[llm_tool] Need to refactor Tool consent config

We need to define a proper way to handle the security of tool usage and user consent before executing a tool. We currently have a model that inserts a system prompt for tools requiring user consent. We need to explore what is the best practice in this kind of use case.