Generates deterministic evaluation cases for agent prompts, including happy-path, ambiguous, and adversarial-input variants.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.