Cursor version 2.1.39 has terminal bug

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

commands are executed inside terminal but actually not executed
npm run builld → no response from terminal at all, so Claude 4.5 Sonnet confusing all time, like “why??”
git status → no response
git diff vite.config.ts → no response
etc, every command

Steps to Reproduce

Just ask your agent to do some terminal stuff

Expected Behavior

responds command execution as same as previous Cursor version

Screenshots / Screen Recordings

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.1.39 (user setup)
VSCode Version: 1.105.1
Commit: 60d42bed27e5775c43ec0428d8c653c49e58e260
Date: 2025-11-27T02:30:49.286Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Claude 4.5 Sonnet

Does this stop you from using Cursor

No - Cursor works, but with this issue

3 Likes

same

Hey, thanks for the report. This is a known issue with the new terminal implementation on Windows affecting many users in recent versions.

Please try the following solution:

  1. Open Cursor Settings Ctrl+Shift+J
  2. Go to the Agents tab → Inline Editing & Terminal
  3. Enable “Legacy Terminal Tool”
  4. Press Ctrl+Shift+P → “Terminal: Kill All Terminals”
  5. Fully restart Cursor

This should restore terminal command execution. The engineering team is aware and working on fixing the new terminal implementation for Windows.

Let me know if this resolves the issue!

1 Like

Thank you for sharing the work-around. Editor mode seems to work fine. Agents mode isn’t. After turning on legacy terminal tool in settings, it seems to work up to 2 terminals in the editor area. Then the bug is back on. Can’t have a 3rd terminal in the editor area no matter what. I’ve tried the following 3 methods, and none of them worked:

  1. drag-n-drop terminal from bottom pane into editor area.

  2. “Ctrl+Shift+5” in the editor area to create a new terminal tab.

  3. “Ctrl+Shift+5” in the bottom pane and then right click on the terminal and select “Move Terminal Into Editor Area”.

Additionally, in Agents mode, once you have 2 terminals open you can’t open any other files in an editor tab. It’ll just trigger an “InstantiationService” error. This bug poses a significant usability issue would appreciate quick fix. Cursor IDE: 2.1.39 on Windows 11 Pro.

1 Like

same , already rollback to 2.1.36

I had the same problem. I asked many models to debug and finally one of them (Opus 4.5) suggested about working_directory params

<parameter name="command">uname -a</parameter>
<parameter name="working_directory">...

So the solution is to ask the agent to include that parameter in all terminal calls. It seems to work now after I put this as a User Rule.

1 Like

I had the same problem. Enabling “Legacy Terminal Tool” didn’t solve the problem. I’ve already rolled back to version 2.0.77.

1 Like

I understand that the Legacy Terminal Tool didn’t fix the issue. This is a critical bug with the new terminal implementation.

Temporary solutions:

  • Roll back to version 2.0.77 (as you already did) - it’s a stable version
  • Or try a workaround: add the working_directory parameter in User Rules for all terminal calls

The team is actively working on a fix. When the new version is released, try updating and let us know if it helps.

If version 2.0.77 works fine, you can stay on it until the fix is out.

I honestly can’t understand how the Cursor development team could ship a release while leaving such a critical bug unresolved. This issue is essentially a repeat of a previous bug that affected many users — as seen here: Agents terminals are read only
— which suggests the team hasn’t learned properly from past failures. Making a mistake once is carelessness, but making the same mistake again and again borders on incompetence.

I get that Cursor wants to sandbox the Terminal for security reasons, and that policy itself makes sense. But it’s hard to believe their processes and quality checks are anywhere near adequate if this is the result. As a user, is it really too much to expect a team to learn from a mistake the first time?

Rolling Back.

1 Like

They don’t do E2E test before shipping Cursor? I think if the test exists you can find the bug easily.

1 Like

I just updated to 2.1.41 hoping this was the release that fixed this issue. Unfortunately, it did not.

For those who haven’t found a workaround, here’s what I’ve been doing to get by:

  1. When a command freezes, click the “Open in external terminal” button.

  2. If the problem is git or something trying to “scroll” then help it along by hitting “spacebar” until you see “END”, then press “q” to quit. And/or if it looks like the command completed, but Cursor isn’t proceeding, then press ENTER at a blank terminal 1-3 times and that helps a lot, too.

