What problem does it solve? When users correct AI behavior or a Skill execution fails, those signals are usually lost. This Skill captures them as structured, deduplicated feedback entries with quantitative scores so an evolution engine can later turn them into concrete improvements. ## Core Features & Use Cases - Signal Detection & Recording: Identifies five signal types (user corrections, uncovered scenarios, repetitive operations, quality issues, capability assessments) and writes them as topic files under a feedback directory. - Deduplication & Indexing: Checks FEEDBACK-INDEX.md before writing, merging recurring failure modes instead of creating duplicates. - Four-Dimension Scoring: Every entry carries Precision, Coverage, Efficiency, and Satisfaction scores (1-5) with anti-inflation rules, plus a failure_class field (skill-defect, execution-lapse, unset) for routing. - Use Case: A user says "the Skill never mentioned X" after a failed run. The feedback-observer dispatches this Skill, which creates a scored feedback entry with a RED observation line, updates the index, and returns a confirmation to the main agent. ## Quick Start Ask the agent to record feedback about the correction you just made so it can be scored and indexed for future improvement.
