Back to Cookbook
KiloClaw

Profiling Playbook

Make profiling routine, not just an emergency tool

Standardize how teams capture and interpret profiles (CPU, memory, IO) and convert results into actionable fixes with verification steps.

CommunitySubmitted by CommunityWork12 min

INGREDIENTS

🔎Web Search

PROMPT

Create a skill called "Profiling Playbook". Inputs: - The service/component and the symptom (CPU, latency, memory) - Environment where profiling is allowed Output: - A profiling capture checklist (safe and repeatable) - Interpretation guide and common pitfalls - A verification plan for the fix (benchmarks/telemetry)

How It Works

Profiling is often delayed until crisis. This recipe makes profiling repeatable: capture,

interpret, act, and verify.

Triggers

  • High CPU or latency incidents
  • "We optimized but it didn't help"
  • Performance regressions after changes

Steps

  1. Define "when to profile" thresholds (p95 latency, CPU saturation).
  2. Capture profiles in a safe environment (staging or controlled prod sampling).
  3. Interpret results:
  • identify hot spots,
  • validate assumptions,
  • propose a minimal change.
  1. Implement fix with measurable before/after evidence.
  2. Add a guardrail test or ongoing measurement.

Expected Outcome

  • Faster performance debugging.
  • Less wasted time on intuition-only optimization.

Example Inputs

  • "Capture CPU profile for API under load."
  • "Find the hot path causing p95 latency spikes."
  • "Memory usage skyrocketed after release."

Tips

  • Every optimization needs a measurement story: before, after, and conditions.
Tags:#performance#debugging#observability#release-management