Anatomy of a command center

A command center is a markdown repo your agent reads as its operating manual. It has seven parts, each with one responsibility.

1. Router — the entry point (CLAUDE.md, AGENTS.md, GEMINI.md, or .cursor/rules/command-center.mdc). Startup routing, always-on rules, and the plan → review → approve → execute gate.
2. Domains — on-demand capsules in domains/, plus domains/_registry.md. Each capsule declares its entry point, where its work lives, its rules, its permissions, and its recent state.
3. Memory — memory/ holds durable facts, indexed by memory/MEMORY.md and governed by memory/_how-memory-works.md.
4. Notes store (optional) — references/ holds the schema for an external knowledge base the agent navigates on demand. 4b. Actions (optional, macOS-first) — actions/macos.md gives the agent verified recipes to perform reminders, calendar events, and email, under a tiered safety model (local actions directly; outward actions like email draft-first).
5. Governance — governance.md: what changes need a ritual, and the source-of-truth hierarchy.
6. Security — SECURITY.md: keep secrets out of the repo.
7. Workspace — workspace/: dated scratch space for plans.

Memory vs. the notes store

Both hold knowledge, but they are different tools:

• Memory is compact, durable facts the agent recalls every session — preferences, decisions, project state. Small, in-repo, always loaded via the index.
• The notes store is the deep knowledge base — analysis, research, source material — the agent navigates on demand. Large, external, never fully loaded.

Rule of thumb: if the agent must know it every session, it’s memory; if the agent should find it when relevant, it’s the notes store.