I'm wondering if that could be part of a prompt injection warn system in the future. Did I get it right, that it does not need an LLM to evaluate sessions?