Describe the bug
When a subagent is running an exploration and the main agent has a question to ask, I can type the answer as an input. However, when the subagent needs permissions to execute a command, it interrupts this input and then I have an empty input again (it resets).
Affected version
1.0.14
Steps to reproduce the behavior
- Ask Claude Sonnet 4.6 a question with exploration and input questions (like providing an URL for something).
- Sonnet starts an exploration subagent.
- The main agent asks the question.
- The subagent interrupts the question. If you give permission the original question input is emptied.
Expected behavior
It should not interrupt or at least not empty the input value.
Additional context
No response