You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We got something like the following -- This status overview should ideally just be a summary of changes & LLM's prediction of the task success, it should NOT say it require external human reviewer
Status Overview:
Pending Issues:
Right side panels extending beyond browser window
Terminal panel resizing problems
Chat panel max-width adjustment to 50%
Current State:
While changes have been implemented, there is no reviewer confirmation that the fixes are working as intended. Multiple previous attempts to resolve these issues were unsuccessful.
Recommendation:
Await reviewer verification before considering these issues closed. Visual confirmation is needed to ensure:
I've been trying to add openhands-resolver.yml github action workflow on my repository.
It manages to work on tasks and create a branch but always fails to make it as a PR. openhands.resolver.send_pull_request returns something like below
This pull request fixes #XX.
While the AI agent claims to have completed the task, there is no concrete evidence provided in the message thread of actual code changes or a Pull Request being created and merged. The agent only describes what they claim to have done, but without seeing
We cannot verify that the issue has been successfully resolved. The agent's message only describes intended or claimed actions, but doesn't provide proof of implementation. To consider this resolved, we would need to see the actual code changes in a Pull Request and confirmation that the changes were successfully merged into the codebase.
A proper resolution would require evidence of the actual implementation and successful merge of these changes.
We got something like the following -- This status overview should ideally just be a summary of changes & LLM's prediction of the task success, it should NOT say it require external human reviewer
We likely need a prompt change to tell LLM do not rely on human for this judgment.
We should add a line in prompt files from this folder:
openhands/resolver/prompts/guess_success
The text was updated successfully, but these errors were encountered: