Powershell optimizations on windows
What feature would you like to see?
Version 0.30 has improved many aspects of the application, and the GPT-5 model on high seems excellent at understanding and creating code.
However, on Windows it seem that GPT-5 is quite limited by its understanding of Powershell. It get the jobs done, eventually, but it sometimes takes 3-4 attempts at powershell. Errors such as missing quotes for the command, assuming tools are present or invalid characters / line breaks often emerge. My impression is that this limits what the model can archive on windows, and it sometimes give up.
WSL is not an option for some commercial settings that requires browser access or use of e.g older .net versions.
It would be great to enhance how codex use powershell on windows, or give it better tools on windows to search files and find information.
Are you interested in implementing this feature?
No sorry
Additional information
_No response_
3 Comments
shell caommand Made a PR to fix Fix shell command win
This modification enhances Codex so that, when running commands on Windows systems, it prioritizes using pwsh statements. It also wraps the commands in script blocks to prevent the outer shell from consuming pwsh statements and causing errors related to symbols and other issues. In practice, there are still occasional cases where one or two commands fail, but this is a significant improvement compared to before, when 5–6 or even more commands would fail consecutively.
Thanks for the bug report. We've shared the feedback with the team members who are responsible for training our codex models. We're tracking this feedback and similar reports so we can make the model more robust in future iterations.
In particular, we know that there is room for improvement on Windows with PowerShell.
If you see other model behavior that you'd like to report, please use the recently-added
/feedbackcommand in the CLI. That will provide us with additional details including logs and conversation text.