karpathy-llm-wiki-template

Karpathy LLM Wiki Template

A ready-to-use implementation of Andrej Karpathy's LLM Wiki pattern — where an LLM incrementally builds and maintains a persistent markdown wiki that compounds over time, instead of re-deriving knowledge from raw documents on every query (RAG).

What this is

You collect sources (articles, papers, notes). You point your LLM at them one at a time. The LLM reads each source, discusses what's worth keeping, then writes and cross-references wiki pages — summaries, entity profiles, concept explanations, synthesis pages. Over time the wiki becomes a compounding knowledge base with backlinks, citations, and structure that no human would maintain by hand.

The LLM is the bookkeeper. You are the thinker.

What's included

The template ships with two pre-ingested example sources (Karpathy's original gist + a community research synthesis) so you can see the pattern in action before adding your own content.

Architecture: thin harness, fat skills

CLAUDE.md stays thin. Each workflow lives in its own skills/ /SKILL.md file that Claude loads on demand. Skill files are plain markdown + YAML — no vendor-specific frontmatter — so they port to any AI agent that can read markdown and follow numbered steps.

The six skills

Session start is handled inline by CLAUDE.md §8 (read order + preflight checks) rather than as a separate skill — the checks are three greps and an ls, which didn't earn a dedicated SKILL.md.