I find that most of my time working in Cursor in my react.js is just explaining to the model what I see in my backend Flask server (run with python3 app.py in a terminal) and my frontend (run with npm start in a terminal) as per what shows up in the browser’s debug console. I’m always just pasting info in from these two places and saying basically, “No, that didn’t work, this is still happening (paste) - try again”
Is there any way to make it so that agents can automatically make use of any of this? It would be like a dream if, for example, I told it to resolve a stacktrace on the backend, and it could try changes, start back up, and if it stacktraces again, automatically see the stacktrace and continue work.
I definitely agree with you, I’ve found I’m spending time on exactly the same things.
I’m actually working on an unofficial side project that attempts to solve some of that for the asynchronous front-end side of the problem. It should be available for beta testing in a few weeks with any luck.
I’ve concentrated on the front end side for now, as most of my Python work has been on apps that run their own self-tests on startup. The agent can call the script, and it’ll fail fast and hard if something’s wrong, with the results directly in the conversation, and the agent can iterate. But there’s definitely the long-running case there where piping later failures directly into (say) a Composer session could be useful. Watch this space.
