Codex Desktop on Linux defaults spellcheck to en-US and flags Polish text as misspelled

Summary

On Codex Desktop for Linux, the composer spellchecker defaults to English (en-US) even when the user writes prompts in Polish. As a result, almost every Polish word is underlined as misspelled, despite the model itself understanding Polish correctly.

Environment

• OS: Ubuntu Linux
• App: Codex Desktop
• Language used in prompts: Polish

Reproduction

1. Launch Codex Desktop on Linux.
2. Type a normal Polish sentence in the prompt composer, for example:

To jest próbne zdanie napisane po polsku, żeby sprawdzić poprawność pisowni.

1. Observe that Polish words are marked with red spellcheck underlines.

Relevant preferences

The active Chromium-style profile preferences were found in:

~/.config/Codex/Default/Preferences
~/.config/Codex/Default/Partitions/codex-browser-app/Preferences

Both contained:

"intl": {
  "selected_languages": "en-US,en"
},
"spellcheck": {
  "dictionaries": ["en-US"],
  "dictionary": ""
}

The partition profile also contained:

"intl": {
  "accept_languages": "en-US,en",
  "selected_languages": "en-US,en"
}

Workaround tested successfully

After fully closing the app, changing the relevant preferences in both files to include Polish and English fixed the issue:

"intl": {
  "selected_languages": "pl-PL,pl,en-US,en",
  "accept_languages": "pl-PL,pl,en-US,en"
},
"spellcheck": {
  "dictionaries": ["pl-PL", "en-US"],
  "dictionary": ""
}

After restarting Codex Desktop, Polish spellchecking worked correctly and Polish text was no longer incorrectly underlined.

Expected behavior

Codex Desktop should either:

• respect the system/application language when configuring spellcheck dictionaries, or
• provide a visible setting for spellcheck languages, ideally supporting multiple active dictionaries such as pl-PL and en-US.

Actual behavior

The application profile defaults to en-US only, making the composer spellchecker effectively unusable for Polish prompts until the preferences files are edited manually.

This appears to be a UI/editor spellcheck configuration issue rather than a model-language issue.