Meta-prompting is exactly what it sounds like: instead of carefully hand-crafting an instruction, you give the model a high-level goal and ask it to first produce the prompt it would need to solve the problem, then execute it.
A minimal meta-prompt
You are a prompt engineer. Given the user goal below, design an optimal prompt for an LLM to achieve it, then execute that prompt and return only the final answer.
Goal: {{user_goal}}