What problem does it solve?
Performance work often fails because teams fix the wrong thing: they profile the wrong phase, trust an instrument that cannot see the real cost, or build a fix before falsifying the hypothesis. This Skill provides a disciplined investigation method for any work whose success is a measured cost — slowness, jank, startup time, blocked JS, memory, render churn, or a slow query.
Core Features & Use Cases
- Cost framing and instrument routing: Define the phase, metric, workload, and symptom first, then route to the right instrument — React profiler, native CPU profiler (xctrace/Perfetto), React DevTools, Metro debugger, or SQL query planners (
EXPLAIN QUERY PLAN, EXPLAIN).
- Waste labeling and hypothesis falsification: Classify waste as TOO EARLY, TOO OFTEN, TOO MUCH, TOO SERIAL, WRONG SHAPE, N+1, RENDER CHURN, or CACHE HAZARD, then build a cheap falsifying control before writing any fix.
- Measured proof and honest reporting: Replay the same flow with paired measurements, revert fixes that cost UX, and report
[blocked] instead of inferred numbers when measurement is impossible.
- Use Case: A React Native app takes 12 seconds to open. Instead of assuming module loading, you arm a drift sampler, discover the JS thread is blocked outside React, run a second-open control to retire the lazy-loading hypothesis, and fix the actual hotspot with before/after blocked-ms proof.
Quick Start
Ask the AI to investigate why a specific screen, query, or startup phase is slow using the profile-and-optimize method, naming the phase and metric you care about.
