Is there a way to get messages actually sent to the llm from when calling a BAML function? For example:

function UserResponseOrCoder(messages: LLMMessage[]...)

If this function has some logic that injects messages condidionally, I would like to be able to update my context with those messages. In effect, I want to persist the context after BAML code injects messages into the context (or even deletes some messages). Is this possible? Or advisable?