This seems very interesting, UX-wise! I first th...
# proj-automation
c
This seems very interesting, UX-wise! I first thought what they call subscription is maybe not necessary, since the agent can wait and babysit/iterate on a PR. But babysitting ends anyway, and reacting to new changes requires another human to prompt, for OpenHands… On the other hand, I’ve experimented with my agent to use a single conversation for all notifications on the same PR/issue and I liked it! (while OH uses a different conversation for the same PR) Maybe we can consider something like this, where the agent subscribes to be notified in the same convo of interesting PR changes? https://x.com/zeeg/status/2077832838150553787?s=61
l
Interesting! I wonder if we can already do this through automations? In other words: 1. create an automation that has webhook for all activity on a repo 2. when it gets activity, it checks if any conversation is associated with that PR and sends a “push” to that conversation
@nutritious-balloon-65098 spec out what this might look like looking at the OpenHands/automations, OpenHands/software-agent-sdk, and OpenHands/extensions repositories. Ideally we’d have all the machinery implemented already in automations and sdk, but if we don’t tell me what’s missing.
c
I think your agent may have got stuck on the LLM, because litellm gets the wrong API key; I understand Saurya fixed this, but it’s probably not deployed yet. If that is so, I found that manually switching LLM to some other profile and back, works to continue.
l
No, it actually got stuck on this facepalm
😅 1
@nutritious-balloon-65098 spec out what this might look like looking at the OpenHands/automations, OpenHands/software-agent-sdk, and OpenHands/extensions repositories. Ideally we’d have all the machinery implemented already in automations and sdk, but if we don’t tell me what’s missing.
n
I'm on it! Graham Neubig can track my progress here.
OpenHands encountered an error: The request to the Slack API failed. (url: <https://slack.com/api/chat.postMessage) The|slack.com/api/chat.postMessage)_The> server responded with: {'ok': False, 'error': 'msg_too_long'} See the conversation for more information.
😅 1
l
gpt-5.5 got a bit verbose 🙂
c
5.5 ?
why not 5.6
l
I don’t actually use the cloud as much now, I mostly work on my own VM using the agent canvas
But I’ll update to 5.6
c
wow 27 pages! 😅