Fix regression from PR 248

[humphd]

In #248 I caused a regression, by getting the messages before adding the current prompt from the user. I need to do it after.

I've updated this code so that it separates the logic for getting the messages for checking if the last one was a Human Message from the part where it sends the new messages.

NOTE: by default, all options to .messages({...}) are true, so I'm removing unnecessary true values in this code to simplify.

To test this, you should be able to get the LLM to respond to a query. Without it, it will ignore what you just wrote.