Getting started

Install the skill for your coding agent, then run the four-step loop: initialize, ingest, query, lint.

Install

There are two install paths. Claude Code gets the full plugin (skill + /wiki:* slash commands + marketplace manifest). Every other supported agent installs the same agentskills.io-format skill through the skills CLI. See the agent support matrix for the full picture.

Install uv first. It is the only prerequisite and creates isolated, pinned environments for the bundled tools.

Claude Code — full plugin

The native path: skill, slash commands, and manifest ship together.

/plugin marketplace add praneybehl/llm-wiki-plugin
/plugin install llm-wiki@llm-wiki

Once installed, the plugin works in any project. Installation does not decide where your wiki lives.

The agents below install the same skill through the skills CLI. Each uses the same command with a different -a value — add -g to install globally across all projects, or drop it to install into the current project only. Slash commands are Claude Code–only; elsewhere, invoke the skill by natural language.

Codex (OpenAI)

npx skills add praneybehl/llm-wiki-plugin -a codex -g

Invoke via /skills, $llm-wiki, or natural language. Bundled scripts run.

Cursor

npx skills add praneybehl/llm-wiki-plugin -a cursor -g

Invoke via /llm-wiki or natural language. Bundled scripts run.

Gemini CLI

npx skills add praneybehl/llm-wiki-plugin -a gemini-cli -g

Invoke via the /skills management commands or natural language. Bundled scripts run.

OpenCode

npx skills add praneybehl/llm-wiki-plugin -a opencode -g

The agent invokes the skill through its native skill tool on natural-language requests. OpenCode also reads .claude/skills/ and ~/.claude/skills/, so a Claude Code install works here too. Bundled scripts run.

OpenClaw

npx skills add praneybehl/llm-wiki-plugin -a openclaw -g

The skill is auto-exposed as a user command. Bundled scripts run.

Pi Agent

npx skills add praneybehl/llm-wiki-plugin -a pi -g

Invoke via /skill:llm-wiki or natural language. Bundled scripts run.

OMP ("Oh My Pi")

OMP reads managed and user skills from ~/.omp/agent/skills/ and surfaces them automatically via skill://. OMP is not yet in the npx skills registry, so install by cloning the repo and symlinking the skill:

git clone https://github.com/praneybehl/llm-wiki-plugin.git
mkdir -p ~/.omp/agent/skills
ln -s "$(pwd)/llm-wiki-plugin/skills/llm-wiki" ~/.omp/agent/skills/llm-wiki

Other runtimes — clone and symlink

Hermes Agent and any other agentskills.io-compatible runtime not yet in the npx skills registry can use the skill by symlinking it into the agent's skills directory:

git clone https://github.com/praneybehl/llm-wiki-plugin.git
mkdir -p ~/.hermes/skills
ln -s "$(pwd)/llm-wiki-plugin/skills/llm-wiki" ~/.hermes/skills/llm-wiki

The wiki itself is agent-agnostic — it is just a directory of markdown files. You can ingest with one agent and query with another.

Choose the wiki location

Skill scope and wiki scope are independent:

• Personal global wiki: one knowledge base at a stable user-level path such as ~/wiki/, with raw sources at ~/wiki/raw/. Add that path to your agent's global instructions. Any project can contribute durable findings and sources to this wiki.
• Project wiki: wiki/ and raw/ live in one project. Add a relative-path pointer to that project's agent-memory file. This keeps knowledge isolated and makes it easy to version with the repository.

Installing the skill globally with -g only makes the skill available everywhere. It does not create a global wiki. Likewise, a global wiki does not automatically crawl projects; the agent ingests project material when asked or when its global instructions require durable findings to be preserved.

Your first wiki

The commands below use Claude Code's /wiki:* slash syntax. In other agents, phrase the same requests in natural language.

1. Initialize the structure. For one personal wiki shared across projects:

   /wiki:init --global

   For a wiki isolated to the current project, run /wiki:init without --global.

   Expected — Bootstraps the wiki structure, then installs FastEmbed 0.8.0, sqlite-vec 0.1.9, and PyYAML 6.0.3; downloads the local BAAI/bge-small-en-v1.5 model if absent; builds the parse cache; and embeds every current section. Setup is complete only when the runtime JSON reports "status": "ready". The agent then walks through the schema and offers agent-memory integration; nothing is written there without approval.

   Already have a wiki from an earlier version? Run /wiki:upgrade instead. It performs the same mandatory runtime setup and incremental full-corpus synchronization while leaving existing pages untouched.

2. Add your first source. Drop anything textual into the configured raw root, such as ~/wiki/raw/ for the global layout, then:

   /wiki:ingest ~/wiki/raw/your-source.pdf

   Expected — The agent resolves the configured wiki, reads the source (chunk-reading if it is large), discusses the takeaways with you, writes a source summary, surgically updates the entity and concept pages it touches, creates new pages (each with at least one inbound link), updates the index, and appends one line to log.md.

3. Ask a question. Once a few sources are in:

   /wiki:query What does my wiki say about X?

   Expected — The agent reads the index to find candidate pages, reads them, synthesizes an answer with [[wikilink]] citations, and offers to file substantive answers back into wiki/synthesis/ so the exploration compounds.

4. Run a health check periodically:

   /wiki:lint

   Expected — Catches orphan pages, broken wikilinks, oversized pages, missing frontmatter, and stale claims. Findings are presented as proposed edits — the wiki is never rewritten silently.

INFO

You don't have to use the slash commands. The agent triggers the same skill on natural-language requests too. The commands are there when you want explicit invocation — see the Commands reference.