Sometimes I’m really sloppy and am really unspecific with my prompts. Can we have a feature that auto generates potential prompt improvements based what it thinks I want and I can select the prompt improvement option from 2 or 3 possibilities before sending it off to the manual or agent mode.
Will be really cool for it to be context aware of my past chats or files I’m working on at the moment and enhance or structure my prompts in a way that’s better for LLMs to tackle.
For example: I load into context a table and filters react component and ask it to “please fix any race conditions errors when applying filters” can enhance it to specify “Please analyze the table and filters React component for race condition issues that may occur when: 1) multiple filter changes happen rapidly in succession, 2) async data fetching overlaps with new filter requests, or 3) component unmounting during filter operations. Provide specific fixes using proper state management, debouncing, or cleanup patterns.”
I’m also sure this generation of multiple versions of prompts and feedback on which prompt patterns people choose can provide some human reinforced data to collect and improve system prompts under the hood.