What problem does it solve? After a long working session, lessons learned — decisions made, friction encountered, workarounds applied — are usually lost in the chat scrollback instead of being captured in the project's persistent documentation. This Skill turns those observations into concrete, reviewable proposals for updating CLAUDE.md, SOP.md, and PROJECT-CONTEXT.md. ## Core Features & Use Cases - Ledger-grounded scan: Reads the retro ledger via retro-report.js for the current month to ground the reflection in recorded evidence rather than memory. - Structured proposal output: Produces a categorized report covering what worked, what hurt, proposed kit changes, ADR backlog items, and tech-debt observations. - Propose-don't-commit workflow: Never edits files directly; every suggestion requires explicit user approval before being applied. - Use Case: After a multi-hour development session where you bypassed a rule twice and made an undocumented architectural decision, run the reflection to get a list of SOP.md rule tightenings, a PROJECT-CONTEXT.md gotcha entry, and an ADR to backfill. ## Quick Start Ask the AI to reflect on this session and propose updates to CLAUDE.md, SOP.md, and PROJECT-CONTEXT.md based on what happened.