Agents Can Hang During Task Execution, Blocking Test Completion

[irvineoy]

When running certain tasks, the agent may encounter a hanging state, which prevents the tests from completing successfully. This issue may be related to several possible factors:

1. Bugs in the task implementation itself.
2. Residual background GPU processes that are not properly cleaned up and interfere with the current run.
3. Insufficient agent prompt design, which does not explicitly describe how to detect or handle such hanging scenarios in advance.