An agent's progress should survive the process that runs it. Its code should run within a policy you can review, and its failures should leave a history you can…