performance-expert

# Performance Expert Skill Backend, database, and infrastructure performance. This skill owns the server-side and delivery-layer of a stack: request latency, data access, caching, background work, and what ships over the wire. Render-time work inside React components — re-renders, memoization, virtualized lists, Profiler traces — is a different diagnosis with different tooling. Route it to `react-component-performance` instead of duplicating it here. ## Contract Inputs: - A performance symptom with a surface: an endpoint, a query, a job, a build output, or a metric that regressed. Outputs: - A ranked list of bottlenecks with the measurement that proves each one, plus a targeted fix per bottleneck. Creates/Modifies: - None on its own. It diagnoses and prescribes; edits happen in the caller's workflow. External Side Effects: - None. Profiling and load-testing commands are prescribed, not executed unattended. Delegates To: - `react-component-performance` for React render hotspots, re-render churn, memoization, and Profiler-driven component work. - `workspace-performance-audit` when the target is a whole monorepo rather than one surface. ## When to Use - Improving API response times - Optimizing database queries, indexes, and aggregations - Diagnosing N+1 query patterns - Implementing caching strategies and invalidation - Moving heavy work into background jobs or queues - Profiling a server process or tracing a slow request - Analyzing shipped bundle size and asset delivery - Configuring CDN, cache headers, or edge caching ## Project Context Discovery 1. Check `.agents/memory/` for performance architecture context (e.g., any `*performance*` or `*architecture*` files) 2. Identify performance tools (APM, load-testing harness, profiler) 3. Review existing optimizations and caching strategies 4. Check for `[project]-performance-expert` skill ## Core Performance Principles ### Backend **API Response Times:** Target < 200ms (p95), caching, background jobs, connection pooling **Query Optimization:** Indexes, projections (`select`), pagination, database-side aggregation **Background Processing:** Queues for heavy operations, async for non-critical tasks, no blocking work in request handlers ### Database **Indexes:** On frequently queried fields, compound indexes, monitor usage **Queries:** Filter in `where`, `select` only needed columns, sort with indexes, avoid sequential scans ### Infrastructure **CDN:** Edge caching, correct cache headers, static assets off the origin **Serverless:** Cold start optimization, memory allocation, provisioned concurrency ### Delivery Layer **Bundles:** Code splitting by route, dynamic imports for heavy modules, tree shaking, drop unused dependencies **Assets:** WebP images, subset fonts, minify CSS/JS, Gzip/Brotli compression **Server rendering:** Static generation for static content, incremental revalidation, framework image and font pipelines ## Performance Metrics ### Backend - **API p50:** < 100ms - **API p95:** < 200ms - **DB Query p95:** < 50ms - **Error Rate:** < 0.1% ### Delivery - **LCP:** < 2.5s - **TTFB:** < 800ms - **Initial bundle:** < 200KB ## Quick Checklist ### Backend - [ ] Database queries optimized - [ ] Indexes created and used - [ ] Caching implemented - [ ] Background jobs for heavy operations - [ ] Connection pooling configured ### Delivery - [ ] Bundle size < 200KB initial - [ ] Code splitting implemented - [ ] Images optimized and lazy loaded - [ ] Compression and cache headers configured Before reporting a measured speedup or regression, vet the number with the `pstack` skill's benchmark checklist reference: name the limiter, confirm every side was tuned, count errors, and repeat the run. --- **For database query optimization code, caching strategy implementation, N+1 query solutions, background-job patterns, infrastructure tuning, performance testing commands, and detailed checklists, see:** `references/full-guide.md`