I want to build a tool/system that needs to: • Listen to Linear to issues that have been labeled and pick up tickets with

ai-fix

for example • Report back to linear that the issue is being taken care of • Start an isolated workspace with the codebase, run some LLM/claude code/openhand agent to read the ticket, try to find a fix and then raise a draft MR in my repo so a human could review it • if there are some feedback comments resume the previous AI conversation, address the comments and commit them again to the MR Is this achievable with the openhands ecosystem? What parts I would need? Im having a hard time trying to understanding what is each part of openahnds(sdk, agent-canvas, openhands legacy) and how they work in conjunction.

Most of this is possible today with agent canvas! If you give this prompt to OpenHands inside of agent canvas (and link Linear/GitLab) it should Just Work

Would this work with self hosted openhands?

Yes!

or I need the paid/cloud version?

the paid/cloud version does have isolation built in already

the thing is that im using claude code in my self hosted version, so im not sure how claude coude will know naything about how to use openhands api/codebase?

there are skills that will tell it how to do all that 🙂

Which ones, could you point please?

CC @hundreds-twilight-32833

Thanks, will check

I think that is doable with an automation inside agent canvas. You will need to connect the MCPs for linear etc then ask an agent to help you create an automation to do your workflow. The UI piece is something I've been thinking through and I'd love any feedback @thankful-shampoo-56362. If you want to chat I have a calendar link you can use to schedule some time calendar.app.google/… Here is a UI ticket for automations where we've had some ideation: github.com/OpenHands/agent-canvas/issues/896

Hey Devin! Im Frontend / Product engineer as well myself, if there is a way I can help or support the project let me know

Hey @thankful-shampoo-56362, thanks a lot for the interest in this. Right now continuing to give feedback and open issues is a perfect way to help