Analyzes a before/after command diff for risky tool-call changes. It flags new actions such as network access and returns a risk score, evidence, and recommendations.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.