What problem does it solve? Feedback given during a coding session is usually lost once the session ends, forcing users to repeat the same corrections. This Skill captures durable preferences from corrections and approvals and codifies them into the Skill files that were active, so the agent improves across sessions. ## Core Features & Use Cases - Signal Detection: Scans sessions for corrections, repeated patterns, and approvals while filtering out one-off or ambiguous feedback. - Quality Filtering: Applies a four-question test to ensure only new, actionable, project-specific information is captured. - Reviewed Proposals: Maps signals to the relevant SKILL.md files and presents HIGH/MEDIUM confidence edits for explicit approval before applying. - Use Case: After a session where you corrected the agent twice to use cn() instead of clsx(), ask it to learn from the session and it will propose a minimal, reversible edit to the appropriate Skill file. ## Quick Start Ask the agent to learn from this session and update the skills based on the corrections you gave.