Context"@" not working when re-prompt previous step in one session

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

When reverting code changes to a specific step, if re-prompt in that step and context is appended using ‘@’, the appending action cannot take effect

Steps to Reproduce

  1. Coding using agent mode with many steps
  2. Reverting to a previous step and re-prompt in that step
  3. using ‘@’ and choose one file/folder, click or enter

Expected Behavior

Context added correctly

Operating System

Windows 10/11

Version Information

Version: 3.0.6 (system setup)
VSCode Version: 1.105.1
Commit: 6e696fa8ae574d6a40e0f1dbf74bd7d823f0b0d0
Date: 2026-04-02T21:42:37.269Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26100

For AI issues: which model did you use?

Auto

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

5

Hey, thanks for the report. This is a known issue. After reverting to a previous checkpoint, the editor state can get out of sync, so @ mentions stop working.

Related thread: Reverting changes does not reset the context

The team is aware. For now, here are a couple workarounds:

  • Instead of revert + re-prompt in the same chat, start a new chat with Ctrl+N and send your request again with the right context.
  • Commit with Git before experimenting, so you can easily get back to the state you want.

I’ve shared this with the team. Let me know if you have any questions.

7

Thanks, but neither is a good workaround.

  • Starting a new chat gets rid of the context. Summarising the context to give it to the new chat is quite sub-optimal.
  • I usually go back to a previous checkpoint because the model did not understand the request and made ineffective or wrong changes. Going back and tweaking the instructions is the right way to do that, not arguing with the model and polluting the context with deviations and arguments. Even if I committed each and every point in the conversation (which is absurd), the proposed workaround doesn’t get rid of the polluted context, which is the main reason to revert to a previous checkpoint, so it’s not a solution.

Please consider elevating the priority of this issue.

Thanks!

8

Hey @jungle, I totally agree, your arguments are valid. A new chat loses context, and revert is meant to roll back context, so git doesn’t help here. The workarounds don’t really solve the problem.

This really gets in the way of a normal workflow, and repeat reports help push it higher in priority.