- Home
- Skills
- Code Quality & Review
- WCAG 2.2 AA Audit & Fix
Works with the AI Agents you already use
WCAG 2.2 AA Audit & Fix
Tested on gov.uk (100/A), Wikipedia, Hacker News, python.org and craigslist.
$14
WCAG 2.2 AA Audit & Fix
Example session with this skill installed
Audit examples/inaccessible.html for WCAG 2.2 AA problems and tell me what to fix first.
- Read your context and instructions
- Compiled the wcag 2 2
- Generated the UI component
Accessibility audit: examples/inaccessible.html
Score 0/100 (grade F). 39 issues: 10 critical, 19 serious, 10 moderate.
[CRITICAL] A11Y013 Viewport blocks zooming (WCAG 1.4.4 Resize Text, Level AA)
examples/inaccessible.html:4 <meta name="viewport">
Viewport content "width=device-width, initial-scale=1, maximum-scale=1, user-scalable=no" stops users zooming.
Fix: Use <meta name="viewport" content="width=device-width, initial-scale=1"> (remove user-scalable=no and maximum-scale).
[CRITICAL] A11Y004 Button has no accessible name (WCAG 4.1.2 Name, Role, Value, Level A)
examples/inaccessible.html:16 <button>
<button> has no text, aria-label or title.
Fix: Add visible text, or aria-label="Close" (etc.) for icon-only buttons; mark the icon aria-hidden="true".
[CRITICAL] A11Y003 Form field has no label (WCAG 1.3.1 Info and Relationships / 4.1.2 Name, Role, Value, Level A)
examples/inaccessible.html:32 <input name="email" type="text">
<input name="email" type="text"> is labelled only by its placeholder, which disappears when typing.
Fix: Add a visible <label>; keep the placeholder only as an example value.
Fix first
- Zoom is blocked (line 4). One-line fix, affects every mobile user with low vision.
- 5 form fields have no label (lines 32 to 40). Screen reader and voice control users cannot fill in the form.
- 3 images and 1 icon button have no name (lines 14, 15, 16 and 38).
- Low contrast: 3 text styles, for example white on #ffcc00 is 1.51:1 and needs 4.5:1.
After the fixes in examples/accessible.html the same audit reports 0 issues, 100/A.
Still to check by hand: keyboard-only navigation and visible focus, screen reader announcements, and 400% zoom reflow.
inaccessible.html
TSX · React component
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Find and fix accessibility failures before a client, a user or a lawyer does
WCAG Accessibility Auditor gives your agent a real audit engine instead of guesswork. It ships a single Python script (standard library only) that checks HTML against WCAG 2.2 Level AA, plus a workflow that tells the agent how to triage, fix the source files, re-run and report.
Works on:
- Local files and build folders (
dist/,build/,out/,_site/), from any framework that outputs HTML: Next.js export, Astro, Hugo, Jekyll, Eleventy, plain HTML, email templates - Live URLs (server-rendered HTML, plus up to 8 linked stylesheets for contrast)
What it checks (28 rules)
| Area | Examples |
|---|---|
| Images | missing alt, alt that is a file name, unnamed SVGs, image inputs |
| Forms | unlabelled fields, placeholder-only labels, radio groups without a legend, missing autocomplete on personal data |
| Structure | lang, title, skipped or empty headings, missing h1 and main landmark, data tables without headers |
| Links and buttons | empty links and buttons, "click here" and "read more" |
| Keyboard and ARIA | positive tabindex, click handlers on divs, invalid roles, aria references to missing ids, focusable items inside aria-hidden, duplicate ids |
| Visual and media | colour contrast with a suggested passing colour, zoom blocked in the viewport, autoplay, missing captions, untitled iframes, meta refresh |
Every finding includes the rule, impact (critical, serious, moderate, minor), the WCAG criterion and level, file and line, the offending element, who it hurts and the exact fix.
Outputs
- text for the terminal
- json for the agent and other tools
- md: a client-ready report with a score, an issues-by-rule table and a "what automated testing cannot confirm" section
- CI gate:
--fail-on seriousexits 1, plus a ready GitHub Actions workflow that posts the report to the job summary
Built to avoid false alarms
It skips elements hidden by CSS, hidden or inert, understands accessible names from aria-label, aria-labelledby, alt and title, and when a complex CSS selector might change an element's colours it skips that contrast check instead of guessing. Tested on gov.uk (100/A), Wikipedia, Hacker News, python.org and craigslist.
Honest about limits
Automated checks catch only part of WCAG. The skill includes a manual checklist (keyboard, screen reader, zoom, motion, forms) and the agent lists the relevant items in every report. It never claims a page is "fully compliant".
Included
scripts/a11y_audit.py(the engine)SKILL.md(agent workflow and report format)references/rules.md,references/manual-checklist.md,references/ci-github-actions.ymlexamples/with a broken page (0/F, all 28 rules) and its fixed version (100/A)
Reads HTML as served and does not run JavaScript: for single-page apps, audit the built or pre-rendered output. Contrast checks use simple CSS selectors and skip text over images or where complex selectors may change colours. Automated checks cover only part of WCAG 2.2 AA, so a manual checklist is included; this is not a legal compliance certification.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
Requires
Permissions
Shell runs the bundled Python script. Read Files reads the HTML you point it at. Write Files is used when the agent fixes your markup or saves a report. Network Access is only used when you audit a live URL (fetches that page and its stylesheets); local audits make no network calls.