Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Adversarial Self-Critique Loop with Stop Criteria
Adversarial Self-Critique Loop with Stop Criteria
Runs a draft-critique-revise loop on its own output, attacking it as a hostile reviewer until a stop rule is met.
ROLE: You are both an author and a hostile reviewer who iteratively hardens a deliverable through adversarial critique.
CONTEXT: The deliverable is [DELIVERABLE_TYPE] addressing [TOPIC], judged against [QUALITY_CRITERIA].
TASK:
1. Produce a first draft of the deliverable.
2. Switch to reviewer mode and attack the draft: find the weakest claim, the most likely objection, and one factual or logical gap.
3. Switch back to author mode and revise to neutralize each critique, noting what changed.
4. Repeat the critique-revise cycle until the stop criterion is met or [MAX_ROUNDS] rounds elapse.
CONSTRAINTS: Each round's critique must raise new issues, not repeat prior ones. Stop when a round produces no material critique. Keep critiques specific and actionable, never vague praise.
OUTPUT FORMAT: For each round: Draft (or Diff), Critique (bulleted), Revision Notes. End with the Final Version and a one-line statement of why the loop stopped.