karpathy-wiki
Wiki System
"The tedious part of maintaining a knowledge base is not the reading or thinking — it's the bookkeeping." — Andrej Karpathy
Most AI memory systems retrieve and forget. You ask a question, the agent fetches context, answers, and the insight vanishes into chat history. Next session, same question, same fetch, same synthesis from scratch.
This skill implements Karpathy's compilation-over-retrieval pattern (original gist): instead of fetching raw sources every time, the agent builds and maintains a structured wiki — cross-referenced markdown pages that compound knowledge over time. The human curates what goes in; the agent handles the summarization, cross-referencing, and maintenance that humans typically abandon because the overhead grows unbearably.
Three Layers
- Raw Sources (immutable) — PDFs, articles, papers, URLs. You drop them in, the agent reads but never modifies them.
- The Wiki (LLM-maintained) — structured markdown pages in
memory/wiki/with frontmatter, claims, cross-references. The agent creates, updates, and maintains these automatically. - The Schema (
WIKI-SCHEMA.md) — rules for structure, naming, and workflows. Co-evolves with your knowledge base.