I was waiting for it so it would be awesome. We might need to put AutoCodeRover and SWE-Agent inside.

Hi @foragerr , I think you started working on this at some point, right? I was looking for the PR that you opened but wasn't able to find it.

@neubig Instructions here: #198 (comment)
tl;dr : add a label dogfood-this to any issue in this repo, it will kick off an Actions workflow to run the Monologue agent and attempt to raise a PR.

PR here: #803

I don't recall if @rbren got around to adding an openAI key into repo secrets.

Ah, that's great. It'd also be nice if we could achieve the UX that I posted above though, just say through the normal UI "fix this issue" linked to a GitHub issue and have opendevin fix it.

Do you know if web browsing (via playwright?) works now?
Being able to access web content is necessary for the workflow you're describing.

Thanks @foragerr, it probably is not, in which case it will need to be fixed. I actually suspect that we'll have to open/close several sub-issues to make this workflow work. Please feel free to open an issue and link it here! (or I can do it later)

I'm looking into this.

@foragerr FYI: https://github.com/OpenDevin/OpenDevin/actions/runs/8685542945/job/23815222805

Looks like we'll need to figure out the sandbox. I know in CircleCI there's a way to access a docker daemon, not sure about GitHub

GiHub actions does docker, SSHBox is new since I last worked on this, there maybe some networking differences preventing the SSH connection.

I'll take a look.

Splitting issue with existing workflow into a separate issue: #1150

This seems to currently be blocked by our browser not working: #1276

The first step to resolving github issues is understanding them, and if OpenDevin cannot browse to issues, then it can't fix them.

@neubig could we use the GitHub API? That might be easier (for the agent) than using browse

Yeah, I think that is a better option