UTF-8 surrogate encoding error on Windows with non-ASCII (Korean) directory paths

Bug Description

When running Claude Code in a directory containing Korean (non-ASCII) characters in the path, a persistent error message appears:

Claude | Error: 'utf-8' codec can't encode character '\udcec' in position 31: surrogates not allowed

This error message overlays the status bar, making it unreadable.

Environment

• OS: Windows 11 Pro
• Platform: win32
• Shell: bash (Git Bash)
• Claude Code model: claude-opus-4-6

Steps to Reproduce

1. Navigate to a directory with Korean characters in the path (e.g., D:\01_총무,재무)
2. Launch Claude Code (claude)
3. The error message appears and covers the status line

Expected Behavior

Claude Code should handle non-ASCII (Korean, Japanese, Chinese, etc.) directory paths without encoding errors on Windows.

Analysis

The surrogate character \udcec (0xec is a leading byte for Korean UTF-8 sequences) suggests that somewhere in the codebase, a file path is being decoded with the surrogateescape error handler (common on Windows/Python) and then re-encoded with strict UTF-8, which rejects surrogate characters.

The fix would likely involve using surrogatepass or properly handling the encoding roundtrip for Windows paths containing non-ASCII characters.