autorefine

AutoRefine improves a SKILL.md by automating eval-grounded autoresearch, combining a design-audit with Hamel's Three Gulfs and a mutation loop to validate changes through judge validation and domain tests.

The mutation loop process for skill improvement applies phase-based mutations to a SKILL.md, routing downstream work through pattern-aware strategies and contract integrity checks to ensure verifiable, eval-grounded gains before session-close cleanup.

A design-audit using Hamel's Three Gulfs evaluates skill workflows by surfacing misalignments between intent and execution, providing structured preflight checks before entering the mutation loop to refine the SKILL.md.

Yes, you can apply automated evaluation and mutation to any skill defined by a SKILL.md, spanning automation to agent workflows, using adapter-domain evaluation and ambient learning to verify improvements across diverse domains.

The best way to validate SKILL.md improvements with domain tests is running a disciplined mutation loop that uses judge validation and domain-metric integration, ensuring changes are verifiable and backed by eval-grounded evidence.

Your skill workflow needs pattern-aware routing and contract integrity checks to maintain verifiable improvements during mutation, ensuring adapter-domain evaluation aligns with ambient learning and preventing structural breakdowns across the SKILL.md.