Agents executing a plan say they will do something then don't

[Evooo]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

When I generate plans in plan mode, then hit Build, at some point the agent will just stop doing anything and will just keep saying things like “I’m working on step 5 and will continue through step 7 as instructed. There is nothing more for you to do right now.” and then it literally ends the message there and does nothing.

I can hit build repeatedly and say things like “please do the next step” and it repeatedly just does the same sort of message without actually doing any of the work it says it is going to do. In thinking, I can see that it is reasoning about how it should tell me it is going to do something rather than how to actually execute what to do.

I’ll end up wasting tokens repeatedly going back and forth with the agent just trying to get it to actually take some action on the plan but all I get is something along the lines of “Got it…rest assured, I’m working on this and will report when I am finished.” or simply “Continuing implementation” then ends the session without doing jack.

Eventually it stalls completely and refuses to do more work at all no matter how many times I hit build, justifying itself by saying:
I can’t silently keep executing the entire multi-phase plan to completion within a single chat turn.

That is also not correct, because it stops to tell me it is about to take action dozens of times without proceeding.

Steps to Reproduce

Just start with a plan and hit build. I’ve found it breaks down by about step 5 on a 20 or some step plan.

Expected Behavior

It actually does the work in the plan instead of telling it is doing something without actually doing it.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.3.35
VSCode Version: 1.105.1
Commit: cf8353edc265f5e46b798bfb276861d0bf3bf120
Date: 2026-01-13T07:39:18.564Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin x64 25.1.0

For AI issues: which model did you use?

Auto

Does this stop you from using Cursor

Yes - Cursor is unusable

[deanrie]

Hey, thanks for the detailed report. This looks like a prompting issue with the agent in Plan Mode. It gets stuck in a reasoning loop instead of taking actions.

We need a bit more info to debug:

1. The Request ID from the broken build: in the chat, top right context menu > Copy Request ID
2. If possible, a screenshot of the thinking phase showing the agent reasoning about “how to say it will work” instead of what to do
3. The exact plan that triggers the issue (screenshot or text is fine)

Workarounds to try for now:

• Switch to plain Agent Mode (no Plan). In the mode menu, pick Agent instead of Plan. This should bypass the problematic Build transition
• Try splitting the plan into smaller chunks (5 to 7 steps instead of 20) and run them one at a time
• If Auto is selecting GPT-5.2, try explicitly selecting Claude Opus 4.5 in the model picker

We have a few open tickets about Build transition issues in Plan Mode, but this specific pattern (“says it will work” but doesn’t do anything) hasn’t been tracked yet. With the Request ID and details, we can send it to the team.