role-reviewer

To validate agent outputs for completeness and intent drift, use an adversarial review process that independently checks completed work against stated goals, detecting incomplete tasks, non-holistic behavior, and validation looseness.

Adversarial output validation in quality assurance is an independent critique mechanism that holistically evaluates agent-produced results for completeness, consistency, edge cases, and potential biases. It surfaces actionable fixes and risk flags by enforcing strict evaluation criteria against the original instruction set.

Review automated planning tasks for glaring issues by applying an independent adversarial critique that checks the output against stated requirements, detecting non-holistic gaps, intent drift, and adherence to best practices, which surfaces actionable improvements for downstream work.

Yes, you can use an adversarial reviewer for both code and document quality assurance. It independently critiques completed work across these formats to verify results, ensuring holistic assessment, detecting incomplete work, and enforcing validation criteria described in the instruction set.

You need holistic assessment for human-in-the-loop processes when verifying results after automation or manual input to ensure completeness and consistency. It detects validation looseness and intent drift, providing guided feedback that surfaces actionable fixes and risk flags for downstream work.

The limitations of using an independent critique subagent for output validation include its reliance on the evaluation criteria described in the instruction set, meaning it cannot assess undefined requirements or discover risks outside its programmed adversarial scope.