Increased frequency of Japanese character corruption (mojibake) on Windows

Description

Japanese characters in Claude Code output are increasingly corrupted (mojibake) on Windows. This appears to have worsened recently.

Environment

• OS: Windows 11 Pro 10.0.26100
• Shell: bash (Git Bash)
• Claude Code: latest
• Model: claude-opus-4-6

Symptoms

• Japanese text in Claude Code's text output occasionally shows corrupted characters
• The frequency of corruption has noticeably increased recently
• Affects inline text output (not just heredoc/bash commands)

Examples

Corrupted characters observed in output text:

• Replacement characters appearing instead of valid Japanese characters
• Long-vowel mark (U+30FC) confused with dash characters

Notes

• The project's encoding rules already document heredoc-related Japanese encoding issues and recommend using temp files with --body-file for gh commands
• This issue appears to be broader than just heredoc - it affects Claude Code's own text rendering