Composability Primitives

Codex is built on a composable architecture where multiple primitives layer together to extend the agent's capabilities. Understanding these building blocks — and how they interact — is the key to unlocking Codex's full power. The five core primitives are AGENTS.md (persistent project context), skills (reusable task-specific instructions), slash commands (quick in-session actions), MCPs (external tool integrations), and sub-agents (delegated parallel work).

These primitives form a hierarchy. AGENTS.md sits at the base — it loads automatically and shapes every interaction. Skills layer on top, activating when relevant tasks arise. MCPs connect to external services (GitHub, databases, design tools). Sub-agents let you spawn parallel workers for independent tasks. Slash commands are the interactive glue that lets you control all of this from the chat interface.

The power comes from composition. An AGENTS.md file can reference skills. A skill can use MCP tools. A sub-agent inherits the parent's AGENTS.md context. Understanding how these layers compose is what separates a Codex user from a Codex power user.

The best Codex configurations are not the ones with the most tools installed — they are the ones where every primitive has a clear purpose and nothing overlaps.

— RC Academy — DRILL