More reliable command execution on windows
Resolved 💬 4 comments Opened Aug 21, 2025 by FreddyE1982 Closed Nov 28, 2025
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
If running on windows the execution of commands by the agent is quite chaotic and faulty.
The problems include that the agent:
- repeatedly tries executing linux commands and only stops if specifically told that we are running on windows
- repeatedly tries running incomplete and faulty commands, even if
- it had run that same or similar command correctly before. this especially applies to working with / in files, finding strings in files etc.
- constantly runs into (powershell) string quoting issues
This leads to tokens beeing used unessecarily because the agent has to keep trying to fix its own blunders.
One way to solve this may be automatically having the agent execute any commands via cygwin or install a small linux distro via WSL and use that....so we would avoid the agent having to execute any windows commands at all.
4 Comments
+1
It's quite unusable, much faster on Windows to use Codex CLI via WSL
Aside from the usability downsides of CLI vs. IDE extension the terminal doesn't auto-scroll (at least for me in Cursor) which is really annoying.
I'm obviously biased, but I'd think Codex being essentially unusable on Windows would be the highest priority issue (especially prior to "DevDay")
We've made many improvements to codex on Windows over the past month. I'm going to close this feature request. If you're seeing specific behaviors that you consider to be bugs that are not already covered by active bug reports, please open a new issue and provide repro steps.