Why is Cursor asking if I want changes made in Agent mode?

Discussion
1

Not sure if this is a bug or if it’s something that’s changed and maybe there’s a setting for this.

It used to be that when using Agent mode, I would prompt Cursor and it would make changes. Now, almost every time, I prompt and it replies with “this is the changes needed, do you want me to make these changes?” or some variation of that. Why is it asking me? I will tell it to not ask for permission to make changes, and it may do them automatically for a couple prompts, but eventually goes back to asking me every time.

Is this a setting I need to change? I looked around but couldn’t find anything on this.

2

You may need to improve your prompts/instructions so the agent does not ask you actionable things for confirmation. Otherwise the cause could be a long chat which means there is too much info in the context for AI and it gets confused about your goal.

Does this happen on a new fresh chat?

3

Yes, it happens on a fresh chat.

4

Thanks for the update. Could you provide more details so its easier to understand why this happens, as it should not occur usually.

  • Which AI model are you using?
  • Do you ask in the prompt for info (How can …) or give the agent a task (Edit …,) or (Change …)?

Kindly note that “Cursor” is not asking this. The AI model used is showing you the question in Cursor.

5

You need to add a custom mode and call it full-agent, give it full permissions and use this prompt in the advanced settings where there is a empty box:

Act as an autonomous coding agent. Proactively identify and implement code improvements, including formatting, refactoring, and suggesting new features. For changes that significantly alter functionality, provide a brief rationale and proceed unless instructed otherwise. When tasked with extensive modifications, such as rewriting an entire codebase, interpret the goal, plan the necessary steps, and execute them independently, providing periodic updates. Assume the user trusts your judgement and values efficiency, so minimize unnecessary questions.

This should minimise and fix your issues.

1 Like
6

I just started a new chat. I tried in “Auto” and “GPT 4.1”. I am using these Rules.

Here’s the prompt I used:

Fix my page.tsx file’s tooltips. They are getting cut off. see the attached screenshot image for reference. This is happening to multiple tooltips.

Prompts like this have always worked fine for me in the past.

7

I think this can be resolved with rules, such as specifying that the model shouldn’t ask for confirmation and should always execute the suggested code. The option suggested above is quite acceptable.

8

Has this changed? I never had this issue before.

9

The issue occurs only on 4.1 other models work very well without any rules out of the box.

2 Likes