{
  "markdown": "# karpathy-llm-wiki\n\n**A reusable skill for building Karpathy-style LLM wikis with Claude Code, Cursor, Codex, and other Agent Skills tools.**\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)\n[![GitHub stars](https://img.shields.io/github/stars/Astro-Han/karpathy-llm-wiki?style=social)](https://github.com/Astro-Han/karpathy-llm-wiki)\n[![GitHub forks](https://img.shields.io/github/forks/Astro-Han/karpathy-llm-wiki?style=social)](https://github.com/Astro-Han/karpathy-llm-wiki)\n[![Agent Skills](https://img.shields.io/badge/Agent_Skills-compatible-blue)](https://agentskills.io)\n[![Install](https://img.shields.io/badge/Install-npx_add--skill-green)](https://github.com/Astro-Han/karpathy-llm-wiki#install)\n\n<p align=\"center\">\n  <img src=\"assets/karpathy-tweet.png\" alt=\"Karpathy's tweet about LLM Wiki\" width=\"560\">\n</p>\n\n`karpathy-llm-wiki` packages [Karpathy's LLM Wiki idea](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f) into one installable [Agent Skills](https://agentskills.io) skill. Your coding agent ingests sources into `raw/`, compiles durable knowledge pages into `wiki/`, answers questions with citations, and lints the wiki for consistency.\n\n## What Is an LLM Wiki?\n\nAn **LLM wiki** is a knowledge system where the LLM maintains structured wiki pages instead of re-searching raw documents on every question. New sources are compiled into durable markdown pages, cross-references are updated over time, and answers cite the wiki pages that already contain the synthesized knowledge.\n\nThis skill gives you three operations:\n\n| Operation | What it does | Output |\n|-----------|--------------|--------|\n| **Ingest** | Collects a source into `raw/`, triages it, then creates or updates wiki articles — or just logs it when nothing is new | New or updated wiki pages |\n| **Query** | Searches the wiki and answers with citations | Grounded answers linking to markdown pages |\n| **Lint** | Checks index integrity, links, and wiki health | Auto-fixes plus reported issues |\n\nSee [SKILL.md](SKILL.md) for the full skill specification.\n\n## LLM Wiki vs RAG\n\n| Approach | Knowledge lives in | When synthesis happens | Good for |\n|----------|--------------------|------------------------|----------|\n| **RAG** | Raw chunks and embeddings | At query time | Broad retrieval across large corpora |\n| **LLM Wiki** | Curated markdown pages | During ingest and maintenance | Compounding knowledge, summaries, and durable cross-links |\n\nThis skill is optimized for the wiki model: knowledge that improves over time instead of re-deriving relationships on every query.\n\n## Usage Stats\n\nBased on a production knowledge base maintained daily since April 2026:\n\n- **94** wiki articles across **13** topic directories\n- **99** source materials ingested\n- **87** operation log entries in the last 7 days\n\nSee [examples/](examples/) for sample wiki pages, source files, and operation logs.\n\n## Install\n\n```bash\nnpx add-skill Astro-Han/karpathy-llm-wiki\n```\n\nWorks with any tool that supports the [Agent Skills](https://agentskills.io) standard.\n\n## Quick Start\n\n### 1. Ingest your first source\n\nGive the skill a URL, a file, or pasted text:\n\n> \"Ingest this article: https://example.com/attention-is-all-you-need\"\n\nThe skill stores the source in `raw/`, then compiles or updates the right knowledge pages in `wiki/`.\n\n### 2. Ask your wiki a question\n\n> \"What do I know about attention mechanisms?\"\n\nThe skill searches the wiki and answers with citations linking back to your markdown pages.\n\n### 3. Keep the wiki healthy\n\n> \"Lint my wiki\"\n\nChecks for broken links, missing index entries, stale cross-references, and related issues.\n\n## How the Workflow Works\n\nThe core idea from Karpathy: the LLM maintains the wiki while the human focuses on choosing sources and asking good questions.