Rephrase ASCII requirement in system prompt: Generated content, not file encoding.
Resolved 💬 3 comments Opened Mar 5, 2026 by cmeeren Closed Mar 5, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What variant of Codex are you using?
CLI
What feature would you like to see?
The system prompts in this repo instructs to prefer ASCII. My interpretation is that this is about what the agent produces, not about the file encoding. On Windows (not WSL), it seems to have the effect that files are read/written with ASCII/default encoding, causing garbled text (mojibake) if a file is already UTF-8 and contains non-ASCII characters (such as æøå).
The prompt should be updated to clarify that the ASCII preference is for generated content, and that all files should be read/written as UTF-8 unless otherwise specified, or something. (UTF-8 is, after all, compatible with ASCII for ASCII-only content.)
Additional information
_No response_
3 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
It's similar, but I don't think it's a duplicate.
The system prompt indicates what the model should do, not the encoding on disk. The latter is the concern of the harness and tools that are called.
If you see examples of incorrect encoding, please use the
/feedbackslash command to upload your session details and file a bug report.