What problem does it solve? Performance complaints like "the app feels slow" are usually handled with guesswork. This Skill replaces guessing with a measure-then-fix workflow: it runs real tooling (Lighthouse, bundle analyzers, heap-snapshot diffing, Node/browser profilers, EXPLAIN ANALYZE, flame graphs) to capture a baseline, diagnoses against Core Web Vitals budgets and algorithmic-complexity patterns, applies a targeted fix, then re-measures to prove the delta. ## Core Features & Use Cases - Baseline measurement before diagnosis: Runs Lighthouse, source-map-explorer, webpack-bundle-analyzer, Node --prof, heap-snapshot pairs, or EXPLAIN ANALYZE depending on the target (page, bundle, function, memory, query, or API endpoint). - Budget-driven diagnosis: Checks measurements against a Core Web Vitals budget table (LCP < 2.5s, TBT < 200ms, bundle < 200KB gzip), an algorithmic-complexity table, and leak-pattern catalogues, citing a concrete number for every finding. - Fix and re-measure loop: Applies the smallest behavior-preserving fix, optionally benchmarks 2-4 fix variants against each other, then re-runs the identical measurement and reports a delta table with PASS/FAIL verdicts per metric. - Backend and database coverage: API benchmark mode (p50/p95/p99 serial and concurrent), constraint-vs-queue analysis, Postgres plan reading, slow-query-log workflow, connection-pool saturation diagnosis, runtime N+1 confirmation, and flame-graph profiling for Python (py-spy), Go (pprof), and Java (async-profiler). - Use Case: A user reports "kenapa halaman ini lemot". The Skill runs Lighthouse for a baseline, finds LCP at 2.9s against a 2.5s budget, traces it to an unoptimized hero image and a sequential await chain, fixes both, re-runs Lighthouse, and saves a report showing the measured delta. ## Quick Start Ask the assistant to run a performance audit on the slow page or endpoint, measure a baseline, fix the bottleneck, and re-measure the improvement.
