gf-evolve

You need RED evidence before applying changes to improve a skill from failures. By requiring RED evidence first, the evolution process enforces validation and tracks progress through the lifecycle, ensuring that patches close actual gaps rather than masking symptoms.

The best way to structure skill evolution patches from user feedback is to apply minimal patch cycles. Observe a failure, collect RED evidence, and implement a minimal evolution patch that updates the relevant skill and its validation artifacts, then revalidate to confirm the fix.

You need to update validation and packaging docs for a skill when applying an evolution patch to prevent drift. Refreshing CHANGELOG, VALIDATION, and package docs ensures that the documentation accurately reflects the applied changes and maintains traceability.

No, this skill evolution process requires no external dependencies to apply patches. It operates by adjusting pressure scenarios and updating RED evidence internally, allowing you to guide minimal patch cycles and revalidate without needing additional tools.

Skill drift occurs when validation gaps and repeated failures remain unaddressed during updates. You can prevent it by using RED-gate driven validation to require RED evidence before applying changes, and by refreshing documentation to ensure patches are traceable.