What problem does it solve? After a coding session, it is hard to know whether friction came from the task itself or from a misconfigured agent environment. This Skill systematically reviews a session's raw transcript to find improvement candidates for guidance files, automated checks, tooling, and information access. ## Core Features & Use Cases - Session Transcript Analysis: Reads the full JSONL session record, examining every tool call and its result rather than relying on conversation summaries. - Five-Category Audit: Checks navigation friction, missing automated checks, guidance bloat in AGENTS.md and skills, tool-call economics, and information access gaps, each backed by session evidence. - Prioritized Recommendations: Presents candidates ranked by severity with suggested changes and costs, waiting for user selection before editing anything. - Use Case: After an agent spent many rounds hunting for a file, run a retrospective to identify the missing AGENTS.md context pointer and propose the exact documentation fix. ## Quick Start Ask the agent to run a retrospective on the current coding session and propose improvements to the agent environment.