Codex Desktop dictation uses ASCII commas instead of Chinese full-width punctuation for Chinese speech

What issue are you seeing?

Codex Desktop voice dictation/transcription for Chinese speech inserts ASCII punctuation, especially ,, between Chinese clauses instead of Chinese full-width punctuation such as ，.

For Chinese users, this makes dictated prompts look unnatural and requires manual cleanup after nearly every dictation.

Environment

• App: Codex Desktop for macOS
• Bundle identifier: com.openai.codex
• Codex app version: 26.506.31421
• macOS app bundle version: 2620
• Platform: macOS / Darwin arm64
• User input language: Simplified Chinese
• macOS selected input source: Simplified Chinese ITABC
• macOS Dictation preferred language includes zh_CN

Steps to reproduce

1. Open Codex Desktop on macOS.
2. Use Codex built-in dictation / global dictation from the composer.
3. Speak a normal Chinese sentence with natural clause pauses, for example:

帮我看看为什么 Codex 的语音输入中文之间的逗号为什么是英文格式的，不是中文格式的逗号。

1. Submit the dictation result into the composer.

Expected behavior

When the detected/transcribed language is Chinese, punctuation inserted between Chinese characters should use Chinese full-width punctuation, for example:

帮我看看为什么 Codex 的语音输入中文之间的逗号为什么是英文格式的，不是中文格式的逗号。

Actual behavior

Codex dictation may produce ASCII punctuation between Chinese clauses, for example:

帮我看看为什么 Codex 的语音输入中文之间的逗号为什么是英文格式的,不是中文格式的逗号。

Local investigation

This appears to be in the Codex dictation pipeline rather than the user's macOS Chinese input method.

Local inspection of the app bundle shows a Codex-specific global dictation route:

• webview/assets/global-dictation-page-*.js records audio with MediaRecorder and posts the result through the Codex transcription flow.
• webview/assets/use-recording-waveform-*.js posts audio to /transcribe.
• The multipart builder supports an optional language field, but the global dictation call path I inspected did not appear to pass zh-CN / zh_CN into the transcription request.
• The cleanup prompt says: Clean up dictation transcripts. Fix likely speech recognition mistakes, punctuation, capitalization, and formatting..., but does not appear to instruct the cleanup step to use locale-appropriate full-width Chinese punctuation when the transcript is Chinese.

So a likely fix would be one or both of:

• Pass the detected/user language such as zh-CN to /transcribe for Chinese dictation.
• Add locale-aware punctuation guidance in the post-transcription cleanup step, e.g. Chinese transcripts should prefer ，。？！；： between Chinese text rather than ASCII ,.?!;:.

Additional context

This issue is distinct from transcription failure. The transcription succeeds, but punctuation style is wrong for Chinese text.