Are there any existing ways to defend against bad stuff that the agent can do in the unmonitored scaffold?