Checks prompts and content for injection risks and malicious instruction patterns.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.