Stop LLM execution when inappropriate words are input in VoicePipelineAgent

[Tanesan]

In the VoicePipelineAgent, I want to stop the LLM execution when inappropriate words are input.

While the text passed to the LLM can be adjusted using before_llm_cb, I am unable to stop the execution of the LLM entirely. Although passing an empty string to the LLM is possible, I believe the execution should be stopped in such cases, and this method feels nonsensical. Is there any other way to stop the LLM execution?