Decompose prompt editor into sections based on typical or popular prompting frameworks
Instead of one big text editor box for the prompt, consider having a customizable start point where you enter each section of the prompt according to a best practice framework. In practice this could be an individual text field for each major section of the prompt with a heading and description reminding the user of what should go into that section.Example: 1. Role - assist the model in accessing the right semantic space by telling it what identity it should adopt.2. Goal - Establish the shared goal between prompt writer and the model3. Context - identified domains or knowledge areas or reference websites or reference repositories or documents, any and all sources of information that the model should consider or explicitly ignore.4. Output format - desired headings content types and modes (E.g. multimedia vs plain text), use of tables, nested bullet point lists, numbered lists etc.5. Before executing, ask me for feedback on what could make this prompt better (this could just be a checkbox with standard boilerplate language thrown in to ensure the model does this recursive check. It's a general purpose element of the prompt so you wouldn't even have to customize it, just tack it on the end of any prompt).
I totally hear what you are saying, but not everybody stores prompts like that. I might consider having templates in the future that follow this and make it a multi-step process, but I don't want to have it locked into that method if users want flexibility.
But definitely good points and something to consider.
2
Chandzo•8/16/2025
Oh no, I wouldn't lock anyone in (sometimes a simple natural language prompt is best 👍), but I think a template approach could be useful to help people follow a structure/framework. Could this be achieved with the {{variables}} already available? Is there a character limit on those?
I could see it being the first thing you decide when you create a new prompt. Is this a "simple" prompt? A structured prompt? I think most of the lay user audience (who use models the same way they used to use Google search) aren't even aware that structure prompts can give them better results. But I acknowledge that that level of complexity isn't necessary, it's just helpful when you're trying to go deep and make your results repeatable.