How to audit AI agent activity

The script ran, exit code 0, but nobody can say why the agent chose it. Standard logs prove an AI agent acted, but accountability needs the whole chain, including the request, the context it drew on, the approval, the action, and the verified outcome.