llm-wiki-skill

Why we included this project

This project packages Andrej Karpathy's llm-wiki method into a skill you can install into Claude Code, Codex, OpenClaw, or Hermes. Instead of re-deriving answers from raw sources on every query, it compiles knowledge once into structured entity and topic pages connected by bidirectional links, then keeps that graph updated as you feed it new material from web pages, PDFs, local files, and social posts. The bundled offline HTML knowledge graph gives you a browsable, searchable map of your notes with confidence labels (extracted, inferred, ambiguous, unverified) so you can see at a glance what still needs checking. It also injects context automatically at the start of each session, so the agent is aware of your knowledge base without you asking. That makes it a practical choice for researchers, writers, and small teams who want a local, agent-driven knowledge base rather than a hosted SaaS wiki.