claude-obsidian
A plugin and skill set that turns an Obsidian vault into a knowledge base with sources, links and grounded answers
Install
git clone https://github.com/AgriciDaniel/claude-obsidian.git
cd "$HOME/Documents/MyKnowledgeVault"
claude --plugin-dir /absolute/path/to/claude-obsidianInitialize the vault first: python3 scripts/claude-obsidian.py init <path> with --generated-at and --operation-id, then rerun with --approved-plan-sha256 and --apply.
This is third-party code. Review the repository files before installing.
What it does
The plugin runs a wiki in a normal Obsidian vault following Karpathy's LLM Wiki pattern. Sources go into an inbox, the agent keeps immutable copies, creates linked pages and maintains source and claim ledgers that flag contradictions. You can then query the vault, search with BM25, lint dead links and orphans, build Canvas maps and run bounded web research with a separate merge. Every write is one recoverable transaction with a preview, and network or destructive actions need explicit consent. It ships 15 skills, including PARA, LYT and Zettelkasten modes.
Who it is for. For researchers, analysts and anyone keeping a personal Obsidian knowledge base who wants the agent to fill it with cited sources.
Good fit when
- You need to turn articles and notes into a linked, cited wiki
- You want answers grounded only in what is already in the vault
- You need regular cleanup of dead links and stale pages
Not a fit when
- You want automatic transcript capture or cloud sync
- You do not use Obsidian and do not want Markdown files
Example request
Ingest the new files from inbox into the wiki with source citationsLimitations
Setup is heavier than a typical skill: clone the repo, initialize a vault with a Python script confirming the plan hash, and run Claude Code with --plugin-dir. PDF and EPUB content is not extracted yet; URL, YouTube and OCR need an external runner. Not a substitute for backups.
How to disable. Run Claude Code without --plugin-dir or remove the plugin via /plugin. The vault stays a plain folder of files.
Security check
- Writes files to the vault and runs Python scripts
- The autoresearch skill uses the web after consent
README in short
The README describes a loop: keep the source, ground claims, connect knowledge and reuse it in answers. Quick start is clone, initialize a separate vault with plan review, and run Claude Code with the plugin; a setup script covers Codex, OpenCode and Gemini. It lists the 15 skills, the transaction model and a candid table of supported inputs. It ends with a CLI reference and upgrade and uninstall notes.
FAQ
Are my notes uploaded?
No, the vault stays local and network requests happen only with explicit consent.
Can I use an existing vault?
Yes, the non-destructive adopt command is covered in the install guide.
Related
A CLI for every Google Workspace API with JSON output and agent skills: Drive, Gmail, Calendar, Sheets and more
Local search over Markdown notes, docs and meeting transcripts: keywords, semantic search and reranking, with an MCP server
A task manager for AI-driven development: breaks a PRD into dependent tasks and guides the agent through them via MCP or CLI
Anthropic's official plugins for knowledge work roles: sales, support, product, marketing, legal, finance, data and enterprise search