---

name: react-performance description: Optimize React rendering, bundle size, and data flow with profiler-led decisions. Use when reducing re-renders, fixing waterfalls, or deciding whether memoization is warranted in React. metadata: triggers: files: - '/*.tsx' - '/*.jsx' keywords: - waterfall - bundle - lazy - suspense - dynamic

React Performance

Priority: P0 (CRITICAL)

Decision Map

• Measure first: use React DevTools Profiler or app-level traces before adding memoization.
• Waterfalls: parallelize independent reads with Promise.all and stream slow branches behind Suspense.
• Rerenders: move state down, split context, and memoize only measured hot paths.
• Bundle cost: lazy-load heavy islands, remove oversized dependencies, and avoid barrel files that fight tree-shaking.
• Compiler-aware: if React Compiler is enabled, do not scatter useCallback/useMemo by reflex.

Recipe

1. Find the hotspot: profiler flamegraph, bundle analyzer, or waterfall trace.
2. Pick the narrow fix: state placement, context split, transition, lazy load, or memoization.
3. Retest after change: confirm the render count or blocking time actually improved.
4. Keep slow work off the main path: workers for heavy compute, streaming for slow I/O.

Verify

• Independent reads are parallel, not sequential.
• Effects are not driving derived state loops.
• Memoization exists because of a measured hotspot or memoized child boundary.
• Large lists are virtualized when needed.
• Heavy client code is split behind lazy or route-level boundaries.

Anti-Patterns

• No export *: Breaks tree-shaking.
• No Sequential Await: Causes waterfalls.
• No memo by superstition: profile before adding useMemo / useCallback.
• No effect-driven derived state: compute during render unless syncing external systems.
• No Heavy Libs: Avoid oversized dependencies when smaller choices exist.

References

• Framework Map
• REFERENCE.md

Canonical response anchors

When this skill applies, preserve the following domain terminology or equivalent concrete examples in the answer when relevant:

• Split Context