What problem does it solve? Agent skills degrade over time as instructions become outdated or fail to address real failure modes observed in sessions. This Skill closes the loop by analyzing actual usage evidence and feedback to propose concrete improvements to a skill's SKILL.md file. ## Core Features & Use Cases - Evidence-gated analysis: Searches agent-mail for feedback and usage messages, checks bead references, and skips refinement entirely when no evidence exists, avoiding wasted agent runs. - Background analysis agent: Spawns a general-purpose agent that reviews the current SKILL.md against collected evidence and writes proposed changes to a cached proposal file. - Proposal caching and diff review: Reuses proposals modified within the last 7 days and presents a before/after diff for user approval before writing updates. - Use Case: After noticing a skill repeatedly misused in sessions, run the refinement workflow to gather feedback, generate targeted DO/DON'T additions, and apply vetted changes to the skill definition. ## Quick Start Ask the agent to refine a specific skill by name, for example by requesting to refine the skill named code-review using recent session feedback.
