@xsynaptic/remark-prose-rules

Checks prose for missing diacritics and off-style terminology, in the remark pipeline, so problems show up in the editor with a suggested fix rather than only in a batch run.

Three rules ported from textlint (diacritics, terminology, pattern, MIT, Artem Sapegin). Not a textlint replacement: no rule ecosystem, no default terminology list, no suppression comments. ESM-only.

import { diacriticsMarks, diacriticsWords, remarkProseRules } from '@xsynaptic/remark-prose-rules';

remarkProseRules({
	// Correct spellings; the misspellings are worked out from them, and casing is kept
	words: [...diacriticsWords, 'mañana'],
	// Letters a spelling may be missing its mark on, plain letter last
	marks: [...diacriticsMarks, 'ōo'],
	// Case-insensitive, and skipped inside a longer word like `postwar-era`
	terms: [['metre', 'meter']],
	// Plain regex with a $n replacement
	patterns: [{ message: 'Use `--` for ranges', pattern: '(\\d)-(\\d)', replace: '$1--$2' }],
	// Nothing inside these is checked
	skip: ['blockquote'],
	// Frontmatter fields to check; `true` reuses the list above, an array replaces it
	frontmatter: { title: { words: true }, 'links[].title': { terms: [['metre', 'meter']] } },
});

Every option is empty by default, so nothing is corrected until asked for. All the rules share one skip list; to scope them differently, add a second instance with its own options.

Only body text is checked, so code, URLs and JSX attributes are left alone. Frontmatter needs remark-frontmatter and is checked only for the fields listed above, where a dot goes into a nested field and [] covers every item of a list. Those fields are edited in place, and a quoted value is reported but never changed.

Only spellings that need a mark from marks can be corrected, and passing marks replaces the built-in groups rather than adding to them. Anything transliterated, like Japanese or pinyin, is better handled by terms, which suggests instead of rewriting.

words and patterns fix as they go, while terms only reports, unless the processor carries data('editorialFixes', true). That keeps wording changes out of format-on-save. Fixes reach the file only if the pipeline ends in remark-stringify and the result is written back; behind an HTML compiler they fix the page and leave the file wrong.

MIT