performance-profiler

Use clinic.js to generate CPU flamegraphs with 'clinic flame -- node dist/server.js' while applying load with autocannon. For memory issues, capture heap snapshots via v8.writeHeapSnapshot and compare them in Chrome DevTools to spot leaks.

Run 'py-spy record -o flamegraph.svg --pid <pid>' against a running process to capture a flamegraph without code changes, or use 'py-spy top' for a live view. For function-level detail, use cProfile with pstats to rank functions by cumulative time.

Enable query logging in your ORM or add a query counter, then assert the query count per request in tests. Fix N+1 patterns by replacing per-row queries with a single JOIN or batched fetch, and confirm with EXPLAIN ANALYZE.

Both are supported for load testing; k6 uses JavaScript test scripts with staged ramp-up patterns and threshold assertions like p(95)<500ms. The Skill's k6 examples include setup/teardown hooks, custom metrics, and error rate tracking.

Optimizing without a baseline risks fixing the wrong bottleneck and leaves no way to verify improvement. Record P50, P95, P99 latency, RPS, error rate, and memory usage first, then re-measure after each single change to confirm causation.

No, the performance_profiler.py script uses only the Python standard library. It scans a project directory for large files, dependency counts from package.json, requirements.txt, and go.mod, and bundle size indicators, with optional JSON output for CI.