Benchmark Optimization Loop

When someone asks you to "make it 20x faster" or run recursive optimizations, this skill turns that into a disciplined benchmark loop with a correctness gate, a measurable metric, and a search budget. It forces you to measure a baseline, test one hypothesis per variant, reject anything that breaks correctness, and promote only the fastest safe option. You track everything in a variant table and codify the winner before moving on. The approach is skeptical of vague speedup claims and insists on repeated measurements, rollback plans, and exact commands in the final summary. It stops you from chasing noise or changing too many variables at once.

Install to Claude Code

npx -y skills add affaan-m/ecc --skill benchmark-optimization-loop --agent claude-code