` block.\n v2's `mintlifyContainer` has no children for these, so only the intended\n component renders.\n\n## Not yet implemented (planned)\n\n- Per-component attribute schemas (currently reuses v1's global allowlist\n for `hProperties`; the full parsed attribute map is kept on the node).\n- Tags inside blockquotes/lists currently require blank lines around them\n (the normalizer only handles top-level tag lines); fixing this means\n splitting multi-line `html` nodes or normalizing per container.\n- Multi-line JSX open tags (an open tag with attributes spread across\n several lines; v1 joins them in `joinMultilineJsxTags`).\n- Attribute naming convention: v1 prefixes hProperties with\n `data--` (e.g. `data-card-href`), presumably because its\n output was raw custom HTML elements read via `dataset`. v2 intentionally\n uses raw names (`href`) since containers render through React components\n that read props directly — this is a deliberate v2 API change, not a gap\n to close.\n- Registry publish. The build (`npm run build` → `dist/` with `.d.ts`,\n exports pointing at it, `styles.css` copied alongside), the `inkstream`\n CLI (see below), and CI (typecheck/test/build/golden, plus a check\n that committed `dist/` matches a fresh build) are already in place;\n this package isn't published to the npm registry yet (`\"private\":\n true`) — consumed via a GitHub commit pin instead, which is why\n `dist/` is committed rather than gitignored: npm's `prepare` script\n (which would otherwise build it automatically on install) only runs\n when the installing environment allows lifecycle scripts, and plenty\n of reasonable npm configs set `ignore-scripts=true`. `prepare` still\n runs `npm run build` for anyone who does allow scripts (and will\n matter again once this is a real registry publish, where it runs on\n the *publisher's* machine regardless of the installer's setting).\n\n## CLI\n\n```sh\nnpx --package=github:catatsumuri/inkstream inkstream [file]\n```\n\nor, once installed as a project dependency, `npx inkstream ...` /\ndirectly via the `inkstream` bin.\n\n| Command | Output | Use case |\n| --- | --- | --- |\n| `inkstream render ` | HTML | RSS feeds, OGP static pages, email bodies. Same pipeline as the golden corpus renderer, so no heading ids (those come from the React heading renderer, not this pipeline) |\n| `inkstream text ` | Plain, human-readable text | Full-text search indexing, excerpts, OGP `description` — the raw markdown source is unusable for these as-is (``, ` ```quiz ` syntax noise) |\n| `inkstream headings [--json] [--prefix=]` | Indented outline, or a JSON array with `--json` | Server-side table-of-contents precomputation; same shape as `extractMarkdownHeadings` |\n\nEvery command reads markdown from the given file path, or from stdin\nwhen the path is `-` or omitted.\n\n`text` resolves the same syntax `render` does (Mintlify tags, Zenn\ndirectives, wikilinks) down to prose: a quiz fence contributes its\nquestion/options/explanation, a chart fence its title and labels (not\nthe numeric values), a tree fence its file and folder names, and a\nwikilink its label (or the path's last segment) — none of these have a\nsensible plain-text form otherwise, so they're handled explicitly rather\nthan silently dropped or dumping raw JSON. Wikilinks resolve without\nneeding a `resolveWikilink` callback, the same way heading slugs do.\n\n## Known limitations (not planned)\n\nEdge cases identified during review, deliberately left unhandled: no\noccurrence in this project's real documents justified the added parsing\nsurface, and each is easy to work around in authoring.\n\n- **Setext headings** (`Title` underlined by `===`/`---` instead of a\n leading `#`) aren't picked up by `extractMarkdownHeadings` — a\n table of contents silently omits them, even though they render fine\n as `
`/``. Write headings with `#`/`##` instead.\n- **`>` inside a tag attribute value** (` b\">`) breaks\n `match-tags.ts`'s line-regex tag matcher, since it isn't a real HTML\n parser. Avoid a literal `>` in attribute values (use `>` or\n rephrase).\n- **Single-quoted attribute values** (``) aren't\n parsed by `parseJsxAttributes`' bare/quoted branches (only `\"x\"` and\n brace-wrapped `{'x'}` are). Use double quotes.\n\n## Golden corpus (`golden/`)\n\nSnapshot regression suite: renders every `golden/corpus/*.md` fixture\nthrough the full pipeline and compares the prettified HTML against the\ncommitted baseline in `golden/baseline/*.html`.\n\n```sh\nnpm run golden # check; exits non-zero on any diff\nnpm run golden -- --update # re-render and accept the current output\n```\n\n- `golden/render-v2.ts` — the pipeline under test\n (`normalizeInkstreamMarkdown` → `inkstreamRemarkPlugins` →\n `remark-rehype` → `hast-util-to-html`), the same exports consumers use.\n- `golden/diff.ts` — prettifies output to one tag per line, line-diffs it\n against the baseline, and writes the current output to\n `golden/output/*.html` (gitignored) for inspection.\n\nFixtures `01`–`09` are synthetic; `10`–`14` are the five real syntax-guide\ndocuments from thinkstream's `SyntaxSeeder` (basic markdown, extended\nmarkdown/GFM, Zenn syntax, Mintlify syntax, thinkstream syntax) — a far\ntruer signal, and the corpus that surfaced the fence-state, inline-code,\nindentation, and array-attribute bugs the current pipeline fixes.\n\nThe corpus originally drove a v1-vs-v2 comparison (rendering the same\nfixtures through inkstream v1's directive pipeline); once every difference\nwas either fixed or decided as an intentional v2 change, the v1 renderer\nand its `file:` dependency were dropped and the v2 output became the\nbaseline.\n","readmeFilename":"README.md"}