Context / Research term
Meta-prompting
Using one prompt to create, analyze, select, or improve prompts or workflow instructions for another task.
A meta-prompt operates on the instruction process itself. It can ask a model to inspect task examples, identify missing information, draft candidate prompts, compare them on an evaluation set, or extract a repeated procedure into a Skill. The generated instruction is a candidate artifact. It still needs evidence against the target task, because fluent prompt analysis does not guarantee better downstream results.
Builder example
Meta-prompting helps when the instruction design itself is repeated or underspecified. It can organize examples and propose variants faster than manual drafting, while held-out cases decide which variant works. Broad standing behavior belongs in durable instructions; that is related but not identical to meta-prompting.
Common confusion: Prompt engineering designs instructions for a task. Meta-prompting makes prompt or procedure design the task. System instructions and preferences may be outputs of that process, but they are not automatically meta-prompts themselves.