performance-profiling

name	performance-profiling
description	Performance profiling principles. Measurement, analysis, and optimization techniques.
allowed-tools	Read, Glob, Grep, Bash

Performance Profiling

Measure, analyze, optimize.

Core Web Vitals

• LCP (Load): < 2.5s
• INP (Interactive): < 200ms
• CLS (Stability): < 0.1

Profiling Workflow

1. BASELINE: Measure first.
2. IDENTIFY: Find bottleneck (Lighthouse, Bundle Analyzer, DevTools).
3. FIX: Targeted change.
4. VALIDATE: Confirm improvement.

Analysis

• Bundle: Split large deps, dedupe, tree shake.
• Runtime: Long tasks (>50ms) block UI.
• Memory: Growing heap = leak.

Quick Wins

1. Compression (Gzip/Brotli).
2. Lazy Load Images.
3. Code Split Routes.
4. Cache Static Assets.

---

🧠 Aletheia Reasoning Protocol (Optimization)

1. Generator (Hypothesis)

• Suspect: "DB query slow?".
• Tool: "explain analyze".
• Target: "Reduce p99 by 50ms".

2. Verifier (Measurement)

• Baseline: "Record metric BEFORE fix".
• Isolation: "Network or CPU?".
• Significance: "Is 1ms worth complexity?".

3. Reviser (Stabilization)

• Regression: "Add test to prevent regression".
• Cache: "Can't fix? Cache it.".

---

🛡️ Security & Safety Protocol (Profiling)

1. Timing Attacks: No optimizations on auth checks.
2. Info Leak: Don't publish profiles (PII/Env vars).
3. Prod Impact: Don't crash prod with profiler overhead.
4. Sampling: Use sampling on high traffic.