What problem does it solve? After a task ends, recurring friction—rediscovered facts, misfired skills, repeated user corrections—goes unrecorded, so the next session repeats the same mistakes. This Skill turns each finished session into a concrete, reversible improvement to the harness (skills, reference notes, memory, kernel instructions). ## Core Features & Use Cases - Trajectory mining: Re-reads actual session events through judgment, tooling, and divergent lenses to classify friction as rediscovery, skill misfire, correction, or stale state. - Smallest-edit discipline: Maps each friction to one minimal edit and one commit, quoting trajectory evidence, with structural enforcement (lints, hooks, scripts, permissions) preferred over prose rules. - Skill description repair: Treats misfired skills as description bugs and verifies fixes by testing routing with a fresh subagent. - Use Case: After a session where the user corrected your git workflow twice, invoke this Skill to locate the friction, add the missing rule to the correct instruction file, and commit it with quoted evidence. ## Quick Start Say "wrap up and refine this session" to mine the trajectory and apply the smallest evidence-backed harness edits.
