Allow Positron Assistant to execute terminal commands?

[donboyd5]

I had switched from Positron to Cursor because I wanted to try Cursor's AI features. I missed Positron's ease-of-use and its usefulness for data and analytic projects so I am trying to switch back. Now I am missing Cursor's agentic features.

As far as I can tell, there is no way (yet) to give PA a list of tasks that include terminal commands (esp. for git) that tell it to commit changes, create a new branch, implement a plan to enhance a quarto project (it can do this as a standalone, of course), render that project (via terminal: quarto render), push to GitHub, etc.. It would be really nice to be able to create a long sequence of tasks that involve editing code, committing, rendering, examining, etc. This seems pretty straightforward in Cursor.

If I am missing how to do this, I would appreciate a pointer. If not, I would love to see it as a planned future feature. It would be great to have the awesome UI/data/analysis features of Positron combined with a lot of the agentic features of Cursor.

Many thanks for considering this.

[testlabauto]

Hi @donboyd5,

Thank you for your feedback! Would you perhaps be open to looking over these two topics and letting us know if this approach would meet your needs?

https://positron.posit.co/assistant-chat-commands-participants.html#custom-slash-commands
https://positron.posit.co/assistant-chat-instructions.html

[donboyd5]

Thanks, @testlabauto

I didn't know of these topics. I'll try them out over the next couple of days and report back. On their face they don't look as fast or easy as what Cursor does but it looks like they may be pretty powerful. I will definitely check them out and let you know how tney work for me.

[donboyd5]

One slightly related comment:

It seems that in ask mode, chat thinks it's more powerful than it is. First, it thought it had terminal access because it gave me what it thought were working instructions to enable access, and thought that subsequent instructions it was offering were actually being executed when they weren't. Second, it thought that after we had discussed a plan in ask mode (my bad for not using plan mode) it could go ahead and implement it. But it couldn't. So it offered lots of code that it couldn't implement and asked me to edit files, apparently not realizing it had another mode that would allow it to edit. Positron probably is not the culprit here; my general experience with AI bots is that they are not very self-aware - they don't know what they can and cannot do. In any event, it can be a little frustrating to provide guidance, be told that the bot can implement it, ask it to implement it and walk away, and then find out it couldn't do what it wanted. But I'm learning not to ask it to do things when I'm in the wrong mode.

[donboyd5]

I see that I was at least partially wrong about terminal commands. Even though I couldn't figure out how to let agent run them, it is asking for my approval for some commands, which I am granting. And I see that on a command by command basis I can grant authority in the future to execute commands without pre-approval (I think). So maybe it has more capabilities than I thought. Will say more if/as I learn more.