Editing agent terminal commands still runs the original command

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

When running an agent and the agent proposes to run a terminal command (I have ‘Ask Every Time’ enabled so I review each command before it runs), if I edit the command before hitting run, the command that is executed is the original proposed command, not incorporating my edits.

Previous behavior was such that I could make inline edits to a command before running it. Now I have to actually send a message to the agent telling it to generate a new command. It is also just misleading UX to have the command be user editable but then not run the edits.

Would be a big improvement to resolve this bug for those of us who don’t let the agent just run wild with terminal commands.

Steps to Reproduce

Start an agent session and ask it to run some command. Then, when it prompts you to run the command it generates, edit the command before accepting. You’ll see it run the original command.

Expected Behavior

When editing a command, the edited version should be the one run in the terminal.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.1.50
VSCode Version: 1.105.1
Commit: 56f0a83df8e9eb48585fcc4858a9440db4cc7770
Date: 2025-12-06T23:39:52.834Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 24.6.0

For AI issues: which model did you use?

This happens with any AI model; it is an issue with the cursor agent framework and the terminal command tool that the models use.

For AI issues: add Request ID with privacy disabled

n/a

Does this stop you from using Cursor

No - Cursor works, but with this issue

2

Hey, thanks for the report. It looks like a regression in confirming terminal commands.

We need a bit of data:

  • Short video: show editing the command, pressing Run, and what exactly gets executed
  • Confirm settings: Settings > Cursor Settings > Agents > Auto-Run is currently “Ask Every Time”. Check if behavior changes when switching to “Run in Sandbox” and back, then restart
  • Diagnostics: try in an empty project folder and with launching cursor --disable-extensions

Send the data - I’ll pass it to the engineers and update the thread.