Detects prompt injection attempts in text and flags malicious instructions.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.