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:

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.

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

destroying concurrency benefits

- 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)

- Implement distributed tracing (Jaeger

Zipkin) to identify cross-service latency bottlenecks

- Run load tests with k6

JMeter

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

py-spy for Python

### 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

- Identify constraints

input/output requirements

- Consider the full system impact of optimizations

not just local improvements

- Remember that premature optimization is counterproductive

but strategic optimization is essential

- Capture percentiles (p50

p95

- [ ] API response times meet SLA: <200ms (p95) for standard endpoints

<50ms (p95) for database queries