A reusable skill for building multi-repo LLM wikis with Claude Code, Cursor, Codex, and other Agent Skills tools.
llm-wiki packages the LLM wiki workflow into one installable Agent Skills skill. Your coding agent ingests sources into sources/, compiles durable knowledge pages into wiki/, answers questions with citations, lints the wiki for consistency, and commits wiki updates with reasons.
An 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.
This skill gives you three operations:
| Operation | What it does | Output |
|---|---|---|
| Ingest | Collects a source into sources/ and compiles it into the wiki |
New or updated wiki pages plus a git commit |
| Query | Searches the wiki and answers with citations | Grounded answers linking to markdown pages |
| Lint | Checks index integrity, links, and wiki health | Auto-fixes plus reported issues, with commits when files change |
See SKILL.md for the full skill specification.
| Approach | Knowledge lives in | When synthesis happens | Good for |
|---|---|---|---|
| RAG | Raw chunks and embeddings | At query time | Broad retrieval across large corpora |
| LLM Wiki | Curated markdown pages | During ingest and maintenance | Compounding knowledge, summaries, and durable cross-links |
This skill is optimized for the wiki model: knowledge that improves over time instead of re-deriving relationships on every query.
The wiki uses two scopes:
projects/for repo-specific knowledgeorganization/for reusable and cross-repo knowledge
organization/ starts intentionally small:
patterns/architecture/topics/
This keeps repo detail close to the source while still letting the wiki accumulate shared lessons across many codebases.
Based on a production knowledge base maintained daily since April 2026:
- 94 wiki articles across 13 topic directories
- 99 source materials ingested
- 87 operation log entries in the last 7 days
See examples/ for sample wiki pages, source files, and operation logs.
npx add-skill ColinThompson1/llm-wikiWorks with any tool that supports the Agent Skills standard.
Give the skill a URL, a file, or pasted text:
"Ingest this article: https://example.com/attention-is-all-you-need"
The skill stores the source in sources/, then compiles or updates the right knowledge pages in wiki/.
"What do I know about attention mechanisms?"
The skill searches the wiki and answers with citations linking back to your markdown pages.
"Lint my wiki"
Checks for broken links, missing index entries, stale cross-references, and related issues.
The core idea behind this workflow: the LLM maintains the wiki while the human focuses on choosing sources and asking good questions.
knowledge-base/
├── sources/
│ ├── external/
│ │ └── 2026-04-03-source-article.md
│ └── projects/
│ └── repo-a/
│ └── 2026-04-03-design-note.md
├── wiki/
│ ├── projects/
│ │ └── repo-a/
│ │ └── concept-name.md
│ ├── organization/
│ │ ├── patterns/
│ │ ├── architecture/
│ │ └── topics/
│ ├── index.md
│ └── log.md
├── projects.md
Each new source can update multiple pages, strengthen cross-references, record contradictions, and produce a git commit that explains why the knowledge changed. That is what makes the wiki compound over time.
Every wiki-writing operation should create a git commit. By default, use one commit per ingest, archive, or auto-fixing lint run. If a single operation changes unrelated topics, split the commits.
When facts change, the commit message should explain why:
wiki: ingest retry-strategies
Reason: new source showed jitter is mandatory for distributed retries, which changes the prior summary.
Files: wiki/projects/repo-a/retry-worker.md, wiki/organization/patterns/retry-strategies.md
This skill follows the agentskills.io open standard:
| Tool | Install method |
|---|---|
| Claude Code | npx add-skill ColinThompson1/llm-wiki |
| Cursor | npx add-skill ColinThompson1/llm-wiki |
| Codex CLI | Copy to .agents/skills/llm-wiki/ |
| OpenCode | npx add-skill ColinThompson1/llm-wiki |
| Other tools | Copy SKILL.md and references/ into the tool's skill directory |
An 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.
Web pages, papers, blog posts, PDFs, markdown files, text files, and pasted text. The skill converts everything into markdown under sources/ and compiles it into wiki/.
The 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.
Implementation of the workflow inspired by Karpathy's LLM Wiki idea. The value here is the reusable workflow, prompt structure, and battle-tested knowledge-compilation rules.
See also: lucasastorian/llmwiki, atomicmemory/llm-wiki-compiler.