Description
When Claude Code streams responses containing Chinese text, occasionally a few characters appear garbled/mojibake (e.g., "看下���果" instead of "看下效果") while all surrounding characters render correctly.
Environment
- Claude Code CLI (macOS)
- Terminal: default zsh
- Model: claude-opus-4-6
Root cause hypothesis
UTF-8 Chinese characters are 3 bytes each. When the streaming response chunks happen to split at a byte boundary within a multi-byte character, the terminal renders the incomplete bytes as replacement characters (�).
Reproduction
- Not reliably reproducible — it's a race condition dependent on where chunk boundaries fall
- Happens occasionally in longer responses mixing Chinese and ASCII text
- Only a few characters are affected per occurrence, not entire lines
Expected behavior
All multi-byte characters should be buffered correctly before rendering, regardless of chunk boundaries.
Description
When Claude Code streams responses containing Chinese text, occasionally a few characters appear garbled/mojibake (e.g., "看下���果" instead of "看下效果") while all surrounding characters render correctly.
Environment
Root cause hypothesis
UTF-8 Chinese characters are 3 bytes each. When the streaming response chunks happen to split at a byte boundary within a multi-byte character, the terminal renders the incomplete bytes as replacement characters (�).
Reproduction
Expected behavior
All multi-byte characters should be buffered correctly before rendering, regardless of chunk boundaries.