llm-wiki-agent

LLM Wiki Agent is a skill for coding agents that builds and maintains a personal knowledge base. You place source documents in a raw/ directory and ask the agent to ingest them; it reads the material, extracts knowledge, and writes a persistent, interlinked markdown wiki. Each new source adds to the existing structure rather than being re-derived per query. Supported agents include Claude Code, Codex, OpenCode and Gemini CLI, each reading its own schema file (CLAUDE.md, AGENTS.md or GEMINI.md). No API key or Python setup is required for the agent-driven workflow. Claude Code additionally exposes slash commands such as /wiki-ingest, /wiki-query, /wiki-lint and /wiki-graph; other agents use equivalent natural-language triggers. The generated wiki contains an index, an append-only log, a living overview synthesis, per-source summary pages, and automatically created entity and concept pages. A graph directory holds persistent node/edge data plus a self-contained vis.js HTML visualization. The graph is built in two passes: deterministic parsing of wikilinks, then agent-inferred implicit relationships marked with confidence. Community detection clusters related topics, and caching limits reprocessing to changed pages. Ingest supports many formats including markdown, PDF, DOCX, PPTX, XLSX, HTML, CSV, JSON, EPUB and audio files, with non-markdown files converted automatically via markitdown. Optional tooling provides higher-fidelity arXiv and complex-PDF conversion. Query answers can be filed back as synthesis pages, and lint reports surface orphan pages, broken links, missing entity pages and contradictions flagged at ingest time. Documented use cases include research tracking, reading a book chapter by chapter, personal knowledge management, team and business intelligence, and competitive analysis. The wiki is plain markdown and can be browsed in Obsidian, including via a vault symlink pattern. Everything runs locally with no server or database. Licensed under MIT.