v0.13.2 — HTML5 normalizer (seven rules) and substitution architecture

Added

• HTML5 normalizer for htmlToOdt() — htmlToOdt(), markdownToOdt(), and the underlying parseHtml() now run input through a Tier 1 normalizer before parsing. The normalizer applies seven spec-grounded text transformations: empties <script> and <style> content; lowercases the doctype declaration; quotes unquoted boolean attributes (e.g. <input checked> → <input checked="">); quotes unquoted attribute values (e.g. <a href=foo> → <a href="foo">); self-closes 14 HTML5 void elements; decodes ~2,120 HTML5 named entities to Unicode; and escapes lone & in attribute values (e.g. href="?a=1&b=2" → href="?a=1&b=2"). Good HTML5 input that previously produced silent empty output now converts correctly. Default behaviour for editor-generated polyglot HTML is unchanged — the normalizer is idempotent on already-polyglot input.
• Substitution architecture — htmlToOdt() and markdownToOdt() accept new normalizer and parser options. Pass false to skip normalization (when input is known polyglot/XHTML), or pass a custom function to substitute either stage. tiptapToOdt() does not expose these hooks because TipTap input is a JSON tree, not an HTML string. See ADAPTERS.md at the repo root for the architecture, naming conventions, and a worked parse5 adapter example.
• odfKitNormalizer — the default normalizer, exported from the root and from the new odf-kit/html-normalizer sub-export. The seven individual rules are also exported: selfCloseVoidElements, decodeNamedEntities, emptyRawTextElements, lowercaseDoctype, quoteUnquotedBooleanAttributes, quoteUnquotedAttributeValues, escapeAttributeValueAmpersands.
• odfKitParser — the default parser (a Parser-conforming wrapper around the existing parseXml), exported from the root.
• Public types — ParsedHtmlTree, Parser, Normalizer exported from the root for adapter authors. NormalizerOption and ParserOption are available via the odf-kit/types path for symmetric architectural use.
• OdtBaseOptions — base interface with shared fields (page format, orientation, margins, metadata, image resolution). HtmlToOdtOptions extends it and adds normalizer and parser. TiptapToOdtOptions extends OdtBaseOptions directly. No user-facing API change for code using the existing options.
• odf-kit/html-normalizer sub-export added.
• ADAPTERS.md at the repo root — documents the substitution architecture: philosophy, six naming conventions, the two-direction adapter principle, skip semantics, contract specifications, versioning promise, sibling-package design, and a complete worked parse5 adapter example.
• 183 new tests (1307 total, 28 test suites).

Changed

• parseXml now fails loudly on malformed input — five tightenings: detects unclosed elements at end-of-input; rejects malformed attribute syntax that the normalizer didn't cover; rejects unescaped & in attribute values not followed by a valid XML entity or numeric reference; rejects ]]> in text content outside CDATA sections; rejects mismatched closing tags. Previous behaviour was silent wrong output. Code that worked in v0.13.1 continues to work in v0.13.2 — the new errors surface latent bugs in inputs that were producing incorrect output. Treat the new errors as bugs in your input that v0.13.2 makes visible.
• package.json — lexical entry in typesVersions had a missing leading ./; corrected.

Migration

No code changes required for typical users. Default behaviour is preserved for editor-generated and polyglot input, and existing options work unchanged. Three situations may warrant attention:

• If your htmlToOdt() calls were silently producing empty or wrong output on hand-written HTML5, v0.13.2 fixes that automatically — the seven-rule normalizer handles void elements, named entities, boolean attributes, unquoted attribute values, ampersands in URLs, and other HTML5-vs-XHTML differences before parsing.
• If your input was triggering one of the malformed-input cases above, you'll now see an explicit error. This is the intended behavior — the previous silent corruption is the bug being fixed.
• If you were calling tiptapToOdt() with normalizer or parser in the options object, those properties were always ignored at runtime and are now compile-time errors. Remove them.