Making agent stop and ask for permission to proceed

Bug Reports
1

I’ve been trying all day now. What I want is pretty simple: I want the AI to make a plan and then ask me for approval or adaptions and then continue. But it just can’t stop… ever - at least in agent mode.

This is my main instruction in .cursor/rules/instructions.mdc with globs *

The AI should:
- Always address me with Milord at the beginning of each message! No exceptions! (Comment: this is just to debug that its using the rule)
- First have a look at @general.mdc, @general-architecture.mdc and all related files as described below that might be relevant to the task at hand
- create a list of very specific and clear requirements subdivided by functional requirements and technical requirements
- should then propose a plan for implementation
- Then ask for approval for the requirements and the plan. It shall not proceed without approval!

[...]

But what happens is that it asks for approval but immediately continues (see screenshot).

The reason I am adding this is to save costs to avoid it running in the wrong direction for too long. Is there no way to make my agent stop? Because in the current way it’s just destined to run the wrong way and spend all my limits on misunderstood requirements.

implementation-plan
implementation-plan489×627 58.6 KB

2

Yep this is a known issue with the current version, but should be fixed in 0.46 with the new ‘ask’ mode feature. This will let you properly plan things out and get approval before executing!

v0.46 is currently rolling out to users, so if you don’t have it available already, it should be with you very soon!

3

Oddly I’m having the opposite issue Cursor asking to proceed It rarely (almost never) proceeds to do the work and keeps asking me “would you like me to proceed with the changes?”