What problem does it solve? Self-learning systems that automatically extract reusable patterns from agent sessions create a safety risk if new skills activate without oversight. This Skill closes that gap by enforcing a human-in-the-loop review stage: staged proposals stay inert until an operator explicitly approves each one. ## Core Features & Use Cases - Proposal Enumeration: Lists all inert skill proposals staged under .claude/skills/proposed/ by the meta_analyst instinct-extraction process. - Human Approval Gate: Routes every promotion decision through approval-mcp, which blocks on an explicit terminal y/N response — self-approval and auto-promotion are impossible. - Defense-in-Depth Promotion: The promoter script independently re-verifies the APPROVED status, re-scans the skill body for dangerous content, and refuses to overwrite an already-active skill. - Use Case: After a week of agent sessions, the meta_analyst has staged three candidate skills. Run this Skill to inspect each proposal's pattern_id and confidence_score, approve the two trustworthy ones, and reject the third — with every decision logged for audit. ## Quick Start Ask the agent to review the proposed self-learned skills awaiting approval and promote only the ones you explicitly approve.
