Parse more forms of apply_patch
Resolved 💬 3 comments Opened Sep 25, 2025 by iamjackharper Closed Oct 20, 2025
What version of the VS Code extension are you using?
0.4.14
Which IDE are you using?
VS Code
What platform is your computer?
MacOS Sequoia 15.6.1 | Darwin 24.6.0 arm64 arm
What steps can reproduce the bug?
Select gpt-5-codex under model selection and ask for any change in the code
What is the expected behavior?
It should behave like gpt-5 and edit the files in the chat so I can see and undo the changes.
What do you see instead?
It only uses shell commands like
zsh -lc ls to list files
zsh -lc 'cat filename' to read files
zsh -lc 'apply_patch <<'"'"'PATCH'"'"'
* Begin Patch
* Update File: filename
Additional information
I'm still using the extension with gpt-5 model instead of gpt-5-codex because of this behavior.
Is this a bug or it's supposed to work like that?
3 Comments
This is happening to me also VSCode Insiders, WSL2 backend
Happening to me as well on cursor
It happens in Windows, WSL, VS Code with WSL2 backend, and VSCode running in Windows. It slows codex down to a crawl and patches are also super inconsistent.