What problem does it solve? Hard-won insights from complex tasks are usually lost when a session ends, and ad-hoc edits to shared skill files risk drift, broken symlinks, and inconsistent behavior across AI tools. This Skill provides a disciplined funnel for turning repeated patterns into reviewed, validated skill changes. ## Core Features & Use Cases - Capture-Validate-Codify funnel: Stage learnings as memory notes first, then promote only patterns that recur, were requested, or fix a real failure. - Two codification paths: Update an existing skill (Path A) or create a new one (Path B), each with branching, diff review, eval gates, and repo validation steps. - Tool drift detection: Identify when MCP tool-adapter skills document stale tool names or examples and route the fix through a PR. - Use Case: After noticing a GitHub MCP tool was renamed mid-task, finish the work, then open a [skill-evolution] PR updating the affected skill with the new tool name and passing validate.sh and agnix checks. ## Quick Start Ask the agent to save what was learned in this session as a reusable skill update in the ai-shared repo.
