Meta-Prompt

A meta-prompt is a prompt whose output is another prompt. For example, you might ask a model to take a rough instruction — “help users with their code” — and expand it into a well-structured system prompt with clear guidelines, edge case handling, and examples. Meta-prompting is useful for scaling prompt development: rather than a human writing every variant by hand, a model can generate a range of candidate prompts that a human then reviews and refines. It’s also used in automated optimization workflows where a model iteratively improves a prompt based on evaluation feedback. For behavior architects, meta-prompting is a productivity multiplier — it makes it feasible to explore more design options than would be practical to write manually.