Parser rejects SKILL.md with leading HTML comments

skills add fails when SKILL.md has anything before the opening ---, even an HTML comment.

Repro:
npx skills add Shopify/agent-skills --skill shopify-hydrogen # Repository cloned
# No valid skills found.

Each Shopify SKILL.md starts with a <!-- AUTO-GENERATED ... --> banner above valid frontmatter. gray-matter, Jekyll, Hugo all skip leading whitespace and HTML comments before locating the block — could this parser do the same? Would unblock any source that prefixes generated banners, not just Shopify's.