Skill: Performance OptimizerAI-consumed reference. Optimized for Claude to read during execution. Human-readable explanation: see docs/architecture/HIERARCHICAL_PLANNING.md or docs/getting-started/ depending on topic.
Identify and resolve performance bottlenecks. Measure first, optimize second.
Principles
- Measure first, optimize second
- Optimize the bottleneck, not everything
- 80/20: 80% of slowness is in 20% of code
Profiling
| Layer | Tools | Metrics |
|---|---|---|
| Frontend | Lighthouse, DevTools | FCP, LCP, TTI, CLS |
| Backend | APM, profilers | Response time, throughput |
| Database | EXPLAIN, slow query log | Query time, index usage |
| Memory | Heap snapshots | Allocation, leaks |
Frontend
Core Web Vitals: LCP < 2.5s, FID < 100ms, CLS < 0.1
Quick wins: loading="lazy" images, lazy(() => import(...)) code splitting, debounce handlers, useMemo for expensive computations. Optimize images (WebP), cache headers, preload fonts.
Backend
| Issue | Solution |
|---|---|
| N+1 queries | Eager loading, batching |
| Missing indexes | Add appropriate indexes |
| Unbounded queries | Pagination, limits |
| Sync blocking | Async/parallel processing |
| No caching | Cache hot data |
Database
EXPLAIN ANALYZE for slow queries. Look for Seq Scan (bad) vs Index Scan (good).
| Query Pattern | Index Type |
|---|---|
| Exact match | B-tree |
| Range | B-tree |
| Full-text | GIN/GiST |
| JSON | GIN |
Caching
| Level | TTL | Use Case |
|---|---|---|
| Browser | Hours-Days | Static assets |
| CDN | Minutes-Hours | API responses |
| Application | Seconds-Minutes | Computed data |
Invalidation: time-based, event-based, or version-based.
Memory
| Leak Cause | Fix |
|---|---|
| Event listeners | Cleanup in useEffect/destroy |
| Closures | Null out references |
| Growing collections | WeakMap, clear periodically |
| Timers | clearInterval/clearTimeout |
Optimization Workflow
- Profile → capture baseline metrics
- Identify bottleneck → focus on hottest path
- Set target metric → define measurable goal
- Apply fix → implement targeted optimization
- Re-measure → compare against baseline
- Verify → confirm no regressions, tests pass