Browse / LLM wiki / llm-wiki-kit

llm-wiki-kit

dotBet

This repository is a Markdown-first implementation of Andrej Karpathy's LLM Wiki pattern, using an OKF v0.2-compatible metadata model: raw sources stay immutable, the LLM

Open dotBet/llm-wiki-kit → Checked against the four rules.

Works with any agent

ships AGENTS.md, the cross-tool instruction file.

In its own words

README.mddotBet/llm-wiki-kit

This repository is a Markdown-first implementation of Andrej Karpathy's LLM Wiki pattern, using an OKF v0.2-compatible metadata model: raw sources stay immutable, the LLM maintains a compact interlinked wiki, and an operating contract tells agents how to ingest, query, and lint the knowledge base.

The goal is compounding knowledge. Each useful source, answer, correction, and synthesis should improve the wiki instead of disappearing into chat history.

OKF Compatibility

The wiki/ directory is an OKF-compatible knowledge bundle. The root wiki index declares okf_version: "0.2"; maintained pages use YAML frontmatter with a non-empty type, descriptive metadata, structured sources, and OKF-compatible lifecycle values (draft, stable, or deprecated).

This repository deliberately adds stricter local conventions: wiki/index.md and wiki/log.md are required, internal dead links are lint findings, and claim provenance follows the balanced citation policy.

Core Loop

  • Add raw material to sources/raw/ and a source note to sources/notes/.
  • Ask the LLM to ingest it using AGENTS.md or the LLM Wiki Ingest skill.
  • The LLM updates pages under wiki/, preserving citations back to sources.
  • Ask questions using the LLM Wiki Query skill, starting from wiki/index.md, then deeper pages, then raw sources only when needed.
  • File reusable answers back into the wiki.
  • Run the LLM Wiki Lint skill before trusting the wiki for important work.

Quoting 53% of the README. Read the rest on GitHub ↗

Read from the repository's git tree on 2026-08-17.