markz

**markz — small, opinionated Markdown.**

A standalone Markdown package for TypeScript/JavaScript: one fixed dialect (GFM's everyday syntax without the parts that need backtracking, plus directives with `{…}` attributes, math, `${…}` expressions and a metadata block), a compact flat source-mapped AST, HTML output, and no configuration. Make the Markdown decision once; use markz everywhere.

import { parse } from 'markz';

const document = parse(markdown);

The dialect is in [`prose/syntax.md`](prose/syntax.md) and the design in [`prose/spec.md`](prose/spec.md).

Not a drop-in Markdown parser

markz parses the Markdown we write, not every Markdown in the wild. Pasted or generated Markdown often uses syntax the dialect leaves out: reference links (`[text][ref]`), bare URLs, raw HTML, setext headings (`Title` over `===`), footnotes and named entities (`&`). markz never guesses at these. Each one stays literal text and adds a diagnostic naming the supported form, so converting a document means fixing what `doc.diagnostics` lists.

Development

# needs Node 26 + pnpm 12.6+ — mise.toml pins both
mise install
pnpm install

`dev` (watch build), `build`, `check` (format, lint, typecheck) and `test` are the whole interface. The site is in [`docs/`](docs/): `pnpm docs` runs it, and its Conformance page shows every spec example against the oracle.