Cursor Pro Agent Mode falsely says code was modified, but no file changes are applied

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I am using Cursor Pro with Agent Mode and the Auto model.

When I ask the agent to modify existing code, it often behaves like Ask mode and only returns an explanation or suggested code instead of actually editing the file.

In some cases, it explicitly says that the code has been updated or fixed, but when I check the target file, no actual changes have been applied.

So the issue is not just that editing fails, but that the agent sometimes falsely reports that the edit was completed even though the file remains unchanged.

Steps to Reproduce

  1. Open a project in Cursor Pro.
  2. Enable Agent Mode.
  3. Use the Auto model.
  4. Ask the agent to modify an existing source file.
  5. The agent responds as if it completed the requested code change.
  6. Check the actual file contents or diff.
  7. The file is sometimes unchanged, or the response is only descriptive like Ask mode rather than a real edit.

Expected Behavior

When using Agent Mode, if the agent says it modified or fixed the code, the actual file should be changed accordingly.

If the agent cannot edit the file, it should clearly say that it could not apply the change instead of responding as though the edit was completed.

Screenshots / Screen Recordings

스크린샷 2026-03-18 133049.png
스크린샷 2026-03-18 133049.png785×780 27.6 KB

Operating System

Windows 10/11

Version Information

Version: 2.6.19 (system setup)
VSCode Version: 1.105.1
Commit: 224838f96445be37e3db643a163a817c15b36060
Date: 2026-03-12T04:07:27.435Z
Build Type: Stable
Release Track: Default
Electron: 39.4.0
Chromium: 142.0.7444.265
Node.js: 22.22.0
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Auto model selection in Agent Mode

Additional Information

This issue is misleading because the agent response makes it look like the code change was successfully applied, even when the file remains unchanged.

It reduces trust in Agent Mode because I have to manually verify whether the edit was actually made every time.

The problem appears during normal code editing requests, where Agent Mode should directly modify existing files rather than only replying with explanation-like output.

Does this stop you from using Cursor

Yes - Cursor is unusable

4

Hey, this is a known issue. The UI can show Agent mode, but the backend actually sends Ask mode instructions to the model. That’s why the agent either just explains things or hallucinates that it made edits when no real file changes happen.

A few workarounds that have helped other users:

  1. Always start a new chat when you want Agent mode. Avoid switching between modes Ask, Plan, Agent in the same conversation. Switching modes mid-chat is the main trigger.
  2. If you notice Agent acting like Ask, close that chat and open a fresh one with Agent already selected.
  3. Try selecting a specific model instead of Auto to check if the issue is mode-related or model-related.

The team is aware of this issue. Your report helps increase visibility.

Can you share a Request ID from one of the affected chats? Click the three dots in the top-right of the chat, then Copy Request ID. This helps us confirm what’s happening server-side.

6

Thanks for the clarification.

Yes, this matches what I am experiencing.
Agent mode appears in the UI, but it sometimes behaves like Ask mode and either only explains the change or says it edited the code when no actual file changes were made.

I will add the Request ID below as well.
Request ID: cfd37637-901e-459c-b4a0-70b7af558da2, 601deccb-f437-425b-8097-fbcd36ca3056

I also tested this while using Cursor Pro, Agent Mode, and Auto model selection.

8

Hey @ehdgjs6203!

The agent falsely reporting code modifications has been fixed in a recent Cursor update. Updating to the latest version should resolve this. If you’re still experiencing this on the latest version of Cursor, please raise a new thread and we’ll take a look.