StateM is an agent-native execution runtime designed for long-running command-line interface (CLI) agents. It externalizes procedural state outside the model context by organizing execution around durable states, phase-local context boundaries, checked transitions, recoverable runbooks, and versioned procedural practices. Unlike conventional orchestrators that break an agent into fragmented, controller-directed sub-prompts or model calls, StateM preserves the primary agent's unified reasoning loop while exposing a shared, machine-operable control surface directly within the agent's native CLI action space.