@faircopy/rules-nlp

Optional NLP-powered ruleset for faircopy using compromise.

npm i -D @faircopy/rules-nlp

Load the ruleset once, then configure rules with bare rule IDs:

rulesets: ['@faircopy/rules-nlp'],
rules: {
  'no-absolute-intensifiers': 'warn',
  'no-adverb-overuse': 'warn',
  'no-expletive-openers': 'warn',
  'no-filter-words': 'warn',
  'no-future-promises': 'warn',
  'no-hedge-words': 'warn',
  'no-empty-transformation-claims': 'warn',
  'no-passive-voice': 'warn',
  'no-qualifier-creep': 'warn',
  'no-redundant-pairs': 'warn',
  'no-vague-comparatives': 'warn',
  'no-weak-modals': 'warn',
  'no-stacked-adjectives': 'warn',
  'no-nominalized-phrases': 'warn',
  'no-noun-strings': 'warn',
  'no-pronoun-led-claims': 'warn',
  'no-buzzword-stacks': 'warn',
  'no-complex-readability': 'warn',
  'no-overly-complex-sentences': 'warn',
  'no-vague-quantifiers': 'warn',
  'no-meaningless-modifiers': 'warn',
  'no-superlative-claims': 'warn',
  'no-llm-speak': 'warn',
  'no-n-things': 'warn',
  'no-weak-verbs': 'warn',
}

Package-qualified IDs like @faircopy/rules-nlp/no-passive-voice still work and are required if another loaded ruleset exposes the same bare rule name.

Rules

no-absolute-intensifiers

Absolute adjectives already express an extreme, so intensifiers such as very are redundant.

rules: {
  'no-absolute-intensifiers': 'warn',
}

Options:

Example:

'no-absolute-intensifiers': ['warn', {
  intensifiers: ['very', 'highly'],
  absolutes: ['unique', 'critical'],
}]

sentence-complexity

Long, clause-heavy sentences are harder to read. This rule flags any sentence that exceeds a configurable word count or contains too many finite-verb clauses, and suggests splitting it into shorter sentences.

rules: {
  'sentence-complexity': 'warn',
}

Options:

Example with custom thresholds:

'sentence-complexity': ['warn', {
  maxWordCount: 20,
  maxClauseCount: 2,
}]

Flagged example:

The engineer reviewed the requirements and wrote the code and ran the tests and deployed the application while the team watched and celebrated the release together.

Suggested fix: split the sentence around each independent clause.

The engineer reviewed the requirements. They wrote the code, ran the tests, and deployed the application. The team watched and celebrated the release together.

no-overly-complex-sentences

Sentences packed with conjunctions are often run-ons or nested too deeply. This rule flags any sentence that exceeds a configurable number of coordinating conjunctions, subordinating conjunctions, or total conjunctions.

rules: {
  'no-overly-complex-sentences': 'warn',
}

Options:

Example with custom thresholds:

'no-overly-complex-sentences': ['warn', {
  maxConjunctions: 3,
  maxCoordinating: 2,
  maxSubordinating: 1,
}]

Flagged example:

I ran and swam and biked and hiked and climbed.

Suggested fix: split the list into shorter sentences or use a bulleted list.

I ran, swam, and biked. Then I hiked and climbed.

Some words, such as since or so, can be prepositions or adverbs in certain contexts. If the rule is too noisy for your copy, add the word to allowList.

no-vague-comparatives

Comparative claims like faster, easier, or more efficient only persuade when the reader knows the baseline. This rule flags comparative adjectives and adverbs that are not paired with than in the same sentence.

rules: {
  'no-vague-comparatives': 'warn',
}

Options:

Example with custom settings:

'no-vague-comparatives': ['warn', {
  comparatives: ['superior', 'inferior', 'better'],
  requireThan: true,
}]

Flagged example:

Our editor is faster. It is also easier to use.

Suggested fix: add a concrete baseline.

Our editor is faster than the old one. It is also easier to use than a spreadsheet.

no-qualifier-creep

Stacked qualifiers or intensifiers dilute a claim and make it sound hedged. This rule flags sequences of two or more qualifiers before an adjective or adverb.

rules: {
  'no-qualifier-creep': 'warn',
}

Options:

Example with custom settings:

'no-qualifier-creep': ['warn', {
  qualifiers: ['very', 'really', 'quite'],
  maxQualifiers: 1,
}]

Flagged example:

The result is very really quite good.

