Obsidian Agent Brain — AI skill for Claude Code

Obsidian Agent Brain

Turn your Obsidian vault into a structured second brain that AI agents (Claude Code, Codex, Aider, etc.) can navigate without you repeating yourself every session.


_What the Obsidian graph looks like after this system is applied to a ~200-note vault. Green nodes are active work; grey are everything else. MOCs act as hubs — you can see them as the denser clusters._

What this is

A set of conventions + scripts that convert "a folder of markdown files" into:

• Maps of Content (MOCs) — one hub page per domain, auto-populated by Dataview
• A strict YAML frontmatter schema — every note tagged with type/domain/status so queries work
• A dense wikilink graph — Obsidian's graph view finally becomes useful
• A hybrid RAG search tool — local FastEmbed + SQLite over every note, for "what did I say about X"
• A SessionStart hook — auto-injects vault context into every Claude Code session
• A living handoff file (PICKUP.md) — mid-project sessions resume without re-explaining

Net effect: a new agent session starts with ~40KB of your relevant context already loaded, plus semantic search over everything else.

Why

If you use an AI coding agent (or any agent that reads/writes a vault), you probably hit one or more of:

• "Random folder problem" — you open the agent in some project dir, then want to work on something else, now you're re-explaining context.
• Flat vault — 100+ markdown files but nothing links to anything. Obsidian's graph is useless.
• Forgotten notes — you wrote down something useful three months ago and can't find it.
• Repetition — you explain "my homelab is…" every new session.

This repo is the opinionated fix for all of the above.

Credit

Architecture inspiration: [coleam00/second-brain-starter](https://github.com/coleam00/second-brain-starter) and Cole Medin's YouTube channel. This repo is a different take — no chatbot, no heart