Accessibility Auditor Skill

A portable AI skill for comprehensive accessibility auditing of websites, GitHub repositories, local project folders, and mobile/desktop app source code.

Built by a blind developer. Based on WCAG 2.1, WCAG 2.2 notes, GOST R 52872-2019, and the Russian Mincifry accessibility order.

What It Does

The skill guides an AI agent through a structured accessibility audit workflow. You give it a target - a URL, a GitHub repo, or a local path - and the agent:

1. Gathers the best available evidence from the current environment
2. Applies the appropriate checklist for the target platform
3. Reports issues by severity with specific fix recommendations
4. Maps findings to WCAG 2.1 / GOST R 52872-2019 criteria
5. Scores the result (0-100) and assigns a letter grade
6. Clearly states coverage gaps and what still needs runtime/manual verification

Portability Model

This skill is meant to work across different agent runtimes such as Hermes, OpenClaw, Codex, Claude Code, and similar systems.

To stay portable, it separates:

• portable core - workflow, references, report format, scoring, coverage rules
• runtime capabilities - fetch page content, render a page, read files, search files, run local audit tools

The skill should describe capabilities and outputs, not hard-code vendor-specific tool names.

Supported Targets

Target	How to provide	Analysis quality
Website	Any public URL	Static-first HTML/CSS/ARIA analysis, optionally richer with browser/audit tooling
GitHub repo	GitHub URL	Source code analysis for web, mobile, desktop
Local folder	Absolute or relative path	Same as GitHub - reads files directly

Supported Platforms

Web / Electron

• WCAG 2.1 AA checklist
• WCAG 2.2 notes where relevant
• GOST R 52872-2019 / Mincifry framing when applicable
• HTML structure, ARIA, keyboard navigation, forms, responsive and contrast heuristics

Mobile Apps (static source analysis)

Framework	What's checked statically
Android (XML + Kotlin/Java)	Missing contentDescription, labelFor, importantForAccessibility misuse
iOS (Swift / SwiftUI)	Missing accessibilityLabel, roles/traits issues, incorrect accessibilityHidden
Flutter (Dart)	Images without semanticLabel, GestureDetector without Semantics wrapper
React Native (JS/TSX)	Missing accessibilityLabel, accessibilityRole, nested touchables

Desktop Apps (static source analysis)

Framework	What's checked statically
Electron	Same as web - WCAG-oriented static checks apply
WPF / XAML	AutomationProperties.Name, HelpText, LabeledBy
PyQt / PySide	setAccessibleName(), setAccessibleDescription(), setBuddy()
wxWidgets / wxPython	SetName(), SetHelpText(), wxAccessible subclass
GTK (Python)	AT-SPI2 / AtkObject-related patterns where visible from source

Static analysis covers only part of accessibility for native apps. The remainder requires runtime/manual testing with screen readers and platform-specific inspectors.

Audit Depth Levels

Level A - zero-setup baseline

Works with the environment as-is:

• URL or source inspection
• static findings
• useful report right away

Level B - richer audit

If the environment already has more capability or tools:

• browser rendering
• better evidence for dialogs, menus, forms, dynamic UI
• richer source coverage

Level C - advanced follow-up

If the environment allows deeper tooling or optional installs:

• approved audit CLI tools
• framework/platform-native linting or testing
• broader coverage across more pages/modules

Approved Optional Tooling

If deeper checks are needed, the skill should prefer existing tools first. If installation is allowed, it should extend only from an approved allowlist.

Web / browser-based

• @axe-core/cli
• pa11y
• lighthouse
• playwright

Source inspection helpers

• ripgrep / rg
• standard framework linters already normal for the target stack

Platform-specific

See references/tools.md for the recommended tools by platform.

Skill Files

accessibility-auditor-skill/
├── README.md
├── SKILL.md
├── PORTABILITY.md
├── plan.md
└── references/
    ├── README.md
    ├── additional-checks.md
    ├── desktop-checklist.md
    ├── gost-checklist.md
    ├── international-standards.md
    ├── mobile-checklist.md
    ├── open-github-a11y-pr-cases.md
    ├── tools.md
    └── wcag-criteria.md

Install / Extension Policy

The skill should follow this order:

1. use the current agent environment first
2. deliver the best partial audit it can immediately
3. if richer checks are needed, extend only through the approved allowlist from SKILL.md and references/tools.md
4. never install arbitrary tooling outside that approved set

This keeps the skill low-friction without turning it into an uncontrolled installer.

Limitations

• Static analysis alone cannot fully prove accessibility
• Color contrast may need rendering or dedicated tools
• JavaScript-heavy sites may need browser rendering for deeper confidence
• Native mobile/desktop apps require runtime/manual testing for full coverage
• Large repositories should be audited by representative flows/components first

Related Project

This skill was developed alongside the Accessibility Auditor - a production service and Telegram bot for website accessibility audits.

This skill is independent. It is meant to work through AI-guided analysis plus optional approved tooling, not through the service API.