Yuhan Lei
637eccaa15
- Translate 2026-03-19-claude-code-statusline-landscape.md (148 lines) - Fix mixed Chinese/English in claude-code-statusline-landscape.md (Impact section) - All example files now uniformly in English
karpathy-llm-wiki
Battle-tested LLM Wiki skill for Claude Code, Cursor & Codex
Battle-Tested Results
Key Takeaways
- ✅ Production-ready: 94 articles, 99 sources, daily use since April 2026
- ✅ Works with Claude Code, Cursor, Codex CLI, OpenCode
- ✅ Agent Skills standard (SKILL.md) compatible
- ✅ Three operations: Ingest → compile sources into wiki, Query → search with citations, Lint → health checks
This skill powers a production knowledge base that's been in daily use since April 2026:
| Metric | Count | Growth |
|---|---|---|
| Wiki articles | 94 | 48 → 94 (nearly doubled) |
| Source materials | 99 | From zero |
| Topic directories | 13 | 8 → 13 |
| Operation log entries | 87 | Every day |
| Recent 7 days activity | 87 entries | Active maintenance |
Top topics: ai-coding-tools (29 articles), ai-research (24), product-design-frameworks (8).
See examples/ for real articles, raw sources, and the operation log. See SKILL.md for the full skill specification.
Quick Install
npx add-skill Astro-Han/karpathy-llm-wiki
Works with Claude Code, Cursor, Codex, and other tools that support the Agent Skills standard. See docs/SPEC.md for detailed architecture and design decisions.
Quick Start
1. Ingest your first source
Give the skill a URL, a file, or paste text directly:
"Ingest this article: https://example.com/attention-is-all-you-need"
The skill fetches the content into raw/, then compiles it into a wiki article under wiki/.
2. Ask your wiki a question
"What do I know about attention mechanisms?"
The skill searches your wiki and answers with citations linking back to your articles.
3. Keep it healthy
"Lint my wiki"
The skill checks for broken links, missing index entries, stale cross-references, and reports potential issues.
How Your LLM Wiki Works
What is this skill?
This skill is a SKILL.md-compatible implementation of Karpathy's knowledge compilation pattern. It turns raw documents into a structured, interlinked wiki that compounds over time.
Inspired by Karpathy's LLM Wiki idea (published April 4, 2026, 2,100+ stars in 12 hours):
"The LLM writes and maintains the wiki; the human reads and asks questions."
What is the Agent Skills standard?
Agent Skills is an open standard for defining reusable skills via SKILL.md files. It's supported by Claude Code, Cursor, Codex CLI, OpenCode, and other LLM coding tools. The standard enables skill discovery, installation via npx add-skill, and cross-tool compatibility.
Architecture
your-project/
├── raw/ ← Immutable source material (you or the LLM add, never modify)
│ └── topic/
│ └── 2026-04-03-source-article.md
├── wiki/ ← Compiled knowledge (LLM maintains)
│ ├── topic/
│ │ └── concept-name.md
│ ├── index.md ← One-page table of contents
│ └── log.md ← Append-only operation log
Three operations:
| Operation | What it does |
|---|---|
| Ingest | Fetch a source into raw/, compile into wiki/, update index and cross-references |
| Query | Search the wiki and answer with citations. Optionally archive answers as wiki pages |
| Lint | Auto-fix broken links and index gaps. Report contradictions, orphan pages, stale content |
The wiki compounds over time. Each new source enriches existing articles, adds cross-references, and flags conflicts.
Tool Compatibility
This skill follows the agentskills.io open standard. It works with any LLM coding tool that supports SKILL.md:
| Tool | Install method |
|---|---|
| Claude Code | npx add-skill Astro-Han/karpathy-llm-wiki |
| Cursor | npx add-skill Astro-Han/karpathy-llm-wiki (auto-converts) |
| Codex CLI | Copy to .agents/skills/karpathy-llm-wiki/ |
| OpenCode | npx add-skill Astro-Han/karpathy-llm-wiki |
| Others | Copy SKILL.md + references/ to your tool's skill directory |
Alternative implementations: See lucasastorian/llmwiki (MCP-based web app) and atomicmemory/llm-wiki-compiler (CLI compiler) for other approaches to Karpathy's LLM Wiki pattern.
Inspired By
This is an unofficial community implementation of the knowledge compilation workflow described in Karpathy's idea file (April 2026). The core value is a battle-tested set of compilation principles and workflow templates, not the code itself.
FAQ
What is an LLM Wiki?
An LLM Wiki is a knowledge management system where a large language model automatically writes, maintains, and queries a structured wiki from your raw source materials. Unlike RAG which retrieves at query time, an LLM Wiki compiles knowledge once into interlinked pages that compound over time.
Which tools are compatible?
Claude Code, Cursor, Codex CLI, OpenCode, and any tool that supports the Agent Skills (SKILL.md) standard. Install with npx add-skill Astro-Han/karpathy-llm-wiki.
How does this differ from RAG?
RAG searches raw document chunks at each query, re-deriving relationships every time. LLM Wiki pre-compiles documents into structured pages with cross-references, entity pages, and an index. Each new source enriches existing articles instead of being searched independently. Knowledge compounds, not repeats.
What sources can I ingest?
URLs (web articles, papers, blog posts), local files (PDFs, markdown, text), or pasted text. The skill converts everything to markdown in raw/ and compiles into wiki/.
Is this production-ready?
Yes. This skill powers a real knowledge base with 94 articles and 99 sources, maintained daily since April 2026. See examples/ for actual wiki pages and operation logs.
How do I start?
npx add-skill Astro-Han/karpathy-llm-wiki
Then tell your LLM: "Ingest this article: [URL]"