Japanese characters occasionally corrupted in output (file writes and terminal)

Bug

Claude Code occasionally outputs corrupted Japanese (CJK) characters in both file writes and terminal output. The corruption appears as replacement characters (U+FFFD / �), typically rendering as ��� in the terminal.

Examples found in a single session

| Original intent | Corrupted output |
|-----------------|-----------------|
| 定番問題セット | 定番問題セッ��� |
| キャンセル | キャ���セル |
| インポート | インポ���ト |
| テンプレート名 | テ���プレート名 |
| 既にありません | 既にあり��せん |

Pattern

• Seems to affect multi-byte characters (Japanese hiragana/katakana/kanji) randomly
• Not every string is affected — most output is fine
• Corruption typically drops 1-3 bytes from a character, resulting in U+FFFD replacement
• Occurs in both Write tool output and Edit tool output
• Observed across multiple files written in a single session
• No obvious pattern in which characters are affected (hiragana, katakana, and kanji all seen)

Environment

• macOS (Darwin 25.3.0)
• Claude Code CLI
• Model: claude-opus-4-6 (1M context)
• Shell: zsh
• Terminal: likely UTF-8 locale

Impact

Silently introduces broken strings into source code. Requires manual review and correction after each session.