Please let me know what you think, about the release of VSCode extension. We should make a release this week, and we have two options • on github, make a separate extension release,

ext-v0.0.1

◦ the extension autodownloads it and autoinstalls ◦ I’m not sure if it messes up with the regular openhands releases, in some way (umm the docker builds? the release notifications?) • on pypi, we build the extension and include the build as a resource in the python library build ◦ the extension reads it and autoinstalls ◦ the build has 22KB, which seems very small, so it doesn’t seem an issue ◦ so it’s part of OH releases. (In the future, we may want to use the marketplace too, after people play with it for a while)

Maybe we can build the extension in the same CI we build our python packages, so it is released with our python library? 👀

So option 2, I like that! Yes, we can, and it’s already implemented I think. The poetry build uses

build_vscode.py

script to build the extension too.

Very excited about that 🙂 have been using the extension for a while and i like it

What’s the next thing you’re missing with it?

one thing i feel we should figure out is the ability to "@" multiple files. right now we do --file, but what it essentially do is dump the content, send it to llm and have llm ask back what do you want. I've been thinking if we can do smth cursor-style, i.e., tag multiple files, and then launch the CLI, we can add new instructions on top of those files, and then hit enter (so at least we save one LLM round-trip)

We can try that either: • before we start OH, which means we need some interface, some place where the user tags one file, a second file, removes the first, tags a third. • during OH run, which means we need to implement in CLI I think, as part of a user message during the agent run

i think we'll need to implement smth in vscode that allow the user to select file / ranges, and i think we can just modify the functionality of --files flag - instead of printing the file and send it to the agent, we should just include the file to the user message AND still ask the user for the prompt (which will get concatenated with the file context and send to the agent when user hits enter)?

OK, so we want two things: • let the user tag multiple files • let the user enter their own message with those files, not our default message

So the release will happen in the same workflow that publishes the pypi package? And so all I have to do is make sure that pypi workflow gets run as part of a release?

Yes

Yep! 🙌

Engel, thanks again for this wonderful extension!

Joe, thank you, that’s a bit too kind, this is just a small start. I think people should get the chance to play with it, give feedback, improve it, and then at some point maybe we could consider publishing it too. As of now, it autoinstalls once you run the CLI in a vscode terminal, which is the same way, I believe, that Claude Code autoinstalled. Claude Code kept it working like this, in a terminal, and improved it iteratively, for a few months, before publishing. Idk, I’m thinking we could do something like that?

That makes a lot of sense to me!

Right now, it builds it during

make build

, and as to the next release (tomorrow?), it should be uploaded to pypi

Hey Engel, not sure if this is my own user error, but when I use Openhands CLI in the VSCode terminal, I don't see it prompting to install the VSCode integration thinky spin Curious if there's a step I may be missing?

Could it be installed already?

Ive installed it manually in the past, and then uninstalled it. So now trying to see if the CLI will install it again 🙂

I suspect it has a flag in ~/.openhands/ that it's installed. 🤔 If you remove it, will it work? I have a PR on making this more reliable here: https://github.com/All-Hands-AI/OpenHands/pull/10307

ah OK, let me investigate!