Use this skill when an operator wants a repeatable way to improve an existing agent skill rather than manually rewriting a SKILL.md from scratch. Darwin Skill frames the work as an optimization loop: evaluate the target skill, propose a bounded improvement, test the result, compare evidence, and keep or roll back the change.

Invoke this instead of editing the skill normally when the goal is to improve a reusable skill with measurable checks and review gates: raising the quality of a Claude Code skill, tightening a SKILL.md that multiple agents use, adding failure-mode guidance, or validating whether a proposed skill rewrite actually performs better.

The scope is one skill-optimization workflow over a single editable SKILL.md asset. It is not a generic prompt-writing tool, a broad skill marketplace entry, or a replacement for the host agent. The boundary is the Darwin evaluation and validation loop for improving agent skills with human approval before changes are retained.

Source-backed installation from the upstream README: npx skills add alchaincyf/darwin-skill

After installation, invoke it when working on a target SKILL.md and keep the run bounded to one skill at a time so the result can be attributed and reviewed.