Overview

claude-obsidian (AgriciDaniel) is a 9.2k⭐ repository for building an Obsidian + Claude Code second brain. It is listed in the vault’s Resources page as a major tool/repository for the LLM Wiki pattern.

Key Facts

  • Stars: ~10.6k
  • Stack: Obsidian + Claude Code second brain
  • Role: bridges Claude Code agent operation with the Obsidian front-end
  • License: MIT
  • Skills: 15 skills in one system
  • Architecture: local-first, file-owned, transaction-safe

Features

Core Architecture

  • Local-first — vault is a normal directory of Markdown, JSON, and source files
  • Transaction-safe — one logical operation = one recoverable transaction
  • Product/vault boundary — product never treats checkout as the default vault
  • 15 skills — wiki, save, wiki-ingest, wiki-query, wiki-lint, autoresearch, canvas, defuddle, wiki-fold, wiki-mode, wiki-retrieve, wiki-cli, obsidian-markdown, obsidian-bases, think

Ingest & Processing

  • wiki-ingest — turns captured sources into linked pages and provenance records
  • Content-addressed capture — immutable copies of source material
  • Inbox workflow — sources placed in inbox/, processed into wiki/
  • defuddle — clean, readable web content before ingestion
  • autoresearch — bounded web research with explicit egress control

Search & Retrieval

  • wiki-query — answers read-only from relevant vault evidence
  • wiki-retrieve — contextual prefixes, BM25, optional cosine reranking
  • wiki-cli — Obsidian CLI reads and search
  • Source-grounded — every claim linked to source evidence

Knowledge Graph

  • Obsidian Graph view — visual navigation of wikilinks
  • Canvas views — wiki-scoped Obsidian Canvas creation
  • Maps of Content — LYT-compatible MOC structure
  • 4 methodology modes — Generic, LYT, PARA, Zettelkasten

Quality & Governance

  • wiki-lint — dead links, orphans, metadata gaps, stale indexes
  • Source/claim ledgers — authority, freshness, support, contradiction, confidence
  • High-risk claims — require two independent sources
  • Contradictions visible — unsupported claims never hidden

MCP / Tools

  • Claude Code plugin — native plugin for Claude Code
  • Agent Skills compatible — works with any Agent Skills host
  • Multi-host support — Claude Code, Codex, OpenCode, Gemini, Cursor, Windsurf

Integration

  • Claude Code native — primary integration target
  • Obsidian — graph view, Canvas, Bases, plugins
  • Cursor + Windsurf — workspace-local skill discovery
  • Canvas — visual knowledge maps in Obsidian Canvas

Data & Storage

  • Markdown files — plain .md, fully portable
  • Content-addressed — SHA-256 hashed source copies
  • Provenance records — every claim traceable to source
  • Obsidian-compatible — works as Obsidian vault from day one

Relationships

  • “operates-with” obsidian — the front-end it targets
  • “operates-with” claude-code — the agent it wires to Obsidian
  • “compatible-with” llm-wiki — second-brain implementation of the pattern
  • “belongs-to” second-brain — its explicit purpose

Implications

claude-obsidian is the direct “Claude Code ↔ Obsidian” bridge — the same pairing this vault uses, but as a packaged repo. Useful reference for the viewer/agent integration the user already runs.

Relationship to This Vault

claude-obsidian provides a reference implementation for the Obsidian + Claude Code integration that this vault uses manually. The vault’s obsidian-skill-commands pages document the commands that claude-obsidian automates as a packaged skill.

Sources