As said in the title, this is happening far too often. I’ve created a GitHub issue a few days ago, no replies. I am disappointed that I have to hold backups of files so I can paste the strings instead of the gibberish every time (all the time).
Can somebody on the team fix this? It’s pretty important for non-English systems.
I am using the VSCode extension with the gpt-5 high model.
Same, still always have problem with non-English strings. I use VSCode + Unity + Codex, GPT-5.2-Codex model. Codex said it can not read my document.
Same thing here, I’m constantly seeing issues related to this. This is how GPT itself explains:”apply_patch is my default, but when the Kotlin file got corrupted earlier (all the {…} strings turned into gibberish) every attempt to use apply_patch produced more malformed quoting—the heredoc kept escaping /" incorrectly and I couldn’t trust the output. Dropping to a tiny Python script let me do the
same textual replacement without fighting the quoting bug. As soon as the edits were in, I deleted the helper script. If the quoting issues settle down I’ll go back to straight apply_patch; this was just the least-risky way to repair the file after the corruption.”
Hey @lubacareer, appreciate you for bringing this to our attention. We’re going to dig into this and see what’s driving it. If you’ve noticed any patterns or you’re seeing specific error messages, feel free to drop them here since that can help us narrow it down faster. We’ll share an update once we learn more.