[M3 Bug] ALLALLALL tags generation

[BEEugene]

Which inference path did you use?

MiniMax API

Inference parameters

No response

Prompt / input

When completing the task with the subagent, M3 reports that ALL tasks have been completed. Sometimes, this may lead to ALLALLALL... tags repeated thinking.

Expected behavior

The model decides that the task has been completed and closes the session.

Actual behavior

2)

Additional context

No response