What problem does it solve? After a development session, project docs, CLAUDE.md/AGENTS.md files, and agent memory drift out of sync with the actual code, and workspace conventions quietly stop being followed. This Skill performs an end-of-session cleanup that reconciles all knowledge layers against the code and audits whether the workspace's own rules (naming, required files, symlink integrity, dead references) are being enforced. ## Core Features & Use Cases - Three-layer knowledge reconciliation: Aligns agent memory, project-root CLAUDE.md/AGENTS.md, and docs/ with the current code, including cross-project downstream documentation. - Anti-bloat governance: Enforces size budgets (e.g., MEMORY.md under 25KB/200 lines), promotes stable memories into docs via a graduation mechanism, and strips historical narrative out of rule files. - Rule execution audit: Extracts mechanically verifiable conventions from hierarchical rule files, checks practice against them, auto-fixes safe violations, and escalates destructive ones for human decision. - Use Case: After shipping a new API with new environment variables, say "sync up" and the Skill updates the route list in CLAUDE.md, the integration guide, architecture docs, and runbook, prunes stale memories, and reports any naming or symlink violations it found. ## Quick Start Tell the agent "sync up the docs and memory after this session" and it will inventory, reconcile, and audit the project knowledge base.
