Guide · Agent skill
How to use the LLM Wiki Knowledge Base agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The wiki skill gives an AI coding agent a persistent markdown knowledge base for a project. It is inspired by Karpathy's LLM Wiki idea: knowledge the agent learns in one session is written down and can be found again in later sessions. This guide explains what the skill does, how to install it, how to trigger it, and what to check before you rely on it. You do not need prior experience with agent skills.
LLM Wiki Knowledge BaseYeachan-Heo/oh-my-claudecode · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the skill ZIP from this page and unzip it so that SKILL.md is inside the skill folder.
- 2Choose an install location. Place the folder in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder for that project only. Alternatively, upload it in the Claude apps under Settings, then Capabilities, then Skills.
- 3Confirm the oh-my-claudecode environment. Make sure the oh-my-claudecode plugin environment is available, since the skill needs its wiki_* tools to work.
- 4Save your first page. Ask the agent to wiki this a decision or finding, or use wiki add for a single page with a title, content, tags and category.
- 5Search in a later session. Use wiki query with keywords or tags to have the agent find and cite existing pages before it starts new work.
- 6Run a health check. Run wiki lint to find orphan pages, stale content, broken cross-references and oversized pages, then fix or remove them.
- 7Review captured notes. Check session-log pages and the log at .omc/wiki/log.md, and adjust wiki.autoCapture in .omc-config.json if needed.
What the wiki skill makes the agent do
Without a persistent store, an agent starts each session with little memory of earlier decisions, architecture choices or bugs you already fixed. This skill lets the agent save key findings, decisions, architecture notes and debugging lessons as markdown pages inside your project folder. In later sessions, the agent searches those pages and cites them in its answers instead of starting from scratch.
The skill covers several operations: ingest (process knowledge into pages, where one ingest can touch several pages), query (search pages by keywords and tags), lint (run health checks), quick add (add a single page), list, read and delete, and log (view operation history).
- Categories: architecture, decision, pattern, debugging, environment, session-log
- Pages are stored as .omc/wiki/*.md with YAML frontmatter
- An index at .omc/wiki/index.md is maintained automatically as a catalog
- An append-only operation history is kept in .omc/wiki/log.md
- Pages can link to each other with [[page-name]] syntax
Install the wiki skill
Download the ZIP file for the skill from this page and unzip it. The skill's main file is SKILL.md. Place the unzipped folder in one of two locations: ~/.claude/skills/ to make it available across your projects, or .claude/skills/ inside a specific project to scope it to that project.
If you use the Claude apps instead, you can upload the skill under Settings, then Capabilities, then Skills.
The skill also requires the oh-my-claudecode plugin environment, because that environment provides the wiki_* tools the skill calls. The source file does not describe how to install that plugin, so check the oh-my-claudecode project's own documentation for setup.
- Option 1: unzip into ~/.claude/skills/ for personal use across projects
- Option 2: unzip into a project's .claude/skills/ folder to share it with that repository
- Option 3: upload it in the Claude apps under Settings, then Capabilities, then Skills
How to trigger the skill with natural requests
The skill declares these trigger words: wiki, wiki this, wiki add, wiki lint and wiki query. Starting a request with or including one of these phrases is the most direct way to invoke it. Because the agent interprets your request, you can also describe the task in plain language, such as asking it to save the decision you just made about authentication, but the listed triggers are the reliable starting point.
Being specific helps. Name the topic, say whether it is a decision or a debugging lesson, and mention tags if you want them used for later search.
- wiki this: capture what we just worked out
- wiki add: save one page quickly
- wiki query: look up what the wiki already says about a topic
- wiki lint: check the wiki for problems
Examples of wiki operations
Ingest is the fullest form of saving. The agent passes a title, content, tags and a category. For a quick note, quick add takes the same kind of fields with less process.
Query searches by keywords and tags and returns matching pages with snippets. The agent then writes an answer and cites the pages it used. Lint reports orphan pages, stale content, broken cross-references, oversized pages and structural contradictions.
wiki_ingest({ title: "Auth Architecture", content: "...", tags: ["auth", "architecture"], category: "architecture" })
wiki_query({ query: "authentication", tags: ["auth"], category: "architecture" })
wiki_lint()
wiki_add({ title: "Page Title", content: "...", tags: ["tag1"], category: "decision" })
wiki_list()
wiki_read({ page: "auth-architecture" })
wiki_delete({ page: "outdated-page" })Keep the wiki healthy and review what it records
Run lint from time to time to catch orphan pages and broken links. Read the operation history in .omc/wiki/log.md to see what was written and when. Because pages are plain markdown files, you can open and edit them by hand in any editor.
Auto-capture is enabled by default. At session end, significant discoveries are saved as session-log pages. You can turn it off through the wiki.autoCapture setting in .omc-config.json. Review session-log pages before sharing the project, since automatic capture does not check for secrets or private details.
- Set wiki.autoCapture to disabled in .omc-config.json if you do not want automatic session notes
- Review session-log pages for secrets or private information
- Pages are git-ignored by default, so add them to version control deliberately if your team wants them shared
Safety and what the skill needs
According to the skill file, it does not run scripts and does not use the network. Its data stays in your project folder as markdown. The main requirement is the oh-my-claudecode plugin environment that supplies the wiki_* tools.
Search is limited. The skill uses keyword and tag matching only, with no vector embeddings. Pages phrased differently from your query may not be found, so use consistent tags and clear titles.
- Runs scripts: no
- Uses network: no
- Requires the oh-my-claudecode plugin environment providing wiki_* tools
- Search is keyword and tag based only
What people use it for
- Keeping architecture notes for a codebase you return to over many sessions
- Recording decisions and their reasons so a teammate can read them later as plain markdown
- Saving debugging lessons and environment setup details so the same problem is not solved twice
- Maintaining a personal, hand-editable notebook that the agent also updates and searches
Frequently asked questions
Is the wiki agent skill free?
The skill is released under the MIT licence. The source file does not describe pricing, so check the oh-my-claudecode project and this directory page for any details about cost or the Claude plan you need.
What do I need to run the wiki skill?
You need the oh-my-claudecode plugin environment, because the skill depends on the wiki_* tools it provides. The skill itself runs no scripts and makes no network calls, according to its file.
How do I install the wiki skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings, then Capabilities, then Skills.
Does the wiki skill use semantic or vector search?
No. It uses keyword and tag matching only, with no vector embeddings. Pages worded differently from your query may be missed, so use consistent tags and clear titles.
Where are the wiki pages stored and are they shared with my team?
Pages are markdown files in .omc/wiki/ inside your project. They are git-ignored by default, so they are not shared automatically. You can choose to commit them if your team wants them under version control.
How can I turn off automatic session notes?
Auto-capture is on by default. Set wiki.autoCapture in .omc-config.json to turn it off, and review any session-log pages it creates for secrets or private details.
How does the wiki skill compare to a regular notes app?
The pages are plain markdown that you can edit by hand, and the agent can write to and search them during a session. Unlike a notes app, search is limited to keywords and tags, and the agent maintains the catalog and runs health checks.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers and teams who work on the same codebase over many sessions and want decisions and architecture notes kept as plain, reviewable markdown. It is a weaker choice for large knowledge bases that need semantic search, or for sensitive projects where automatic session notes could capture secrets unless you turn auto-capture off and review what it writes.
Download the skill