Summary

On Windows, a skill can disappear from openclaw skills list if SKILL.md starts with a UTF-8 BOM (EF BB BF).

This happens when the file is edited with PowerShell / .NET WriteAllText using the default UTF-8 encoding behaviour. The parser in extractFrontmatterBlock fails silently because it checks whether the content starts with ---, but the BOM comes first.

Reproduction

1. Install a skill on Windows.
2. Edit SKILL.md using PowerShell WriteAllText without explicitly using new UTF8Encoding($false).
3. This writes the file with a UTF-8 BOM.
4. Run openclaw skills list.

Actual behaviour

The skill disappears from openclaw skills list.

Expected behaviour

The skill should still be discovered even if SKILL.md begins with a UTF-8 BOM.

Root cause

extractFrontmatterBlock in frontmatter-CAnFi2D6.js checks whether the file content starts with ---.

When the file starts with a BOM, the first character is \uFEFF, so the frontmatter check fails.

Proposed fix

Strip a leading BOM before checking for frontmatter:

content = content.replace(/^\uFEFF/, '')

Then proceed with the existing content.startsWith('---') logic.

Why this matters

This is an easy Windows footgun because PowerShell / .NET file-writing paths commonly produce BOM-prefixed UTF-8 unless explicitly configured otherwise. The current failure mode is silent, which makes the bug hard to diagnose.