Convert still requires semantic description

The cp-skill-convert skill is mostly structural — add frontmatter with fixed structural values (type: kb/types/note.md, traits: [], tags: []), leave user-verified absent, and align the filename to the title while repairing backlinks.

But description breaks this pattern. Writing a good description requires reading and understanding the content — it is a semantic judgment. We kept it because the note schema requires both description and type; frontmatter without a description is invalid, not merely hard to retrieve. The requirement also serves progressive disclosure: agents decide whether to load a note from its title and description.

Options if we want to make convert fully syntactic: - Leave the artifact as text until a later semantic step can supply a valid description - Auto-generate from the first paragraph (mechanical, but often poor quality) - Accept that description is the one semantic judgment convert must make

Currently we accept the third option. If a script replaces the skill, this is the part that still needs LLM involvement.