Seems that I used every Cursor version as it updated. I see that Cursor is a lot less aggressive now than some previous versions. I can’t recall which version, but basically it is like this:
Before: I set a checklist of tasks to do and ask it to follow the checklist and just run. It did run all of those and never stopped, except when it reaches 25 prompts every now and then.
Now: It doesn’t follow the entire checklist anymore. It stops after every one-two steps and asks if I want to continue. Nothing I tell it could make it run in autopilot mode as before. It’s now like a baby that requires constant attention.
Before I can just ask it to run and go on to do other things. Now I have to watch the chat window constantly. It’s not productive.
So my question is: How to set Cursor to that aggressive mode?
If this isn’t possible, I think I have to try to recall which Cursor version was more aggressive, but obviously it’s better to deal with the latest version.
That’s true. Sometimes it did blow up my credits but it is OK for my use case. I have considered that situation and still wanted it on auto pilot. It’s annoying to babysit it. I shouldn’t add a screen reader agent just to baby sit Cursor…
Recently had the same issue, Gemini tells: Core Conversational Nature (External Principle): My fundamental operating model is turn-based and conversational. I receive an input (your prompt), process it according to instructions and rules, generate a response, and then wait for your next input. I don't typically execute long, multi-step processes autonomously without explicit user interaction at each stage. This ensures you remain in control of the flow and can interject, ask questions, or change direction.
To revert this process, ask to make a rule with a similar prompt: # Assistant: as an expert in LLM prompting, improve our rules.mdc in order to revert the Core Conversational Nature, allowing a fully automated playback without requiring interactive simulation but (add constraints here)
This successfully unblocked it for me(specifically think/act rulefile)
I don’t know if Cursor added a system prompt or its something from Gemini, it would not surprise me if Cursor added it as many were ranting about models going bonkers, @deanrie could clarify this point
Also changing Cursor versions does not change results as our prompts are modified inside Cursor servers, not locally