It's useful for quick scaffolding and prototyping, but I sometimes get annoyed by not having the terminal separate, and sometimes it will start running a different server instance every few edits in and before I noticed that I was scratching my head at why nothing changed on the frontend until I realised I had to check localhost:5001, :5002, :5003, etc. :/
it will start running a different server instance every few edits
I think this an issue generally with the agent losing its orientation lately, I've had that happen in the last few days which is out of character... like it did mkdir -p to the same location it had been working in for like 10 responses in a row until i explicitly told it not to do that, wonder if a .cursorrule would fix it
I sometimes get annoyed by not having the terminal separate
I got a terminal in a standalone cursor window so it feels separate, do you mean for taskbars sake? or the way the composer is still a separate pane? On the latter I agree totally, the thing I linked in the above comment is optimal IMO but I haven't bothered to arse with any API subscriptions to use it with
like it did mkdir -p to the same location it had been working in for like 10 responses in a row until i explicitly told it not to do that
Yep, I've had weirdness like that too. But what I meant was if the agent would spawn a terminal as part of a response, it would also try do things like npm run dev over and over from response to response, almost like it doesn't really keep the terminal it spawned itself in context. I'd have to test using the in-editor-but-not-in-chat terminal pane more. I've just been using a regular external shell most of the time.
Welp, as we like to say around these parts, we're still early~
Yea I think that's exactly the same problem, I'm actually about to make a post on the forums because I can't get multiple repo's to work with the agent in a single workspace
All the code snippets and cheatsheets I've carried over from backup to backup, from cloud storage to cloud storage, like some unhinged hoarder... all that code can finally rest in peace 🥲
Most of my cheatsheets have become useless since the rise of the LLMs...
Yeah I imagine these "smart" terminals will obsolete them fully: https://www.warp.dev/
This looks cool. I could see myself using it...
The latest cursor agent has terminal windows now, haven't had time to get in a groove with it yet but it's promising
Enjoyed playing with this but performance wasn't quite worth it with locally run models https://github.com/BuilderIO/ai-shell
It's useful for quick scaffolding and prototyping, but I sometimes get annoyed by not having the terminal separate, and sometimes it will start running a different server instance every few edits in and before I noticed that I was scratching my head at why nothing changed on the frontend until I realised I had to check localhost:5001, :5002, :5003, etc. :/
It's getting there though
I think this an issue generally with the agent losing its orientation lately, I've had that happen in the last few days which is out of character... like it did
mkdir -pto the same location it had been working in for like 10 responses in a row until i explicitly told it not to do that, wonder if a .cursorrule would fix itI got a terminal in a standalone cursor window so it feels separate, do you mean for taskbars sake? or the way the composer is still a separate pane? On the latter I agree totally, the thing I linked in the above comment is optimal IMO but I haven't bothered to arse with any API subscriptions to use it with
Yep, I've had weirdness like that too. But what I meant was if the agent would spawn a terminal as part of a response, it would also try do things like
npm run devover and over from response to response, almost like it doesn't really keep the terminal it spawned itself in context. I'd have to test using the in-editor-but-not-in-chat terminal pane more. I've just been using a regular external shell most of the time.Welp, as we like to say around these parts, we're still early~
Yea I think that's exactly the same problem, I'm actually about to make a post on the forums because I can't get multiple repo's to work with the agent in a single workspace
All the code snippets and cheatsheets I've carried over from backup to backup, from cloud storage to cloud storage, like some unhinged hoarder... all that code can finally rest in peace 🥲