Cursor Agent Mode calls agent terminal to execute, but no response is generated. The Agent marks the termial as success and fast forwards without any real terminal feedback. Windows Machine
Steps to Reproduce
Just start a Agent (Opus 4.5) and give him tasks like execute tests with uv
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
I have the same problem. Auto agent running but no output from commands in terminal. Request id: 8395af00-4c17-4e81-bb81-b03e3da1d59d. PowerShell. Local project. I enabled legacy terminal.
I’m a big Cursor fan and I’m always promoting it. I get the “move fast and break things” mindset, but maybe the update process could reflect that work style a bit better. I don’t care about disk space—just keep the latest five Cursor versions installed on my machine. That way, if something breaks, I can roll back to a working version in one to three clicks.
Also, the terminal issues has been the most prevalent problem for me as far as I can remember in my cursor live, so it should get higher priority.
Title: Shell tool returns empty output on Windows PowerShell
Description:
The AI agent’s Shell tool executes commands successfully (exit code 0) but returns completely empty output. This makes it impossible for the agent to verify command results or read any shell output.
Environment:
OS: Windows 10 (10.0.26200)
Shell: PowerShell
Cursor Version: 2.1.49
Steps to reproduce:
Have the agent run any shell command (e.g., git status, gh pr list, etc.)
Command exits with code 0
Output is completely empty: Command output: \n\n``\n\n```\n**Expected:** Shell output should be captured and returned to the agent.**Impact:** Agent cannot verify actions, read command output, or complete workflows that depend on command results. Forces user to manually run commands and paste output back.
Same issue here. It started a few days ago (around Dec 2 or 3). I updated to the latest version today (2.1.49), this caused my Cursor to complete reset its config. After the reset it briefly worked, but then reverted back to the same behavior after about an hour: agent runs commands thinking they are successful but the command didn’t run and/or it can’t see the output.
