Итеративная оптимизация метрик кода

★ 7.5 · engineering

ralph is a Claude Code skill that runs an iterative, measurement-driven improvement loop following the cycle: measure → profile → mutate → re-measure → commit. It works for performance, bundle size, test coverage, complexity, or any quantifiable metric. Each cycle breaks the target metric into components, identifies the biggest contributor, applies a single targeted change, and commits only when the improvement exceeds noise (3% threshold for timing). The skill opens a feature branch with a draft PR early, maintains a `docs/<target>-ralph-report.md` with a running optimization log, and documents dead ends as explicit findings. By default it runs 20 cycles and halts after three consecutive no-improvement rounds — built for developers who need a reproducible, evidence-based process to reduce load times, memory usage, or other measurable characteristics.