Performance Debugging

Find and verify the cause of an observed performance regression in a known application or service. This skill uses hypotheses, representative workloads, runtime profiles, and before/after measurements; it does not turn a benchmark into a production load test or treat a flame graph as proof without a reproduced symptom.

When to use

• Investigating known application CPU, heap growth, allocation, lock contention, GC, throughput, or latency regressions
• Profiling a service, worker, CLI, or library after a reproducible slow path is identified
• Comparing a change against a stable baseline and checking for performance regressions
• Turning a production observation into a safe local or staging reproduction

When NOT to use