llm_wiki_skill

This repository is a small package that teaches an AI coding assistant, such as Claude Code or Codex, how to talk to a separate desktop application called LLM Wiki. The desktop app runs on the user's own machine and exposes a local HTTP API. The skill is documentation only: it contains three markdown files that describe the API contract, with no client library, no SDK, and no compiled code. The AI agent reads the documentation and then makes plain HTTP calls using whatever tools it already has, such as curl or a fetch function.

Once installed, the skill lets the user ask their AI agent things like 'what does my wiki say about X', 'show me the neighborhood of node Y in my graph', 'read me the page about Z', or 'I just added new docs, re-index them'. Most actions are read-only. The only one that changes state is a rescan command that asks the wiki to look for new files. The agent can also fetch a structural overview, such as the file tree and a top-level index page.

Installation is offered in three ways. The recommended path is a one-line command using the skills CLI, which fetches the latest version from GitHub and registers it with the agent runtime. Users can also clone the repo and symlink it into the Claude Code skills folder, or drop the folder anywhere their agent runtime looks for skills. There are no dependencies and no build step.

For the skill to work, the LLM Wiki desktop app must be running, since the API only binds to a local port while the app is open. The user also needs to turn on the API server in the app settings and configure an authentication token, either through the settings panel or through an environment variable. Without a token, every endpoint refuses the call. The README includes a smoke test using curl, a troubleshooting table for common errors, and a short security model that explains the local-only listener, path traversal protection, file whitelist, and rate limits.