  1. Ta da! (Usually.) Rinse and repeat as needed.

Hope this helps others! And hoping dev fixes this one soon. It’s slowed me down a bit this weekend.

Cheers! –Matt

Update to 2.1.46. Issue resolved.

2.1.46 still had the issue for me. Cursor could not read any terminal output, like from commands like ‘dotnet test’. It seemed like the command never even ran at all. Legacy terminal mode had no effect. I uninstalled and rolled back to 2.1.36 and it works.

1 Like


In SSH linux, it works

Same issue here. Support has not responded to multiple reports. The terminal is broken in the latest 2 updates. Best to request a credit refund, until the cursor resolves the issue.

1 Like

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Describe the Bug *

The Cursor Shell tool completely fails to execute commands. While the tool reports that commands have been executed (exit code 0), commands are actually not executed at all - no processes are spawned, no files are created, and no system changes occur. The Shell tool appears to silently fail without any indication that execution did not happen.

Key Issues:

  • :cross_mark: Commands do NOT execute - No processes are actually spawned or run
  • :cross_mark: No command output - Commands never run, so there is no output to capture
  • :cross_mark: File operations fail - Files are not created because commands never execute
  • :cross_mark: No error indication - Tool incorrectly reports success (exit code 0) when nothing happens
  • :cross_mark: Silent failure - No indication that commands failed to execute

This is a critical bug that completely breaks the Shell tool functionality, making it impossible for AI agents to execute any commands or scripts through Cursor’s Shell tool.

Steps to Reproduce

Test Case 1: Basic Echo Command

  1. Open Cursor IDE
  2. Use the Shell tool (or AI agent attempts to execute a command)
  3. Execute: echo "test"
  4. Expected: Output should contain “test”
  5. Actual: Output is completely empty (exit code: 0)

Test Case 2: PowerShell Output

  1. Execute: powershell -Command "Write-Output 'hello'"
  2. Expected: Output should contain “hello”
  3. Actual: Output is completely empty (exit code: 0)

Test Case 3: PowerShell Get-Date

  1. Execute: powershell -Command "Get-Date; $PSVersionTable.PSVersion"
  2. Expected: Should display current date/time and PowerShell version
  3. Actual: Output is completely empty (exit code: 0)

Test Case 4: File Operations

  1. Execute: powershell -Command "Set-Content -Path 'test-output.txt' -Value 'Test content'"
  2. Expected: File should be created with content
  3. Actual: File is not created, no output returned (exit code: 0)

Test Case 5: Directory Listing

  1. Execute: dir /b or pwd
  2. Expected: Should display directory listing or current path
  3. Actual: Output is completely empty (exit code: 0)

Detailed Test Results:

Command: echo "test"
Exit Code: 0 (reported, but command did not execute)
Output: (empty)
Evidence: Command did not run - no process spawned

Command: powershell -Command "Write-Output 'hello'"
Exit Code: 0 (reported, but command did not execute)
Output: (empty)
Evidence: PowerShell process was never started

Command: powershell -Command "Get-Date; $PSVersionTable.PSVersion"
Exit Code: 0 (reported, but command did not execute)
Output: (empty)
Evidence: No system interaction occurred

Command: powershell -Command "Set-Content -Path 'test-output.txt' -Value 'Test content'"
Exit Code: 0 (reported, but command did not execute)
Output: (empty)
File Created: No
Evidence: File does not exist - command never executed

Critical Finding: Commands are NOT being executed at all. The exit code 0 is misleading - it appears the Shell tool reports success without actually executing anything.

