Hey :wave: first version of the command `/goal` ...
# proj-agent-canvas
m
Hey šŸ‘‹ first version of the command
/goal
https://github.com/OpenHands/agent-canvas/pull/1492 you can use it as follows: •
/goal <objective>
, or •
/goal --max <n_of_iterations> <objective>
For instance, the judge is always the LLM of the main agent. This can be changed in the future, i.e., you can chose the profile you want to use as judge. But I didn't now because we are migrating to agent profiles. PS: It probably needs a little tuning. Therefore, testing it and trying to break it is highly appreciated! šŸ™‚
šŸ™Œ 4
c
This is so cool!
h
This looks great so far! šŸŽ‰
m
@hundreds-twilight-32833 Could you test it?
h
Sure! I can test today @modern-fireman-99755
Finally got some time to test this fully. Really fun to use! I'm trying to let minimax create a light theme for agent canvas and its just chugging along. The only hiccup I had in testing was switching to the branch killed my automations server when I restarted. Probably unrelated but I wanted to call it out. Once this loop finishes I'll try some with --max and make a note on the PR šŸ˜„
Ran into an issue resuming running conversations. Made a video on the ticket
m
Hey @hundreds-twilight-32833 I cannot reproduce your bug. The only thing that I see when I change of discussion and I come back is that it takes a moment to reload the messages. (I'm fixing this)
I fixed the issue that you saw in the video. Now it should be better. Could you re-test?
h
Oh maybe it was a timing thing then. Sure I can retest!
m
thank very much... if it makes the same error: could you also share the trajectory? Perhaps I have more info on what is going on
šŸ‘ 1
h
Done. No longer seeing that same issue āœ…
m
thank you for your help šŸ™‚
you can accept the PR if you feel comfortable (no pressure šŸ˜‰ )
āœ… 1