<skill_overview> <purpose>Improve runtime performance through measurement-driven changes</purpose> <triggers> <trigger>Profiling slow functions or scripts</trigger> <trigger>Optimizing hot paths</trigger> <trigger>Reducing memory pressure</trigger> <trigger>Improving startup time</trigger> </triggers> <sources> <source url="https://nodejs.org/api/perf_hooks.html">Node.js Performance Hooks</source> </sources> </skill_overview> <measurement> <rules> <rule>Measure before and after every optimization</rule> <rule>Use performance marks and measures instead of Date.now()</rule> <rule>Benchmark with representative data sizes</rule> </rules> <example> <code> import { performance, PerformanceObserver } from "node:perf_hooks";
const obs = new PerformanceObserver((items) => { const entry = items.getEntries()[0]; console.log(entry.name, entry.duration); performance.clearMarks(); performance.clearMeasures(); }); obs.observe({ type: "measure" });
performance.mark("start"); doWork(); performance.mark("end"); performance.measure("doWork", "start", "end"); </code> </example> </measurement> <hot_paths> <rules> <rule>Avoid unnecessary allocations inside tight loops</rule> <rule>Precompile and reuse regular expressions</rule> <rule>Prefer simple loops over chained array methods in hot paths</rule> <rule>Do not JSON.parse or JSON.stringify inside loops</rule> </rules> </hot_paths> <data_structures> <rules> <rule>Use Map/Set for large dynamic key sets</rule> <rule>Use arrays for dense numeric indices</rule> <rule>Use plain objects for small fixed key sets</rule> </rules> </data_structures> <async_throughput> <rules> <rule>Run independent tasks in parallel with Promise.all</rule> <rule>Avoid serial awaits when concurrency is safe</rule> <rule>Limit concurrency to prevent overload</rule> </rules> <example> <code> const [user, orders] = await Promise.all([ fetchUser(userId), fetchOrders(userId) ]); </code> </example> </async_throughput> <memory_management> <rules> <rule>Clear timers and intervals when no longer needed</rule> <rule>Remove event listeners to avoid leaks</rule> <rule>Bound caches with size or TTL</rule> </rules> </memory_management> <anti_patterns> <avoid name="premature_optimization">Optimizing without profiling data</avoid> <avoid name="chatty_allocations">Creating new objects in tight loops</avoid> <avoid name="unbounded_cache">Growing in-memory caches without limits</avoid> </anti_patterns>