Expected Behavior

The Shell tool should:

  1. :white_check_mark: Execute commands and capture their standard output
  2. :white_check_mark: Return command results as strings
  3. :white_check_mark: Persist file operations (create, modify, delete files)
  4. :white_check_mark: Display error messages if commands fail
  5. :white_check_mark: Support both cmd.exe and PowerShell commands

Example of Expected Output:

Command: echo "test"
Exit Code: 0
Output: "test"

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.1.46 (system setup)
VSCode Version: 1.105.1
Commit: ab326d0767c02fb9847b342c43ea58275c4b1680
Date: 2025-12-02T03:59:29.283Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

Describe the Bug *

The Cursor Shell tool completely fails to execute commands. While the tool reports that commands have been executed (exit code 0), commands are actually not executed at all - no processes are spawned, no files are created, and no system changes occur. The Shell tool appears to silently fail without any indication that execution did not happen.

Key Issues:

  • :cross_mark: Commands do NOT execute - No processes are actually spawned or run
  • :cross_mark: No command output - Commands never run, so there is no output to capture
  • :cross_mark: File operations fail - Files are not created because commands never execute
  • :cross_mark: No error indication - Tool incorrectly reports success (exit code 0) when nothing happens
  • :cross_mark: Silent failure - No indication that commands failed to execute

This is a critical bug that completely breaks the Shell tool functionality, making it impossible for AI agents to execute any commands or scripts through Cursor’s Shell tool.

For AI issues: which model did you use?

Model name: Sonnet 4, Tab… (if applicable)

Note: This issue affects the Shell tool directly, which is used by AI agents when they need to execute commands.

For AI issues: add Request ID with privacy disabled

Request ID: f9a7046a-279b-47e5-ab48-6e8dc12daba1 (if applicable)

Note: This is a system-level issue with the Shell tool itself, not specific to AI requests.

Additional Information

Environment Details:

  • Workspace Path: E:\MadPoster
  • Shell Type: PowerShell (Windows)
  • Terminal Type: Windows PowerShell / cmd.exe

Workarounds Currently Used:

  1. :white_check_mark: Using Cursor integrated terminal (Ctrl + `) - This works correctly
  2. :white_check_mark: Creating .bat or .ps1 scripts and executing them directly - This works correctly
  3. :white_check_mark: Using external PowerShell terminal - This works correctly

Important Notes:

  • :white_check_mark: System PowerShell works perfectly - The issue is specifically with Cursor’s Shell tool integration
  • :white_check_mark: Cursor integrated terminal works correctly - Commands executed there run normally and produce output
  • :cross_mark: Only the Shell tool (used by AI agents) is affected
  • :warning: Commands do NOT execute - Despite reporting exit code 0, commands never actually run
  • :warning: No process creation - No child processes are spawned when using Shell tool
  • :warning: Silent failure - Tool reports success but nothing actually happens

Impact Assessment:

  • :red_circle: Severity: High
  • :red_circle: Affects: All automation tasks that require command execution
  • :red_circle: Workaround Available: Yes (use integrated terminal or external scripts)
  • :red_circle: Blocks Core Functionality: Yes - Makes Shell tool unusable for automation

Related Documentation:

  • Test report: e:\MadPoster\shell-test-report.md
  • Fix guide: e:\MadPoster\FIX_CURSOR_SHELL.md
  • Diagnostic script: e:\MadPoster\diagnose-cursor-shell.bat (if exists)

Attempted Solutions:

  1. :white_check_mark: Restarted Cursor (as administrator)
  2. :white_check_mark: Checked terminal settings
  3. :white_check_mark: Verified PowerShell configuration
  4. :white_check_mark: Tested with multiple command types
  5. :white_check_mark: Confirmed system PowerShell works correctly

Does this stop you from using Cursor

Yes - Cursor is unusable

1 Like

the integrated tools terminal only work once, when fresh install

1 Like

Same issue, no output for the agent. exitcode 0. The terminal is broken in the latest 2 updates. Best to request a credit refund until the cursor resolves the issue.
tried every fix online. deleted workspace history, edited settings.json… None effective

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

[1 tool called]

Report it here: forum.cursor.com


Copy-paste this bug report:


Title:

Environment:

  • OS: Windows 10 (10.0.26100)
  • Shell: PowerShell
  • Cursor Version: [check Help → About]

Bug:
When using AI Agent mode, the Shell tool executes commands (exit code 0) but captures zero stdout/stderr output. The terminal files created in .cursor/projects/*/terminals/ show only headers and footers with empty content between them.

