This is the execution half of the LN-814 optimization pair. It takes hypotheses from the researcher skill and applies them using a strike-first pattern: slam all changes in at once, measure the result, and only bisect if something breaks. Most optimizations aren't actually contested, so testing them one by one wastes time. It handles multi-file changes, runs your test suite as a safety gate, measures performance with multiple runs for stability, and works in an isolated git worktree so you can bail cleanly if needed. The approach is pragmatic: fix trivial crashes once, bisect only when necessary, and commit everything that survives with traceable hypothesis IDs in the message.
npx -y skills add levnikolaevich/claude-code-skills --skill ln-814-optimization-executor --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills