https://docs.all-hands.dev/ logo
Join Slack
Powered by
<Release v1.48.0>
# proj-agent
m
šŸ‘€ 2
c
Did anything unusual happened with this PR? https://github.com/OpenHands/software-agent-sdk/pull/4516 A check claims it wasnā€™t merged cleanly, as if, 
main
was pushed to or rebase to, forcefully, but that canā€™t be right?
m
don't know. let me give a look
I think it is correct. It was approved and then the bot fixed a thing. So the workflow is flagging that. Or ?
c
Noā€¦ maybe Iā€™m dense but the strange thing is, if I click on the last hash, in the comment, on every PR it shows what happened after approval. Example: this But on that PR, it shows ā€¦ uh this
Maybe itā€™s nothing, but I wonder why it happened
m
Ah yes now I see... Yes it is probably this
litellm_proxy/openai/gpt-5.5
integration tests are failing because of the key. @millions-crayon-32880 do you have access to litellm proxy? Can you check and/or replace the key?
I don't think we need to re-run the integration tests, but at least for enxt tim we do have all the test working šŸ˜‰
c
It must have been because of the subsequent merge+squashā€¦
I went through allā€¦. ish. Sigh. This thing doesnā€™t scale
do you mean both of these?
and these
l
We need this to deploy a low CVE image for a customer
Bot approved but it'd be good to get a review from a human too
c
I took a quick look, but I canā€™t assess this quickly: removal of curl, of wget, would seem to me to kneecap the agent. But Iā€™m not sure.
I have no problems if you want to try to move forward with it
(it might be that theyā€™re included transitively anyway šŸ¤·ā€ā™‚ļø)
l
These ones are probably safe to keep so I can maybe add those back, even though they're not strictly required to run the agent
m
@calm-artist-7206 No. I was speaking about Integration ones not examples. Examples were not done when I posted the msg. But I will look into them.
l
curl and wget increased high-priority CVEs by 10, so I think we can leave them out for now
@dry-book-58714 approved the image PR so Iā€™ll go ahead and merge that in. to be clear, this wonā€™t change behavior for most users, only for ones who intentionally choose binary-minimal
šŸ‘ 2
m
I will cherry pick the commit once is merged
šŸ™ 1
l
itā€™s merged!
m
About the examples which are failing: ā€¢ 
23_responses_reasoning.py
and `04_llm_specific_tools/01_gpt5_apply_patch_preset.py`: OpenAI credentials. These are the only examples using gpt models. So the same problem as for integratin tests. ā€¢ `02_remote_agent_server/17_convo_with_agent_sandbox_server.py`: this is a new example which cannot run because 
Error: ModuleNotFoundError: No module named 'k8s_agent_sandbox'
. It seems that this example need specific infrastructure. If it is ok for everyone I will add it to 
_EXCLUDED_EXAMPLES
list ā€¢ 
28_ask_agent_example.py
ask_agent() finds no ask-agent-llm in the registry, so it falls back to 
self.agent.llm.model_copy(..., deep=True)
. That copy crashes with 
TypeError: cannot pickle 'generator' object.
This example was also failing in the last release. This need a fix.
šŸ‘ 1
l
sg
c
What seems to be the problem with OpenAI models?
I ask because itā€™s a curious error, ā€œthis API key is not tied to the organizationā€ or something
oooh I didnā€™t realize it might be from OpenAI
l
we rotated the key but itā€™s not rotated on the eval proxy yet, iā€™ll fix the eval proxy one soon
šŸ‘ 1
šŸ™Œ 1
m
āœ… 1
I cherry-picked the three desired commits: the fix requested by Graham, and the two generated from my previous message above. I'm currently re-running the test examples, but I don't think we need to wait for the key rotation since we already have other integration tests (if everyone is okay with that).
šŸ‘ 1
l
Sounds good!
Letā€™s go ahead
m
Just to let you know: still three examples failing but because of the eval proxy keys. So I'm merging and releasing
šŸ™Œ 2
l
I think the eval proxy should be fixed fyi
šŸ‘ 1
gratitude thank you 1
3 Views
Open in Slack
PreviousNext
Powered by