Phase 1 — Benchmarking Tools
Cross-platform benchmark runners for Python, Node.js, PowerShell, and Bash.
A CLAUDE.md template that turns Claude Code into a strict performance engineering expert. Drop it in your project root. No install, no plugin, no config.
$ cp performance-deity/CLAUDE.md ./CLAUDE.md
✔ Done. Open Claude Code in your project.
user: optimize this function
→ Running baseline benchmark... P95: 450ms
→ Identified O(n²) nested loop. Rewriting to HashMap lookup...
→ Benchmarking optimized code... P95: 12ms
✔ Baseline: 450ms → Optimized: 12ms (-97.3%)
Claude Code reads CLAUDE.md automatically at session start. The file contains strict, phased instructions that override Claude's default behavior for performance-related tasks.
cp CLAUDE.md /your/project/CLAUDE.md
It reads CLAUDE.md on every session start. No flags, no setup.
Say "optimize this", "memory leak", or "slow query" — the right skill activates automatically.
Each skill is a phased methodology with mandatory benchmarking, proof requirements, and prohibited patterns.
Benchmarks baseline, analyzes Big-O, rewrites the algorithm, and proves improvement with a before/after table. Will not present a change without numbers.
Instruments memory with tracemalloc or process.memoryUsage(), proves the leak grows across iterations, then fixes the uncollected reference.
Fires 1,000–10,000 concurrent requests at the target function, proves it fails, then applies the minimal synchronization primitive needed.
Runs EXPLAIN ANALYZE, identifies missing indexes, and rewrites ORM loops into batch joins or eager loads with exact CREATE INDEX statements.
Bombards code with 1GB payloads, null values, malformed JSON, and simulated network failures. Documents every crash, then hardens the boundary.
Profiles payload byte size, enforces Brotli compression and HTTP cache headers, and migrates heavy REST endpoints to GraphQL or protobuf when over 1MB.
Fixes Docker layer ordering, enables Webpack filesystem caching, and adds GitHub Actions dependency caching keyed on lockfile hashes.
Wraps critical paths in OpenTelemetry or Sentry spans with execution time, query tags, and payload sizes. Always includes a concrete Datadog or Prometheus alert rule.
Audits React re-renders, migrates animations to GPU-composited transform, and virtualizes lists over 50 items. Target: 60 FPS.
cp CLAUDE.md /your/project/CLAUDE.md
Claude Code reads it automatically.
cp CLAUDE.md /your/project/.cursorrules
Rename to .cursorrules for Cursor.
cp hooks/pre-commit .git/hooks/pre-commit
Blocks commits with unresolved TODO: optimize markers.
Where this project is headed.
Cross-platform benchmark runners for Python, Node.js, PowerShell, and Bash.
All nine skills compiled into a single, deployable CLAUDE.md with global rules and prohibited patterns.
A reusable workflow that runs the benchmark suite on pull requests and fails the build if P95 regresses beyond a configurable threshold.
Adapted rule files for .cursorrules and Windsurf's memory system, maintained in sync with the canonical CLAUDE.md.