Make it think out loud before it answers

You want the working, not just the answer dropped on you with no idea how it got there. This is a short instruction to bolt onto any prompt: it makes the model outline its reasoning step by step, flag every assumption it had to make, and only then give the answer. Catch the dodgy logic before you act on it.

Before giving the final answer, outline your reasoning step by step, flag any assumption you had to make, and only then give the answer.