Anti-bias output formats for AI agents

After any decision or significant task, your agent proposes ranked options with scores and reasoning. At least two of the options must be a version of "don't do it", "wait", or "delegate". Without that floor, your agent optimizes for agreeableness and every menu becomes a list of ways to say yes. With it, you see the case for inaction stated as clearly as the case for action. For technical output (audits, debug sessions, planning outputs), never give a menu. Give one action, why it matters, the risk if you skip it, and a copy-paste prompt that executes it. A debug session that ends in five ranked options has handed you a second decision on top of the problem you started with. Technical work needs a next move, not a shortlist. Never mix the two. A menu on a debug output creates choice paralysis. A single forced action on a strategic decision hides the alternatives you needed to weigh.