Agent response included one Russian word rather than the English equivalent

What version of Codex is running?

codex-cli 0.94.0

What subscription do you have?

pro, I think

Which model were you using?

gpt-5.2-codex

What platform is your computer?

kubuntu

What terminal emulator and version are you using (if applicable)?

_No response_

What issue are you seeing?

The agent response used the Russian word разрешить rather than the English equivalent "allow". The Russian word is actually correct (semantically), but for some reason (classic developer expression for confusion) the model used Russian for that one word only and English for the rest.

Here's the full response:

On your question: there’s nothing in OpenAI’s public Codex docs that says a repo file like prompt.md is automatically loaded at the start of every
Codex conversation. The docs describe Codex as operating on the prompts you send and then executing actions, but they don’t mention a global,
always‑loaded prompt.md from the open‑source repo. (developers.openai.com (https://developers.openai.com/codex/prompting?utm_source=openai)) If you
share the file or let me browse GitHub, I can check how it’s referenced in code and answer definitively.

If you want me to proceed, pick one:

1. Paste prompt.md here.
2. Explicitly разрешить GitHub browsing for this question (I’ll open it and trace its usage).

What steps can reproduce the bug?

Uploaded thread: 019c20b7-c7ac-7be2-888d-8e374c4b1788

What is the expected behavior?

_No response_

Additional information

_No response_