SecondBrain

A portable template for an AI-maintained markdown knowledge base using the LLM Wiki pattern.

The core idea: keep source material in raw/, let AI agents compile durable insights into wiki/, and preserve everything as plain Markdown so the system stays portable, auditable, and easy to inspect. A gated tools/ layer of small maintenance scripts keeps the derived pages (indexes, cross-links) in sync with the source-of-truth frontmatter, instead of relying on an agent to hand-edit them correctly every time.

What's included

• SecondBrain-Setup-Guide.md — setup and replication guide
• CLAUDE.md — agent schema/instructions for maintaining the vault
• INGEST-PIPELINE-CONTRACT.md — operational contract for an automated ingest pipeline or agent
• tools/ — versioned, gated maintenance scripts (index rebuild, link regeneration, lint)
• wiki/patterns/ — a starter set of documented tooling gotchas and their fixes
• .gitignore — ignores local Obsidian/cache artifacts

Quick start

• Copy this repository or use it as a template.
• Create/open it as an Obsidian vault.
• Customize CLAUDE.md with your own context, domains, and rules.
• Add source files to raw/.
• Ask your AI agent to read CLAUDE.md before ingesting or querying. If a separate automated

pipeline runs ingest on a schedule, give it INGEST-PIPELINE-CONTRACT.md directly.

• After any content batch, run python3 tools/regen_links.py, then python3 tools/rebuild_index.py,