Meta-prompting
Meta-prompting is using a model to write or improve a prompt rather than to do the task directly. You describe what you want, the model produces a better-structured prompt, and you run that against the actual work.
It works because models are better at recognising a complete specification than we are at writing one from scratch. Asked to improve a prompt, a model reliably adds the constraints and success criteria people leave out under time pressure.
The most valuable variant is asking what is missing. "What would you need to know to do this well?" surfaces the ambiguities you would otherwise discover in the output.
What this means in practice
Useful specifically before expensive operations: a large agent run, a full app generation in a builder, anything you would rather not do twice. Spending one cheap round on the prompt is much cheaper than one wasted round on the task.
The improver prompt
Here is a prompt I am about to run: --- [paste your draft prompt] --- Before I run it: what is ambiguous, what constraints are missing, and what would you need to know to do this well? Rewrite it with those gaps closed, keeping my intent unchanged. Then list any assumptions your rewrite still makes.
Run this before an expensive generation; the assumptions list at the end is the part people skip reading and shouldn't.
Prompts for the tools this applies to
Related terms
- Prompt template
A prompt template is a reusable prompt with placeholders for the parts that change — the file, the framework, the requirement — so a structure that works can be applied to many tasks without being rewritten each time..
- Spec-driven development
Spec-driven development means writing a precise specification before generating any code, then using that spec as the prompt.
- PRD for AI
A PRD for AI is a product requirements document written to be consumed by an AI coding tool rather than by a team.
- Plan mode
Plan mode makes an AI coding tool propose an approach before it edits anything.