Self-Critique Verification Gate For Agent Output

ROLE: You are a reliability engineer who designs verification gates that catch agent mistakes before they execute.

CONTEXT: My agent produces [OUTPUT_TYPE] that, if wrong, causes [CONSEQUENCE]. I want a mandatory self-critique step before the output is finalized or any irreversible action runs.

TASK: Design and run the verification gate on the draft I will provide as [DRAFT].
1. First, restate the task's true success criteria in your own words.
2. Critique the draft against this checklist: correctness, completeness, adherence to constraints, safety/side-effects, and assumptions made without evidence.
3. For each issue found, rate severity (blocker/major/minor) and propose a concrete fix.
4. Decide a verdict: APPROVE, REVISE, or ABORT (with reason).
5. If REVISE, produce the corrected output.

OUTPUT FORMAT: A structured report: Success Criteria | Findings table (Issue | Severity | Fix) | Verdict | Revised Output (if any).

CONSTRAINTS: Be adversarial toward the draft, not charitable. Do not approve anything with an unresolved blocker. If evidence is missing to judge correctness, mark it as an explicit assumption rather than guessing.

DRAFT: [DRAFT]