The Agensi Peer Review Stress Test is a skill that functions as a pre-submission quality gate for AI agent output. It instructs an agent to adopt an adversarial reviewer’s posture towards its own generated text. Instead of simply drafting content, the agent critically evaluates its work for weaknesses. The output of this skill is a structured review verdict, not a rewrite. This verdict includes identifying the single weakest point in the argument, flagging any overclaims and suggesting downgrades, listing missing limitations, and surfacing unstated assumptions. It also generates the three toughest questions a skeptical reviewer would ask. This process involves the agent performing five distinct review passes: a weakest-claim hunt, an overclaim scan, a missing-limitation check, an unstated-assumption audit, and a hostile-question rehearsal. The final decision provided is either REVISE, ADD CAVEAT, or PASS, accompanied by a one-line justification. This process helps to separate the writing and reviewing roles within the AI model, encouraging a more critical self-assessment and improving the quality of the final output before human review.

Use cases

Use it to review research summaries for logical weaknesses.

Use it to stress-test recommendations before presentation.

Use it to evaluate technical explanations for clarity and completeness.

Use it to pre-screen customer-facing answers for potential misstatements.

Use it to scrutinize analyses for unstated assumptions.

When to use it

When factual accuracy and absence of overstatement are critical.

When generating outputs that will be quoted or acted upon.

When drafting content where being incorrect is costly.

When an agent's confident-sounding prose needs scrutiny for unearned certainty.

When a structured, critical self-review is needed for text output.

When not to use it

When the primary need is external fact-checking or data verification.

When the agent needs to access the internet or run code for validation.

When the task requires rewriting or rephrasing the original text.

When the capability is expected to certify factual claims as true.

When grounding the output with external evidence is the main objective.