- brain/graph/learn-loop.sh: rebuild+cluster -> (LLM-tier) deep-extract corpus
-> merge -> measure & heuristically optimize (orphan rate, community balance,
size) -> record metrics.json + learnings.md -> sync to Windows
- brain/corpus/{diagrams,designs,notes}/: drop sources for the brain to learn
- cron agentic-os-brain-rebuild now runs the learn-loop (compounding)
- README documents the learning tiers (offline structural vs LLM semantic)
Verified: 614 nodes, 1024 links, 0.2% orphans after optimization; no node loss.
- AGENTS.md single source of truth with 51+10 features
- FastAPI backend with 20+ REST endpoints
- 15 skills with self-improvement (eval + learnings)
- Web SPA dashboard with 12 pages
- Agent configs for opencode, Hermes, Gemini CLI
- APScheduler, plugin registry, standards system
- 10 prompt templates, cost analytics, backup/restore