core

Move dirty changes to an auto-named branch, run save-learnings + docs-sync drain + AI review chain (generic baseline + team specialists), open PR. Optional --auto-merge enables GitHub auto-merge with polling + auto-fix loop. End-of-work returns the user to the target branch.

Drain marker-driven docs queue, run drift-detector against the changed scope, regenerate TR mirrors via parity-checker, present accept/reject drafts. Boring sessions free (pre-flight short-circuit). Invoked from /create-pr Step 4.5 OR manually for catch-up audits. Composite design from the docs-sync-automation brainstorm (Q1+Q3+Q4+Q5+Q7).

Generate a full code-level Mermaid class diagram of the entire project. Shows all classes, interfaces, relationships, dependencies, and inheritance. Output saved to .atl/docs/code-diagram.md

Persist conversation learnings to journal + wiki + agent children + skill learnings, with auto-rebuild of agent.md / skill.md Knowledge Base sections from frontmatter. Multi-transcript input via --transcripts. Updates the learning-capture state file at completion (closes the auto-trigger loop). AskUserQuestion gates only for new structure creation: new agent, new skill, new rule, agent identity change, skill core change. Everything else is automatic.

Project knowledge base — living, cross-referenced, always current. Initialize, ingest new knowledge, query existing, lint for staleness. Powered by Karpathy's LLM Wiki pattern.