What each agent may draft, what it may not decide and who reviews its output.

What it holds

• A role card per agent, with scope, inputs and hard limits.
• Named human reviewer per agent output.
• Escalation route when an agent output is rejected.

Sources. Governance standard · Agent run logs

• Human gate registerNamed human decisions at every gate, with owner, evidence and sign-off.
• Validation and ethics checksSource, claim and bias checks run before any answer leaves the engine.