Windows: valid Git worktree scans fail for non-ASCII target paths

[pradigmaz]

Environment

• Windows
• Codex Security plugin: 0.1.11
• Security workspace UI: 0.1.63
• Python default locale encoding: cp1251
• Target: <Windows path containing Cyrillic characters>

Reproduction

1. Use a valid Git worktree whose path contains non-ASCII characters.
2. Start a standard scan for the worktree.

Workspace setup reports:

{
  "isGit": true,
  "isWorktree": true,
  "hasHead": true
}

Observed

The scan fails with:

Scan target must stay inside its Git working tree.

git rev-parse --show-toplevel returns the correct UTF-8 path, but Python decodes the output using the Windows locale. The resulting path is corrupted, so Path.relative_to() raises ValueError.

Setting PYTHONUTF8=1 makes the helper pass, confirming the decoding issue.

Validation

The issue was reproduced and verified with temporary Git repositories using root and nested targets. Tested path components included:

• Cyrillic
• Simplified and Traditional Chinese
• Japanese
• Arabic
• Korean
• Thai
• Lao
• Khmer
• Burmese
• Georgian
• Armenian
• Indic scripts
• Persian and Urdu
• Amharic
• Mongolian
• Tibetan
• Ukrainian and Serbian
• Latin diacritics
• Combining characters
• Emoji

After decoding text-mode Git output as UTF-8, all cases passed and returned the correct pathspec. Binary Git output remained unchanged.

Suggested fix

Set an explicit encoding for text-mode Git subprocess calls:

encoding="utf-8" if text else None,

The scan-start waiter also remains pending after this startup error. A retry returns:

A scan-start waiter already exists