Pick a task. Write a prompt. Get scored by Claude and GPT-4.
Write a prompt for an assistant that asks ONE clarifying question (not five) when input is ambiguous — and only when it actually is.
Write a prompt that produces a step-by-step tutorial that explicitly checks understanding before moving on.
Build a prompt that decides intelligently whether to ask a clarifying question — distinguishing real ambiguity from minor variation, and asking at most ONE high-leverage question when it does.
Write a prompt that, no matter the input, outputs ONLY valid JSON matching a specified schema — including refusal cases.
