[Codex App] Unexpected non-Chinese Unicode fragment appended to a Chinese response

What version of the Codex App are you using?

26.820.7780.0 (local package build; bundled app-server 0.150.0-alpha.8)

What subscription do you have?

Not disclosed

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

In a Chinese-language conversation in the Codex desktop app on Windows, an otherwise normal assistant response ended with an unrelated mixed-script Unicode fragment:

വിശദ♀♀

The fragment was not present in the user's prompt and was appended after the intended Chinese answer. It appeared in the visible assistant message itself, not in a terminal command or generated project file.

This was observed once on 2026-08-30 (Asia/Shanghai). The exact trigger is unknown.

I am reporting this conservatively to determine whether the fragment was already present in the raw model response or was introduced by client-side rendering, streaming, encoding, or post-processing.

What steps can reproduce the bug?

1. Open the Codex desktop app on Windows.
2. Continue a long, multi-turn conversation in Chinese.
3. Ask the assistant for a normal Chinese answer.
4. In the observed case, the answer was coherent, but the following unrelated fragment appeared at the very end:

വിശദ♀♀

Reproducibility: observed once; no deterministic minimal reproduction is currently available.

The public report intentionally omits the session/thread identifier. It can be supplied through a private OpenAI feedback channel if that channel becomes reachable.

What is the expected behavior?

The assistant response should contain only the intended content. Unrelated Unicode fragments should not be appended to the end of a response.

If anomalous text is returned, Codex should preserve enough diagnostic information to distinguish raw model output from client-side rendering, streaming, encoding, or post-processing.

Additional information

Related report: #20054 (unexpected anomalous characters appended to Codex CLI output). This occurrence differs in that it was observed in the Codex desktop app on Windows, surrounded by Chinese text, and contains Malayalam-script characters followed by two U+2640 symbols.

Observed code points:

U+0D35 U+0D3F U+0D36 U+0D26 U+2640 U+2640

An in-product feedback submission was attempted without logs, but it timed out twice and produced no success receipt, so this public issue is being filed instead.

No logs, screenshots, local paths, account identifiers, IP addresses, or private thread identifiers are attached.

No source file, test, or entry point is named. Start by reproducing the issue in the Windows desktop app with a long Chinese conversation, then compare the raw response with the rendered message and capture diagnostics for streaming, encoding, and post-processing. Done means identifying where the fragment is introduced or obtaining evidence that it originates in model output.