What problem does it solve? After a long Claude Code session, valuable lessons and skill improvement ideas are easily lost. This Skill closes out a session by extracting what actually happened, identifying which skills have gaps, and routing lessons to the right place before they are forgotten. ## Core Features & Use Cases - Session Forensics: Extracts user messages, assistant replies, timelines, or compact transcripts from multi-megabyte Claude Code session JSONL files via a Python script, writing results to disk instead of loading everything into context. - Skill Iteration Assessment: Maps each discovered problem or lesson to a specific skill defect, missing rule, or capability gap, then produces a prioritized optimization list with value, cost, and recommendation for user approval. - Cross-Repository Lesson Reflux: When the session operated on another project repository, writes empirically verified lessons into that repository's own AGENTS.md following its commit conventions. - Use Case: After finishing a debugging session in a client project, ask for a session summary to get a traceable recap, an update to that project's AGENTS.md with newly proven pitfalls, and a confirmed list of improvements to apply to the skills repository. ## Quick Start Ask the assistant to summarize this session and evaluate whether any skills should be improved based on what happened.
