GFM subset, not full CommonMark. The parser implements the syntax most used in practice: heading 1-6, bold, italic, strikethrough, inline and fenced code with language hint for external syntax highlighting (Prism, highlight.js), ordered and unordered lists, blockquote, links, images, pipe-syntax tables with alignment, HR. It does not support footnotes, task lists, definition lists, math LaTeX, iframe embeds, GitHub-style autolinks (#123, @user). For strict CommonMark spec compliance you need a dedicated server-side parser.
HTML -> Markdown reverse. The reverse conversion walks the DOM tree via the browser's native DOMParser and produces equivalent Markdown for headings, paragraphs, bold, italic, code, links, images, blockquote, lists and tables. All inline styles, CSS classes, container divs and semantically empty spans are ignored: the semantic content is extracted and the presentation chrome dropped. The behavior is intentional for content portability. It does not handle complex nested HTML, iframe, script, form.
Operational privacy. Parsing is entirely local to the browser: the content of a draft post, internal documentation or personal notes does not transit through third-party backends. Useful in contexts where the content itself has value (NDA material, enterprise documentation, pre-publication assets).