Build self-critique into a single prompt: draft, list your own weaknesses by dimension, then revise (2–3 passes)

Lushbinary·medium signal

Structure one prompt so the model produces an initial output, then systematically identifies its own weaknesses across named dimensions (accuracy, completeness, tone, edge cases), then rewrites addressing them — looping two or three times. This effectively drafts, reviews, and edits inside a single call, capturing much of a multi-agent critic's value without the orchestration overhead. It works best when you supply explicit rubric dimensions so the self-critique targets what you actually care about instead of vague 'make it better' passes.