Perf Profiler — Persona

The problem

Engineering teams often waste weeks implementing clever caching layers or micro-optimizations based on gut feeling rather than data. Without empirical baselines, these changes frequently add technical debt or, worse, make the system slower in production.

What it does

• Enforces a strict measurement-first protocol by demanding baseline metrics before discussing any code changes.
• Analyzes flame graphs, profile logs, and benchmark outputs to identify bottlenecks affecting end-user experience.
• Ranks optimization candidates based on user-felt impact versus implementation costs like memory and readability.
• Validates benchmarking methodology to ensure results aren't just statistical noise or environment artifacts.
• Proposes specific, data-backed hypotheses for performance improvements without modifying code prematurely.

Why this beats prompting it yourself

General-purpose LLMs often suggest "best practice" optimizations that don't apply to your specific bottleneck. This skill acts as a skeptical performance engineer that rejects vague claims of speed and forces you to maintain rigorous benchmarking standards.

Use cases

• Analyzing CPU profiles to find the actual hot path in a service.
• Evaluating whether a proposed architectural change justifies its complexity cost.
• Debugging inconsistent benchmark results and identifying sources of variance.
• Auditing legacy codebases for "premature optimizations" that are no longer effective.

Known limitations

Does not assist with infrastructure procurement, vendor selection, or load testing against third-party systems you do not own.