Japanese IME: digits are dropped during composition in Codex CLI (e.g., “2つ” → “つ”)

[umekun123]

What version of Codex is running?

0.30.0

Which model were you using?

GPT-5 high

What platform is your computer?

Mac 15.6

What steps can reproduce the bug?

Steps to reproduce the bug

Open Terminal and run a Codex command, e.g. codex ask "test" so the input prompt is focused.

Switch input source to Japanese (Hiragana).

Type 2つ (i.e., press the digit key 2, then type tu for “つ”).

Observe the input field.

Expected behavior The typed digits should be preserved. The string should appear as 2つ.

Actual behavior The digit 2 disappears and only つ remains. This reproduces 100% of the time in my environment. Typing with the English layout doesn’t show the issue; only Japanese IME is affected.

What is the expected behavior?

2つ

What do you see instead?

つ

Additional information

macOS “Japanese – Romaji” (also tested with Live Conversion on).