criterium

Run `npx skills add hugoduncan/criterium --all -g -y` in your terminal to install this skill globally for your AI agent.

Use the criterium.bench/bench macro, which automatically handles JVM warmup, GC isolation, and statistical analysis to produce trustworthy timing results.

Naive timing ignores JIT warmup, garbage collection, and dead code elimination. Criterium corrects for all of these with warmup phases and bootstrap confidence intervals.

Yes. Domain analysis lets you benchmark multiple implementations across input sizes and even fit complexity models like O(n) or O(n log n).

Yes. The included SKILL.md teaches agents like Claude Code how to run benchmarks, choose bench plans, and interpret results from plain-English requests.