Documentation
¶
Overview ¶
Command authority shows authority-as-governed-state: a principal's delegated limit is a state variable, so the compliance invariant is scoped to whoever is acting, "exposure must not exceed THIS desk's limit," and one machine-checked policy covers every possible limit at once. The limit is seeded at run start from the identity/authority the deployment binds (agent.WithIdentity), and every governed action's leaf commits to who acted, on whose behalf, under what authority, and the resulting state.
No LLM or network: the "agent" here just calls the governed buy tool in a loop, so the example runs offline and exercises the governance + identity + audit machinery. Swap the loop for a real agent and the guarantees are identical.