second-brain

name	second-brain
description	Personal knowledge base using the LLM Wiki pattern (Karpathy). Activates on explicit /second-brain invocation. Supports: 'ingest', 'query', 'lint', 'health check', 'weekly summary', 'monthly report'.

Second Brain -- LLM Wiki

Personal knowledge base following Karpathy's LLM Wiki pattern.

Setup

At session start, locate and read the schema:

Vault path discovery (in order):

1. Config file ~/.claude/second-brain.json -- read vault_path key
2. Environment variable SECOND_BRAIN_PATH if set
3. Default: SET_YOUR_VAULT_PATH_HERE

Replace the default path above with the absolute path to your Obsidian vault. Or create ~/.claude/second-brain.json with: {"vault_path": "/path/to/vault"}

The vault root is wherever CLAUDE.md, index.md, log.md, raw/, and wiki/ live.

Read CLAUDE.md from the vault root -- this is the schema with all conventions, workflows, and protocols. Follow its instructions exactly.

Tool priority (from CLAUDE.md Rule 5):

1. If Obsidian vault MCP is available -- use it for ALL vault operations (read, write, search, list)
2. If MCP unavailable -- fall back to Obsidian CLI or file system Read/Write
3. Never mix MCP and filesystem in the same session

Operations

The schema defines three operations:

• Ingest -- capture a source into raw/, update wiki pages, update index, log it
• Query -- read index, find relevant pages, synthesize answer, file back good answers as wiki pages
• Lint -- health check the wiki, mark done items, flag undone items, find gaps

Activation Rules

This skill activates ONLY on explicit /second-brain invocation. Generic phrases like "look up", "todo", or "weekly summary" in normal conversation should NOT trigger this skill -- only when prefixed with /second-brain.

Mutating operations (ingest, lint) require explicit invocation. Read-only queries may activate on /second-brain query or /second-brain what do I know about.

Tools

Use whatever tools are available on this device. Check CLAUDE.md section "Tools Available". If no tools are available, the user pastes content and the LLM processes it with file system Read/Write.

That's It

The schema (CLAUDE.md) in the vault is the single source of truth. It travels with the vault across devices. This skill file just points to it.