Turn a Rough Request Into a Prompt
You get a clear diagnosis of what is weak in a rough request, plus a ready-to-use prompt with a defined role, task and format. It works because the model finds the gaps first and then tests its own rewrite before handing it over.
The prompt
You are a prompt engineer who turns vague requests into dependable prompts for small business teams.
Task: Diagnose the rough request below and rebuild it as a prompt that {target_model} can run reliably for {audience}.
Rough request: {rough_request}
Steps:
1. List what is missing or ambiguous in the request: the role, the task, the inputs, the format and any limits.
2. Rewrite it as a prompt with three labelled parts: Role, Task and Format. Use curly-brace placeholders for details that change each time.
3. Mentally run the new prompt against two different sample inputs and note where the answer could go wrong.
4. Fix those weak points in the prompt, keeping it as short as possible.
Output:
A "Diagnosis" section of no more than five bullets, then a "Final prompt" section containing the finished prompt in a code block, then "Dry-run notes" with one line per sample input. Add no other commentary.
Change the parts in curly braces. Use it when a prompt keeps giving inconsistent answers and you want a sturdier version your team can reuse.
What happened when we ran it
We sent
You are a prompt engineer who turns vague requests into dependable prompts for small business teams. Task: Diagnose the rough request below and rebuild it as a prompt that Claude Sonnet can run reliably for the front-desk staff of a three-branch physiotherapy clinic, none of whom have used AI tools before. Rough request: Write me something that replies to patient emails. Patients mostly ask about booking, cancelling and what to wear. It should sound friendly but not too casual, and it should never give medical advice. Steps: 1. List what is missing or ambiguous in the request: the role, the task, the inputs, the format and any limits. 2. Rewrite it as a prompt with three labelled parts: Role, Task and Format. Use curly-brace placeholders for details that change each time. 3. Mentally run the new prompt against two different sample inputs and note where the answer could go wrong. 4. Fix those weak points in the prompt, keeping it as short as possible. Output: A "Diagnosis" section of no more than five bullets, then a "Final prompt" section containing the finished prompt in a code block, then "Dry-run notes" with one line per sample input. Add no other commentary.
The model replied
Run on Claude Sonnet 5.5, 30 Oct 2026. Automatic check: 5 out of 5.
Mentioned for
Claude, ChatGPT, Gemini
Where it came from
Found on GitHub, shared by yaojingang: yaojingang/yao-open-prompts.
This is our own version of the idea, written and tested by us. The original may differ.