https://docs.all-hands.dev/ logo
Join Slack
Powered by
I'm seeing "`Unknown event`" in OpenHands cloud an...
# proj-agent
e
I'm seeing "`Unknown event`" in OpenHands cloud anytime a file is about to be read (e.g. "`Reading README.md`"). No problems after the file is read and the contents are returned, (e.g. "`Read README.md`") Known issue? Or should I open an issue with more details?
c
I'm guessing maybe it's a visual weird thing, not affecting behavior. Not sure if it's been reported before though. @nutritious-balloon-65098 search the open issues for this, and if you can't find it, investigate where it's coming from and make a new issue on it. It's a bit disconcerting from a UI/UX perspective.
n
I'm on it! can track my progress here.
## Summary I investigated and reported the "Unknown event" UI issue you observed in OpenHands Cloud. ### What I Did 1. Searched existing issues: No matching issues were found for "Unknown event" in the OpenHands repository. 2. Investigated the source: Traced the "Unknown event" message to the 
EVENT$UNKNOWN_EVENT
i18n translation key, which is displayed as a fallback in 
frontend/src/components/v1/chat/event-content-helpers/get-event-content.tsx
when an event fails type guard validation. 3. Created a new GitHub issue: [#12865](https://github.com/OpenHands/OpenHands/issues/12865) - "Unknown event" shown momentarily before file read operations complete ### Key Findings The issue occurs because: - Events pass initial V1 validation (
isV1Event
) but fail more specific type guards (
isActionEvent
) - The SDK's 
ActionEvent
allows 
action: None
, which would fail the type guard - The fallback 
title || i18n.t("EVENT$UNKNOWN_EVENT")
triggers when title is an empty string ### No Code Changes I did not make any code changes to the repository - this was purely an investigation and issue creation task. The issue is now documented for the maintainers to address.
👍🏼 1
e
Another UI issue.. may or may not actually affect functionality: there is a Pydantic error apparently related to a 
pending
status for the task list udpates:
Copy code
Input should be 'todo', 'in_progress' or 'done' [type=literal_error, input_value='pending', input_type=str]
c
Did the agent correct itself next steps, for this one?
It looks like the LLM sent a ‘plan’, and got an attribute slightly wrong, ‘pending’ task doesn’t exist. Normal behavior would then be, we send it this error, it sees it, and corrects itself next time
e
I think it corrected or gracefully failed and then recovered / learned, since it eventually was able to mark all tasks as completed.
👍 1
c
I think it’s normal, the LLM makes a blunder sometimes, but it corrects itself
👍🏼 1
e
Got it. Just wanted to make sure the "pending" was an LLM oops instead some code/config error
nod 1
ty 1
b
I also notice this from time to time when using the web app but I did not notice this in cli, why is it ? Even if it self correct does it mean more token consumption ?
Open in Slack
PreviousNext
Powered by