A user spends significant time structuring product ideas and implementation steps for prompts before pasting them into Cursor, often using external tools for diagrams like Mermaid. They need integrated capabilities within Cursor to support this planning and visualization process, streamlining their prompt engineering workflow.
It's not unusual for me to spend hours crafting a prompt in a text file before I ever paste it into Cursor. What takes time isn't wordsmithing. I'm not trying to craft the perfect way to say "You are a senior developer" or whatever (I actually never include those kinds of things in my prompts). What I'm putting thought into is the big picture structure of the product itself ("how will this thing work, what is the goal, what are the inputs and outputs, what are the relationships") and the step by step breakdown of how to implement it. A lot of that time is spent doing things like drawing diagrams in a notebook so that I can understand and explain how different pieces connect. I'll sometimes then generate a Mermaid diagram to share along with the prompt if it seems helpful. It's surprising to me when I read people saying they vibe coded a working app in three hours. I mean, that's awesome! But that's not how it generally works for me. If you're struggling to get good results out of tools like Cursor, my top recommendations are: (1) make sure you're using a reasoning model like gemini-2.5-pro or claude-4-sonnet, (2) don't ever make more than one request in the same conversation — once one thing is done, start a new chat, and (3) put the same level of thought into writing your prompt that would writing a Jira ticket for a human engineer (which is, uh, hopefully quite a bit of thought).