[Placeholder: not written yet.]
Intent: the policy written before the agent. How the models run locally, exactly what the monitor can read, how read-only access is enforced, the log of what it sees and reports, and the limits of running models locally.