Prompt Test Set Builder
Builds a small test set for a prompt you keep fixing, with the cases split so the ones you tune on are not the ones you judge by.
You fill in paste the prompt and one…paste them, or describe them
Reads your prompt as a newcomer who can't ask questions and lists every gap a model would fill by guessing: empty input, clashing rules, vague words, no picture of the finished answer. Use it when a prompt gives different results from run to run, before you blame the model.
Reads your prompt as a newcomer who can't ask questions and lists every gap a model would fill by guessing: empty input, clashing rules, vague words, no picture of the finished answer.
I want you to act as a new colleague who has been handed my prompt and can't ask me anything about it. Don't run the prompt. Read it and find every place where you'd have to guess. What the prompt is for: [one sentence on its job] Model it runs on: [ChatGPT, Claude, Gemini or another] Answer each question from the prompt text alone. Quote the line that answers it, or write "not covered". 1. What should happen when the input is empty, partial or off-topic? 2. Which instructions could pull against each other, and which one wins? 3. Which words could be read two ways, such as short, simple, detailed or professional? Give both readings. 4. What does a finished answer look like: length, format, what to leave out? 5. When should the model stop and ask instead of carrying on? Then, for every "not covered" and every word with two readings, suggest one line I could add to the prompt. Keep my wording everywhere else, and don't rewrite the whole prompt. My prompt: [paste your prompt]
Builds a small test set for a prompt you keep fixing, with the cases split so the ones you tune on are not the ones you judge by.
You fill in paste the prompt and one…paste them, or describe them
Paste a prompt written for an older AI model and get it rewritten for today's models: workarounds removed, real requirements kept, changes listed.
You fill in paste the prompte.g. ChatGPT, Claude, Gemini
Paste a piece of text you'd like more of and the model works backwards: it names what makes it work, writes a reusable prompt that produces it…
You fill in for example product…ChatGPT, Claude, Gemini or…placeholders+2 more
Before writing tests, the model lists every place the spec or code is ambiguous, and for each expected value it says whether it came from the spec…
You fill in for example Python + pytest…one sentencepaste code+1 more
One email a week: the best new prompts and one short guide. Confirm by email and get the free Prompt Cheat Sheet (PDF). Unsubscribe any time. Privacy policy