Does not take action

Pro user. Cursor on Mac : Version: 0.49.6 (Universal) : VSCode Version: 1.96.2

Seems to be rather standoff-ish now. It used to immediately suggest code changes then ask me to accept them, now it just gives me advice on what I need to change then I need to follow up with details on how to fix a problem.

1 Like

Hey, which model are you using?

1 Like

defaults.

Does this issue also occur in other projects? Have you tried creating a new chat?

Yes, i’ve created several new chats. it seems to err toward suggesting changes rather than actually doing them. IDK why
Same in other projects

Are you in agent mode? Do you use any rules in your project?

I have a similar issue. I get a lot of errors about not being to search my codebase and then that leads to suggesting codes changes as well.

yes i’m in agent mode. i have only one rule which tells ai to always use a particular script to start up the servers. that’s it.

I had the same problem today after the update.
I uninstalled Cursor, cleared the cache and reinstalled Cursor, then everything worked again.

Yep, same issue here.

Agent is no longer an agent. It just suggests what I should do and I have to tell it to please find the files and complete the code for me. Sometimes after telling agent to do that, it will then search and find the files but then ask me again, are these the files you want me to apply changes to?

YES!

And so on.

1 Like

I’m having the same issue here. The Agents also asked me to update code in other tools and when I ask it for the script it tells me that it provided the script but doesn’t actually provide it.

Lots of strange things happening over the past couple of days

Same issues. Telling him to put a plan into a file. In it’s last paragraph it tells: Let me create an updated plan … and stops. Tried this 5 times with differnt prompts - the same - no action!

I have the same issue. It’s Auto mode which causes it . Auto just makes the Agent into Ask mode.

Agent will work again if you explicitly choose a model, such as Claude 3.7
with a model selected the Agent will enact changes ( and will stop Cursor from charging me money for a code suggestion I could have got for free !!!)

This is indeed what’s happening. The other models don’t take action. You should use Claude for it to work properly. The only issue is that those models take forever to do something, which makes you lose a lot of time anyway. I don’t know why I’m paying for Pro. The Anthropic models take forever to answer, and instead of fixing it, they thought, “Let’s give Cursor Pro away for free to students, that will lighten the load?” What a (excuse me) retarded move.

I’m wondering if the problem is me. As the complexity of my app has increased from green field, Cursor seems to have become more likely to suggest I make changes rather than just doing them for me. I think maybe my prompting style has shifted from being directive “create a startup script which first checks the servers are not already running” to pointing out issues in the negative “the startup script is not working”. This would mean it has more work to do to infer what I want to happen. I’m not sure this is the problem as it used to be good with me vaguely suggesting things but I’m going to try and see if there’s any difference.