markdown-plus-author

Provide the source document and ask for a Markdown+ rewrite. The Skill classifies the input, plans a block manifest, assigns types from a controlled vocabulary, and outputs a single CommonMark-valid file with block IDs, metadata, captions, and prose companions.

Markdown+ is plain Markdown plus three habits: substantive sections become bullet-list blocks with `**#id**` headers and inline-code metadata, binary media uses external relative paths, and every figure or table gets a prose companion. It stays valid CommonMark, unlike `:::` directive dialects.

Yes. Markdown+ source is valid CommonMark, so standard viewers render it as nested bullet lists, inline code, italic captions, and standard tables with no unrendered directives. A dedicated viewer can additionally project it into rich HTML with nav, pills, and cards.

Avoid it for slide decks, one-off emails or chat replies, long-form prose essays where block structure harms the writing, and short documents with fewer than two substantive sections. Plain Markdown or a dedicated format fits those cases better.

LLM rewrites commonly flatten directory trees into paragraphs, drop fence markers, or re-align whitespace, permanently destroying structure. The Skill treats fenced code, box-drawing trees, and aligned art as byte-for-byte protected content inside fenced blocks.

All binary media and SVG are referenced by relative path, never base64 or inline. Tables over roughly 30 rows move to an external CSV via `data-source:`, and large chart specs are extracted to external JSON files.