Suggested fix: keep the strongest qualifier or replace the phrase with concrete evidence.

The result is excellent.

no-non-inclusive-language-nlp

Non-inclusive terms can alienate readers. This rule flags common non-inclusive language and suggests neutral alternatives using NLP-aware matching so it understands word boundaries and part-of-speech context.

rules: {
  'no-non-inclusive-language-nlp': 'warn',
}

Options:

Example with custom terms:

'no-non-inclusive-language-nlp': ['warn', {
  terms: [
    { term: 'guys', alternatives: ['everyone', 'team', 'folks'] },
    { term: 'blacklist', alternatives: ['denylist', 'blocklist'] },
  ],
  allowedTerms: ['whitelist'],
}]

Flagged example:

Hey guys, add this to the blacklist and do a sanity check.

Suggested fix:

Hey everyone, add this to the blocklist and do a quick check.

The rule uses compromise to avoid partial-word matches (e.g., guyses is not flagged) and skips verb-only usages of ambiguous words such as master in master the skill.

no-llm-speak

LLM-generated copy often falls back to padded, abstract phrases that sound authoritative without saying much. This rule flags common LLM clichés and suggests concrete replacements.

rules: {
  'no-llm-speak': 'warn',
}

Options:

Example with custom phrases:

'no-llm-speak': ['warn', {
  phrases: [
    { phrase: 'delve into', alternatives: ['explore', 'examine'] },
    { phrase: 'robust', alternatives: ['strong', 'resilient'] },
  ],
  allowedPhrases: ['crucial'],
}]

Flagged example:

It's important to note that we will delve into the robust, intricate tapestry. Furthermore, the system is crucial.

Suggested fix:

Note that we will explore the strong, detailed pattern. Also, the system is critical.

The rule uses compromise for word-aware matching, so robustness is not flagged when the configured phrase is robust.

no-weak-verbs

Vague action verbs such as make, perform, conduct, and carry out hide the real action. This rule flags weak verbs used as verbs and suggests stronger, more specific alternatives. It uses compromise to identify verb inflections, so made, performed, and carried out are also flagged, while noun usages such as the make of the car are ignored.

rules: {
  'no-weak-verbs': 'warn',
}

Options:

Example with custom verbs:

'no-weak-verbs': ['warn', {
  verbs: [
    { verb: 'utilize', alternatives: ['use'] },
    { verb: 'leverage', alternatives: ['use', 'exploit'] },
  ],
  allowList: ['make'],
}]

Flagged example:

We need to make a decision, perform a review, and carry out the plan.

Suggested fix:

We need to decide, review, and implement the plan.

Because the rule matches by verb inflection, it may also flag common phrasal verbs and idioms such as make out, make do, or made it. Add those to allowList if they are acceptable in your style guide.

no-noun-strings

Three or more nouns in a row create dense, hard-to-read compounds. This rule flags consecutive noun strings and suggests rewriting them with verbs or prepositions.

rules: {
  'no-noun-strings': 'warn',
}

Options:

Example with custom settings:

'no-noun-strings': ['warn', {
  maxConsecutiveNouns: 4,
  allowedPhrases: ['user experience design process'],
}]

Flagged example:

The user experience design process is slow.

Suggested fix:

The process for designing the user experience is slow.

The rule uses compromise for part-of-speech tagging and skips proper nouns such as place names.

no-n-things

A count plus a placeholder noun — Three things., 5 things to know before launch — borrows its number for rhythm instead of meaning. The frame is a documented AI-writing tell: the count is chosen because enumerated lists feel punchy, not because the content demanded that many items. Name the items in the heading, or drop the count and state the claim.

rules: {
  'no-n-things': 'warn',
}

Options:

Example with custom settings:

'no-n-things': ['warn', {
  minCount: 3,
  nouns: ['thing', 'things', 'tips', 'takeaways'],
}]

Flagged example:

Three things you should know about memory.

Suggested fix:

How Ditto remembers your conversations.

The rule uses compromise for part-of-speech tagging. Only cardinal counts trigger it — idioms like First things first, All things considered, and a few things are left alone, as are concrete counted nouns like three tabs.