Write a policy for an agent

What this shows

Policy is where governance actually gets decided, so this walkthrough starts there: writing a Cedar policy for a specific agent and watching it take effect.

The policy is authored against canonical action classes rather than raw endpoints, which is what makes it portable across tools and agent vendors. It is then published from the Authority and distributed to every connected sidecar, where it is enforced locally without a network round-trip. The next call the agent makes is evaluated against the new rules, and the resulting decision appears in the audit log with the policy that resolved it.

The point of the demo: the loop from “we need a rule about this” to “that rule is enforced on every agent in the organization” is a short one, and every step of it leaves a record.