Requirements Wiki — Starter

You bring the raw material; the AI agent files it.

The pattern

This vault follows the Karpathy three-layer LLM-wiki pattern. raw/ holds your immutable sources — interview notes, transcripts, specs — which you drop in and never edit. wiki/ holds the structured, interlinked requirements knowledge (glossary, stakeholders, use cases, stories, and more) that the AI agent reads sources into and keeps current. A schema layer — CLAUDE.md, _templates/, _system/anchors/ — defines the content types and conventions both layers share, and evolves with you over time.

Quick start

1. Run ./dashboard.sh to start the live dashboard (Docker) at http://localhost:8080 — a browsable view of the wiki, useful even while it's still empty. No Docker on this machine? ./dashboard.sh local runs it directly with Python at http://localhost:8000.
2. Open this folder in Obsidian (Open folder as vault) to navigate and graph-view the wiki directly.
3. Point Claude Code at this same folder — it reads CLAUDE.md automatically and discovers the bundled skills in .claude/skills/.

For a full walkthrough, see docs/GETTING-STARTED.md. If you're running this as a workshop, see docs/WORKSHOP.md.