Find agent skills by outcome

Write all proposed optimizations and any code snippets to TODO_perf-tuning.md only. Do not create an...

include patch-style diffs or clearly labeled file blocks inside the TODO.

In TODO_perf-tuning.md

include:

- Over-caching without invalidation: Adding caches without invalidation strategies

serving stale data and creating consistency bugs

- Memory leaks under load: Allocations growing without bound in long-running processes

leading to OOM crashes in production

Coding & Debugging

- Synchronous blocking in async code: Blocking the event loop or thread pool with synchronous operat...

destroying concurrency benefits

- Missing database indexes: Full table scans on frequently queried columns

causing query times to grow linearly with data volume

- Monitor slow query logs and set appropriate thresholds (e.g.

50ms for OLTP queries)

- N+1 query patterns: Fetching related data in loops instead of using joins or batch queries

multiplying database round-trips linearly

Education

- Use EXPLAIN ANALYZE to inspect query execution plans and identify sequential scans

hash joins

- Implement distributed tracing (Jaeger

Zipkin) to identify cross-service latency bottlenecks

Database (Query Analyzers

Index Tuning)

- Run load tests with k6

JMeter

Coding & Debugging

- Use language-specific CPU and memory profilers (pprof for Go

py-spy for Python

Backend (APM

Profilers

- Deploy Application Performance Monitoring (Datadog

New Relic

Frontend (Chrome DevTools

Lighthouse

- Address scalability: proactively consider expected input size

memory limitations

- Provide alternative solutions when trade-offs exist (in-place vs additional memory

speed vs memory)

- Identify constraints

input/output requirements

- Remember that premature optimization is counterproductive

but strategic optimization is essential

- Capture percentiles (p50

p95

- Consider the full system impact of optimizations

not just local improvements

- [ ] All optimizations are documented with rationale

measured impact

- Use reproducible benchmarks with consistent hardware

data volume