performance-optimizer_skill

Overview

This skill is a performance optimization expert for applications and infrastructure, focused on practical analysis and actionable improvements. It combines measurement, profiling, code and database tuning, caching, concurrency, I/O optimizations, and monitoring to deliver measurable speed and resource improvements. Use it to find bottlenecks, apply targeted fixes, and verify impact with repeatable metrics.

How this skill works

I start by establishing baseline metrics (latency percentiles, throughput, CPU/memory, error rate) and run targeted profilers for the runtime and database. I identify hotspots using flamegraphs, CPU/heap profilers, slow query logs and tracing, then propose and implement fixes at code, database, caching, concurrency and I/O layers. Finally I verify changes with load tests and benchmarks and iterate until targets are met.

When to use it

• When p95 or p99 latency exceeds your SLA
• When CPU or memory steadily climbs under load
• When database queries dominate request time
• Before scaling hardware to avoid wasted cost
• When an application shows N+1 queries, blocking I/O, or excessive allocations

Best practices

• Measure first: capture p50/p95/p99, throughput, resource usage and trace samples
• Use appropriate profilers (flamegraph, perf, massif) and interpret call stacks before changing code
• Prefer algorithmic improvements and reduced allocations over micro-optimizations
• Introduce caching and connection pooling with clear invalidation and size limits
• Run realistic load tests and benchmarks to validate gains and prevent regressions

Example use cases

• Reduce API p95 latency by identifying and removing blocking database calls with JOINs and bulk fetches
• Lower memory usage by eliminating unnecessary copies and pre-allocating collections
• Increase throughput by converting sequential awaits into concurrent tasks with tokio::join! or rayon
• Fix N+1 query patterns by refactoring to joins or bulk queries and adding appropriate indexes
• Add multi-level caching (L1 in-memory, Redis, CDN) to offload database and improve tail latency