Decodes on-chain tx intent and returns a signed risk verdict: safe, review, suspicious, or dangerous.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.