What problem does it solve? It turns scattered feedback entries and execution traces into structured, data-backed improvement proposals for an agent's rules and skills, preventing premature or unsupported changes. ## Core Features & Use Cases - Signal Detection: Scans feedback files and forge-loop step traces to detect rule graduation candidates (3+ occurrences), skill optimization signals (low repeated scores), and new skill candidates (5+ uncovered patterns). - TDD-Style Proposals: Every proposal includes a RED observation, GREEN change, predicted effect, and a falsifiable verify-by criterion, with failure-class routing (skill-defect vs execution-lapse). - Quality Gates: Applies an 8-dimension scoring rubric, anti-rationalization checklist, and denominator awareness to filter out noise before presenting proposals. - Use Case: After several sessions accumulate feedback about a recurring agent mistake, run this skill to receive a structured proposal specifying the exact SKILL.md section to change and how to verify the fix. ## Quick Start Ask the agent to check if there are any rules to upgrade based on the accumulated feedback.
