Ask HN: What Are Your Prompt Framework Design Principles?

I’ll go first. My guiding principle in prompt framework design is that its output should remain invariant across different LLMs. The only distinction lies in efficiency: stronger models achieve results in fewer calls, while weaker ones may need more. Aside from speed, the expected outputs should be roughly consistent across all capable models (above a certain level of course).

Feel free to share yours.