Find agent skills by outcome

- Starts with measurement

not assumptions

- Considers full-system impact

not just local improvements

Before finalizing

verify:

Use checkboxes and stable IDs (e.g.

`PERF-ITEM-1.1`):

- Trade-offs: Complexity

maintainability

Use checkboxes and stable IDs (e.g.

`PERF-PLAN-1.1`):

- Baseline metrics: response time (p50

p95

In `TODO_perf-tuning.md`

include:

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

serving stale data and creating consistency bugs

Coding & Debugging

Write all proposed optimizations and any code snippets to `TODO_perf-tuning.md` only. Do not create any other files. If specific files should be created or edited

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

Coding & Debugging

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

destroying concurrency benefits

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

causing query times to grow linearly with data volume

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

multiplying database round-trips linearly

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

leading to OOM crashes in production

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

>50ms for OLTP queries)

### Database (Query Analyzers

Index Tuning)

- Implement distributed tracing (Jaeger

Zipkin) to identify cross-service latency bottlenecks

- Run load tests with k6

JMeter

Coding & Debugging

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

py-spy for Python

- Deploy Application Performance Monitoring (Datadog

New Relic

### Backend (APM

Profilers

### Frontend (Chrome DevTools

Lighthouse

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

speed vs memory)

- Address scalability: proactively consider expected input size

memory limitations