Agent shell commands never execute on Windows 11 (no exit status)

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

The AI agent’s terminal/shell tool runs commands but always returns “no exit status.” The commands never actually execute — files they are told to create are never created, and no output is captured. This makes the agent unable to run any shell commands (git, python, file listings, etc.). A normal manual terminal works fine in the same window.

Steps to Reproduce

  1. Open Cursor on Windows 11 with a workspace folder.
  2. Ask the agent to run any shell command (e.g. count files, run a python script).
  3. Agent reports “the shell command returned no exit status.”
  4. Verify the command did nothing (e.g. an output file it should have written does not exist).
  5. Open a manual terminal (Ctrl+) and run the same command (e.g. echo hi`) — it works and prints output, confirming the shell itself is fine.

Expected Behavior

The agent’s terminal tool should execute the command, capture its stdout/stderr, and return an exit code — the same way the manual integrated terminal does.

Operating System

Windows 10/11

Version Information

Windows 11 24H2 (build 26100)

For AI issues: which model did you use?

Not model-specific — this is an agent terminal/shell-integration issue, reproduces regardless of model.

Additional Information

Already ruled out on my side:

  • Reloaded the window multiple times.
  • Fully restarted Cursor multiple times.
  • Execution policy is not the cause (LocalMachine = Unrestricted; also set CurrentUser = RemoteSigned). No change.
  • settings.json is clean (no terminal overrides): only window.autoDetectColorScheme, security.workspace.trust.enabled, cursor.general.disableHttp2.
  • Manual terminal (Windows PowerShell 5.1) works and prints output; only the AGENT shell fails.
  • Agent’s file read/edit tools work — only command execution is broken.

Does this stop you from using Cursor

Yes - Cursor is unusable

6

Hey @Omri_Chen ,
This is a Windows behavior we’ve seen before: the agent’s Shell tool can silently return “no exit status” and run nothing (no output, no files created), while your manual terminal keeps working fine. On Windows the agent’s sandboxed command path isn’t supported, so the command stream can close without ever returning a result. The integrated terminal uses a different path, which is why it’s unaffected. Your “settings.json is clean” note fits too, since the setting behind this isn’t stored there.

Quickest way to unblock yourself:

  1. Turn on the Legacy Terminal Tool. Open Cursor Settings, go to Agents, and toggle Legacy Terminal Tool on (it’s under the Inline Editing & Terminal section, or just type “Legacy Terminal” into the settings search). This restores the agent’s ability to spawn a terminal and is the most reliable fix.
  2. Check Auto-Run (it may be labeled Run Mode) under Cursor Settings → Agents. If it’s set to a sandbox mode, switch it to Use Allowlist. On recent versions the sandbox option is hidden, so if you don’t see one, switch the mode to another option (for example Auto-review) and back again to reset it, then start a new chat.
  3. Fully quit Cursor (close it completely, not just Reload Window) and reopen.

Could you let me know if turning on the Legacy Terminal Tool gets the agent running commands again? If it does, that confirms it. If it doesn’t fix it even after a full restart, grab a Request ID from one of the failed commands (open the ⋯ menu on that agent message, then Copy Request ID) and drop it here so I can trace your specific case.