Steps to reproduce:

  1. Open a workspace on Windows
  2. Start an Agent chat
  3. Ask the agent to run any shell command (e.g., echo hello, dir, git status)
  4. Observe: exit code 0, but output section is completely empty

Expected: Command output should be captured and returned

Actual: Commands execute but all output is lost. Terminal log files show:

=== Background Shell XXXXX ===
Command: echo "test"
Working Directory: C:\Users\...
Started: ...
============================================================


============================================================
Exited with code: 0
Ended: ...

Impact: Agent mode is completely unusable for any task requiring command output (builds, tests, logs, debugging).

Notes:

  • Other tools (LS, Read, Write, Grep) work fine
  • Only Shell tool output capture is broken
  • Tried CMD, PowerShell, background mode - all same result

Steps to Reproduce

  1. Open a workspace on Windows
  2. Start an Agent chat
  3. Ask the agent to run any shell command (e.g., echo hello, dir, git status)
  4. Observe: exit code 0, but output section is completely empty

Expected: Command output should be captured and returned

Actual: Commands execute but all output is lost. Terminal log files show:

=== Background Shell XXXXX ===
Command: echo "test"
Working Directory: C:\Users\...
Started: ...
============================================================


============================================================
Exited with code: 0
Ended: ...

Impact: Agent mode is completely unusable for any task requiring command output (builds, tests, logs, debugging).

Notes:

  • Other tools (LS, Read, Write, Grep) work fine
  • Only Shell tool output capture is broken
  • Tried CMD, PowerShell, background mode - all same result

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

  1. Open a workspace on Windows
  2. Start an Agent chat
  3. Ask the agent to run any shell command (e.g., echo hello, dir, git status)
  4. Observe: exit code 0, but output section is completely empty

Expected: Command output should be captured and returned

Actual: Commands execute but all output is lost. Terminal log files show:

=== Background Shell XXXXX ===
Command: echo "test"
Working Directory: C:\Users\...
Started: ...
============================================================


============================================================
Exited with code: 0
Ended: ...

Impact: Agent mode is completely unusable for any task requiring command output (builds, tests, logs, debugging).

Notes:

  • Other tools (LS, Read, Write, Grep) work fine
  • Only Shell tool output capture is broken
  • Tried CMD, PowerShell, background mode - all same result

For AI issues: which model did you use?

  1. Open a workspace on Windows
  2. Start an Agent chat
  3. Ask the agent to run any shell command (e.g., echo hello, dir, git status)
  4. Observe: exit code 0, but output section is completely empty

Expected: Command output should be captured and returned

Actual: Commands execute but all output is lost. Terminal log files show:

=== Background Shell XXXXX ===
Command: echo "test"
Working Directory: C:\Users\...
Started: ...
============================================================


============================================================
Exited with code: 0
Ended: ...

Impact: Agent mode is completely unusable for any task requiring command output (builds, tests, logs, debugging).

Notes:

  • Other tools (LS, Read, Write, Grep) work fine
  • Only Shell tool output capture is broken
  • Tried CMD, PowerShell, background mode - all same result

For AI issues: add Request ID with privacy disabled

  1. Open a workspace on Windows
  2. Start an Agent chat
  3. Ask the agent to run any shell command (e.g., echo hello, dir, git status)
  4. Observe: exit code 0, but output section is completely empty

Expected: Command output should be captured and returned

Actual: Commands execute but all output is lost. Terminal log files show:

=== Background Shell XXXXX ===
Command: echo "test"
Working Directory: C:\Users\...
Started: ...
============================================================


============================================================
Exited with code: 0
Ended: ...

Impact: Agent mode is completely unusable for any task requiring command output (builds, tests, logs, debugging).

Notes:

  • Other tools (LS, Read, Write, Grep) work fine
  • Only Shell tool output capture is broken
  • Tried CMD, PowerShell, background mode - all same result

Does this stop you from using Cursor

Yes - Cursor is unusable

hi @Someguy333 could please clean up this thread? it has some information multiple times posted which is confusing. Please avoid using AI to write the content. Its ok to use AI to translate it.

Also following path is not recognized by Cursor and you should not use it .cursor/projects