If you’re talking about Gemini responding, using a tool, and then responding to prompt again, I think that’s just a feature of Gemini. I also thought it was a bug, but in fact he just doesn’t get enough new information after the request.
Sometimes he answers right after the prompt, then after the first too call, and then again at the end of the work.
There is no way that is a feature because it is just repeating what has been said in the first part of the response with no new information. It is also wasting output tokens and driving up the costs + poisoning the context window.
Gemini has just given me an answer to the questions from the old request at the beginning of the response to the new request. This is most likely how Gemini works with its context window, not how Cursor works with it.