Describe the Bug
In a Python project, if I prompt the agent to run a command to run the project’s unit tests and to fix any errors that are observed it seems to have difficulty seeing the output of the command. This seems to occur in many circumstances where a command needs to be run to assess something. I have had some limited success in asking it to redirect output to a file rather than looking at stdout. That trick seemed to work for a while but now it seems intermittent at best if works.
Steps to Reproduce
A command such as this one (which it concocted on its own based on me telling it errors were seen when running “tox -e py3 – ironic.tests.unit.conductor” and telling it to redirect to a file if it can’t see the output (because this is an ongoing problem):
“cd /home/user/src/python/opendev.org/ironic && tox -e py3 – ironic.tests.unit.conductor -v 2>&1 | tee /tmp/conductor_test_errors.txt”
Results in a response like this:
“The command seems to have run but I didn’t see the output. Let me check the file to see what errors occurred.”
…but that was immediately after the command was launched and that particular commands takes a little while to finish so clearly it didn’t wait at all for it to complete.
It then suggests running this command to check the output which I wait for the original command to complete before I acknowledge/approve it.
“cat /tmp/conductor_test_errors.txt”
Which results in a response like this:
“It seems the tee command didn’t capture the output properly. Let me try running the test command again directly to see the output.”
which doesn’t seem valid since that file actually does contain the output of the command.
Then it suggests running it this way:
“cd /home/user/src/python/opendev.org/ironic && tox -e py3 – ironic.tests.unit.conductor-v”
which this time it does seem to wait for it to complete but still can’t see any output at all so suggests yet a different way to run it which goes on, and on, and on.
Expected Behavior
If it runs a command I expect it to see the output and deal with it correctly. If it has to pipe to a file then I expect it to still wait for the command to complete before checking the result.
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Version: 1.2.2 (user setup)
VSCode Version: 1.99.3
Commit: faa03b17cce93e8a80b7d62d57f5eda6bb6ab9f0
Date: 2025-07-07T06:19:13.016Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.26100
Additional Information
Running on Windows but with WSL2 (Fedora 40)
The OSS project is: https://opendev.org/openstack/ironic.git
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor