Refine rough prompts into structured, model-specific prompts with a guided optimizer. Keep them in your own library and share the ones that work.