What problem does it solve? Existing skills can contain wrong instructions, outdated steps, or missing edge cases, and fixing them without discipline leads to bloated rewrites or unjustified changes. This Skill enforces an evidence-driven, minimal-patch workflow so every skill edit traces back to a concrete, verified problem. ## Core Features & Use Cases - Evidence collection and attribution: Gathers concrete failure traces or user feedback, then classifies whether the fault lies in the skill text, the agent's behavior, or a changed environment before editing anything. - Minimal patch generation: Produces surgical old_string/new_string edits with documented problem, evidence, and rationale instead of full rewrites. - Mutable vs. immutable handling: Edits user and agent skills directly, while routing built-in or project skill changes through a worktree and merge request. - Use Case: A user reports that a deployment skill references a renamed CLI command. The Skill verifies the evidence, patches only the outdated command line, and re-reads the file to confirm the frontmatter and overall coherence survived. ## Quick Start Ask the agent to refine the skill named deployment-helper because its instructions reference a command that no longer exists.
