Second Brain Builder — AI skill for Claude Code

Second Brain Builder

A Claude skill that stands up and runs a knowledge base that maintains itself, then grows it into a **company brain**: the core truth, decisions and knowledge every agent in your business starts from.

You curate what goes in. The LLM files it, links it, keeps it tidy, and has a brief waiting for you each morning.

**No backend. No vector database. No accounts.** A folder of markdown in git, a few scheduled jobs, and an LLM doing the upkeep. You can read, version and correct every file.

Why this exists

Most knowledge bases die the same way. Capture is easy. Maintenance is a second job, and it loses to real work every time. So the links pile up unread and the calls evaporate by Friday.

This skill treats notes as a by-product of the work, not a task on top of it:

• Capture is passive. Your sources are fetched on a schedule, overnight, whether you show up or not.
• The LLM maintains, you curate. It structures, cross-links and self-heals. You decide what matters.
• Retrieval stays cheap. A routing table points the model at two pages, not fifty, so the base doesn't get slower as it grows.
• Nothing is deleted. Stale items fall off the daily surface and stay findable forever, which is what makes it safe to put everything in.
• Expensive steps sit behind a human yes. The system drafts freely. Anything costly or public waits for you.

What's in it

File	What it does
SKILL.md	The skill: the five rules, setup, and the routines an agent runs.
references/schema-template.md	The operating contract you copy into your vault as schema.md: layers, routing table, and every operation (ingest, query, call ingest, promote outcomes).
references/morning-brief-routine.md	The daily brief and the end-of-session content extraction.
references/self-heal-checklist.md	Eleven rotating checks. Structural issues fixed on the spot, content issues flagged for you.
`references/architecture-and-auto