\n\n```text\nyour-project/\n├── raw/            ← Immutable source material\n│   └── topic/\n│       └── 2026-04-03-source-article.md\n├── wiki/           ← Compiled knowledge pages maintained by the LLM\n│   ├── topic/\n│   │   └── concept-name.md\n│   ├── index.md    ← Global table of contents\n│   └── log.md      ← Append-only operation log\n```\n\nEach new source can update multiple pages, strengthen cross-references, and record contradictions. That is what makes the wiki compound over time.\n\n## Tool Compatibility\n\nThis skill follows the [agentskills.io](https://agentskills.io) open standard:\n\n| Tool | Install method |\n|------|----------------|\n| Claude Code | `npx add-skill Astro-Han/karpathy-llm-wiki` |\n| Cursor | `npx add-skill Astro-Han/karpathy-llm-wiki` |\n| Codex CLI | Copy to `.agents/skills/karpathy-llm-wiki/` |\n| OpenCode | `npx add-skill Astro-Han/karpathy-llm-wiki` |\n| Other tools | Copy `SKILL.md`, `references/`, and `scripts/` into the tool's skill directory |\n\n## FAQ\n\n### What is the difference between an LLM wiki and a personal wiki?\n\nAn LLM wiki is maintained by the model. It updates summaries, cross-links, index entries, and contradictions as new material arrives. A normal personal wiki depends on manual editing.\n\n### What sources can I ingest?\n\nWeb pages, papers, blog posts, PDFs, markdown files, text files, and pasted text. The skill converts everything into markdown under `raw/` and compiles it into `wiki/`.\n\n### Is this production-ready?\n\nThe workflow is based on a real knowledge base with 94 articles and 99 sources maintained daily since April 2026. The repo includes examples, templates, and a design spec.\n\n## Design Boundaries\n\nDeliberately not built, after three months of production logs and a survey of the ecosystem (LLM Wiki v2, llm-wiki-compiler, OKF, agent-memory literature):\n\n- **Source-hash freshness tracking** — raw/ is immutable, so hashes guard against events that cannot happen. Genuinely new information arrives as new sources through normal ingest.\n- **Persisted line-number citations** — every observed fidelity error was \"value absent from the source\", which a whole-file grep catches. Anchors only disambiguate a failure mode that has not occurred, and the annotation friction makes agents skip the rule.\n- **Numeric confidence or quality scores** — false precision with no calibration behind it. Evidence strength belongs in the prose.\n- **Per-article review dates** — nobody can predict at compile time how fast a domain moves. Maintenance is driven by whole-wiki lint, not per-page timers.\n- **Access-based decay** — frequently asked is not the same as true.\n- **Retract / bad-source machinery** — has not happened yet. Handle it manually until it does.\n- **Automatic hooks and scheduled runs** — those belong to the agent harness, not a tool-agnostic skill.\n- **Vector or graph search** — at 50K–100K tokens of curated wiki, grep and read are more reliable. Add search tooling only when recall measurably degrades.\n- **Typed relationship ontologies** — link semantics live in the prose around the link.\n- **OKF conformance** — the spec is a v0.1 draft with a minimal tooling ecosystem. Tracked; will be revisited.\n- **MCP servers, UIs, output subsystems** — outside the boundary of a tool-agnostic skill.\n\n## Inspired By\n\nUnofficial community implementation of the workflow from [Karpathy's LLM Wiki idea](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f). The value here is the reusable workflow, prompt structure, and battle-tested knowledge-compilation rules.\n\nSee also: [lucasastorian/llmwiki](https://github.com/lucasastorian/llmwiki), [atomicmemory/llm-wiki-compiler](https://github.com/atomicmemory/llm-wiki-compiler). We are tracking Google's [Open Knowledge Format](https://github.com/GoogleCloudPlatform/knowledge-catalog/tree/main/okf) draft and will evaluate compatibility once the spec and tooling mature.\n\n## License\n\n[MIT](LICENSE)\n",
  "bytes": 7756,
  "sha": "67dc43c26a747ece5ed56323a7e0a16dc66b50f85e47ef55f7c35309f67de847",
  "repo_slug": "astro-han/karpathy-llm-wiki",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/skl_astro_han_karpathy_llm_wiki_karpathy_llm_c5f40711/readme"
}