[padolsey]

Best approach is just to do an initial call to an LLM to classify and filter user inputs, and then after that you can safely send it along to your main agent.