Content standards checker for dev environments

Check content is a command line interface (CLI) tool that scans code bases for content issues. First it extracts text from HTML, JSX, or TSX files, then it auto-detects industry, and finally it sends extracted content paired with a system prompt to a GPT-4o API. The API evaluates the text against industry best-practice content standards, inclusive language, plain language, and accessibility standards, returning violations with suggestions.

It's one of my answers to the decline in cross-functional consultation.

Go on, have a demo

Pattern matching industry detection

Check content attempts to automatically detect an app's industry so that it can provide industry-specific content standards.

First, it scans for definitive indicators like stock symbols or medical terms that clearly identify an industry. If it doesn't find any, it falls back to weighted pattern matching across content, dependencies, and file structures.

function detectIndustry(codebaseData) {
  const allText = combineAllSources(codebaseData);

  for (const [industry, patterns] of industries) {
    // Check definitive indicators first
    const definitiveMatches = patterns.definitiveIndicators
      .filter(term => allText.includes(term.toLowerCase()));

    if (definitiveMatches.length > 0) {
      return {
        industry,
        confidence: 0.9 + bonusCalculation(definitiveMatches),
        source: 'definitive_indicators'
      };
    }

    // Fall back to pattern matching
    const patternScore = calculatePatternScore(codebaseData, patterns);
  }
}

The tool currently supports six industries, each with specialized rules. Healthcare apps get HIPAA compliance content standards and person-first language checks. Finance gets regulatory compliance and risk disclosure rules. E-commerce apps get product clarity and conversion optimization guidance.

Refinement: fintech trading platform

I tested Check content with a friend's cryptocurrency trading prototype. I was intrigued to see that the industry detection algorithm consistently classified it as a gaming app with 70% confidence even though there was clear financial terminology.

After investigating, it turned out that application contained gaming terms like "Alpha Leaderboard", "Win Rate", and "Trading Arena" that overwhelmed financial indicators like "Stock Symbol", "TSLA Position Alert", and "Total Portfolio", resulting in an inaccurate analysis.

// Content that triggered gaming detection:
"Alpha Leaderboard"     // Strong gaming signal
"Win Rate"              // Gaming terminology
"Trading Arena"         // "Arena" suggests competitive gaming

// Financial signals that were overwhelmed:
"Stock Symbol"          // Should indicate finance
"TSLA Position Alert"   // Clear stock reference
"Total Portfolio"       // Financial terminology

I used Claude Code and some common sense to enhance the definitive indicator system to prioritize financial terms and added special handling for stock symbols.

Standards and analysis

The standards are based on three authoritative sources:

Atlassian Design System - Bold, optimistic, practical voice with inclusive language
Atlassian Inclusive Writing Guidelines - Gender-neutral pronouns, avoiding exclusionary terms
California Innovation Hub Principles - Plain language (8th grade reading level), conversational tone

The system prompt is a ~210-line constant that is passed to GPT-4o as the system message in the OpenAI chat completion call.

Key Standards Categories

Category	Severity	Examples
Inclusive Language	error	guys→everyone, blacklist→blocklist, manpower→workforce
Pronoun Guidelines	warning	he/she→they, his/her→their
Plain Language	warning	utilize→use, facilitate→help, paradigm→model
Active Voice	warning	"was created"→"created", "is being processed"→"processes"
Missing Alt Text	error	Images without alt attributes
Non-descriptive Links	warning	"click here"→descriptive text

Prohibited Terms (Partial List)

The system prompt includes 50+ prohibited terms:

Gender-exclusive:

guys, dudes, bros → team, everyone, folks
mankind → humanity, people
manpower → workforce, staff
businessman, freshman → business people, first-year student

Tech legacy terms:

master/slave → primary/secondary, main/replica
blacklist/whitelist → blocklist/allowlist

Ableist language:

crazy/insane → surprising, unexpected
blind spot → gap, oversight
lame → disappointing, weak

Jargon to simplify:

utilize, leverage → use
facilitate → help, enable
optimize, streamline → improve, simplify
synergize, ideate → work together, brainstorm
paradigm, methodology → model, method