Name: argent-react-native-optimization
Author: software-mansion

Optimizes a React Native app by profiling first to find real bottlenecks, then sweeping for mechanical issues. Entry-point for all performance work. Use when the app feels slow, user asks to optimize, fix re-renders, reduce jank, or improve startup. Delegates to argent-react-native-profiler for meas

Rules Do not apply shotgun optimizations. Measure first, define what "good enough" looks like (target metric + threshold), fix the top offender, re measure honestly. Quick scan : react profiler renders for a live render count table. Identifies hot components instantly. Deep measure : load argent react native profiler skill. react profiler start → interact → react profiler stop → react profiler analyze . Inspect : react profiler component source per finding. react profiler fiber tree to trace component ancestry and render cost. Verify correctness before fixing, recollect information from steps above and make a logical conclusion whether the approach is worth undertaking. Fix : apply one fix. Validate with debugger evaluate before committing. Re measure : report whether the target metric improved, regressed, or stayed flat. Check for regressions in other areas. If no net benefit or unacceptable tradeoffs, revert. Profile for discovery, not only verification. Use the profiler to find issues static analysis missed, not only to confirm fixes. 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 e2e skill) before the first run so all subsequent cycles replay identical steps. React Compiler : if react profiler analyze reports reactCompilerEnabled: true , do NOT propose useCallback / useMemo / React.memo unless you confirmed compiler bail out via react profiler fiber tree (absent useMemoCache ). 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: Phase 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. See [references/lint rules.md](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](references/semantic checklist.md) for full checklist. Phase 3: Visual profiling 1. Load argent react native profiler skill, start dual profiling 2. Exercise key user flows (navigate screens the user specified, or all major flows) 3. Analyze with react profiler analyze + native profiler analyze + profiler combined report 4. Cross reference profiling results with Phase 1–2 findings 5. Fix highest impact issues. Re profile after architectural changes; batch mechanical fixes. If a recorded e2e test breaks after a fix (e.g., UI layout changed), follow argent create e2e skill to repair the e2e test rather than silently discarding it. Phase 4: Verify no regressions Navigate every screen and UI flow within scope, confirm each renders without errors. If no scope was specified, verify the entire app: cover all reachable screens via argent device interact . Use debugger log registry to check for runtime errors (if it returns status: "not connected" there is no log file; follow its guidance to reconnect first) and take screenshots to check for red/yellow error screens. Check for regressions introduced by fixes (e.g., fewer re renders but higher CPU, or new jank in a different screen). Main agent only. App wide optimization 1. Phase 1 : run lint centrally (one command), dispatch sub agents to fix per file in parallel 2. Phase 2 : one sub agent per checklist item for semantic sweep 3. Phase 3 : main agent profiles top offending screens; fixes architectural issues top down 4. Phase 4 : main agent navigates all screens to verify nothing crashes After the entire run, run lint again to verify no new issues were introduced with your changes. This also helps ensure you haven't missed any issues which could've been fixed.