What problem does it solve? Teams repeatedly make the same mistakes because corrections and successful patterns from conversations are never captured. This Skill analyzes conversations for learning signals and proposes structured updates to squad knowledge files so insights survive across sessions. ## Core Features & Use Cases - Confidence-Ranked Signal Detection: Classifies learnings as HIGH (user corrections), MEDIUM (praise, edge cases), or LOW (accumulated preferences) with explicit thresholds for when to propose changes. - Targeted Knowledge Routing: Routes learnings to the right destination: agent-specific history.md files, team-wide decisions inbox, or skill improvement recommendations. - Approval-Gated Persistence: Always shows proposed changes in a structured summary and waits for user approval before writing to any file. - Use Case: After a user says "No, use the code search tools first, grep is too slow", the Skill captures a HIGH-confidence constraint and appends it to the agent's history.md so the mistake is never repeated. ## Quick Start Ask the assistant to reflect on this session and capture any learnings from corrections or feedback into the squad knowledge files.
