations, async patterns, effect cleanup, state hygiene, context architecture. Dispatch one sub-agent per checklist item. See references/semantic-checklist.md for full checklist.
Phase 3: Profiling
- Load the
argent-react-native-profilerskill and follow its workflow (React and native profiling together) on the key user flows: the screens the user named, or all major flows. It covers analysis, drill-down, and re-profiling after each fix. - Cross-reference the profiling results with the Phase 1–2 findings, and fix the highest-impact issues top-down. Re-profile after architectural changes; batch mechanical fixes.
Phase 4: Verify no regressions
Navigate every screen and UI flow within scope (the entire app if no scope was given, via argent-device-interact) and confirm each renders without errors. Check debugger-log-registry for warnings and errors (Grep its log file). Argent hides LogBox toasts once a debugger-* tool connects, so screenshots catch only full-screen red errors. Look for regressions the fixes introduced (e.g. fewer re-renders but higher CPU, or new jank on a different screen). Main agent only.