Checking whether generated outputs follow required rules and constraints during the agent's execution, not just at the end.