accessibility-audit

This skill should be used when evaluating a design, page, or component against accessibility standards and producing a severity-scored report. Trigger phrases include "audit accessibility", "check a11

SKILL.md

GitHub

Accessibility audit

Evaluate a design, page, or component against WCAG 2.2 Level AA and produce a severity-scored report with a remediation for every finding. This skill diagnoses and grades; when a fix means writing code, hand off to the accessible-components skill.

When to use

Use when the user asks to "audit accessibility", "check a11y", run a "WCAG audit", "check color contrast", confirm "is this accessible", or do an "accessibility review before handoff". Also use proactively before a design ships to engineering.

Inputs

Work from whatever is available, and say which mode you used:

  • Source code (HTML/JSX/Vue/etc.) — inspect markup, roles, attributes, and styles directly. Most thorough.
  • A screenshot — assess visual checks (contrast, target size, focus indication if captured, text legibility); flag what cannot be verified visually (keyboard, names, sequence) as "needs manual test".
  • A description — reason about likely issues and list the manual checks the user must run.
  • A live URL or local HTML file — run automated checks (see below) AND the manual walkthrough.

Method: POUR

Walk the four WCAG principles in order. The full per-criterion checklist with criterion numbers and test steps is in references/wcag-2.2-aa-checklist.md — load it and work through it.

  • Perceivable — text alternatives (1.1.1), captions/audio (1.2.x), info-and-relationships and meaningful sequence (1.3.1, 1.3.2), use of color not alone (1.4.1), contrast (1.4.3 text, 1.4.11 non-text), resize/reflow/text-spacing (1.4.4, 1.4.10, 1.4.12), content on hover/focus (1.4.13).
  • Operable — keyboard (2.1.1) and no trap (2.1.2), enough time (2.2.x), no flashing (2.3.1), bypass blocks / headings / focus order / link purpose / multiple ways (2.4.x), focus not obscured (2.4.11), focus appearance, pointer gestures and dragging movements (2.5.7), target size 24px minimum (2.5.8).
  • Understandable — language (3.1.x), on-focus/on-input no surprise (3.2.x), consistent help (3.2.6), error identification/labels/suggestions (3.3.1–3.3.4), redundant entry (3.3.7), accessible authentication (3.3.8).
  • Robust — valid parsing, name/role/value (4.1.2), status messages (4.1.3).

Manual checks the tools cannot catch

Always do these by hand (or instruct the user to):

  1. Keyboard-only walkthrough — unplug the mouse. Reach and operate every control with Tab/Shift+Tab/Enter/Space/arrows. Nothing should be unreachable or trapped.
  2. Focus order — Tab order must follow a logical, meaningful reading sequence; no jumps that lose the user.
  3. Visible focus — every focusable element shows a clear focus indicator that is not clipped or hidden behind sticky headers (2.4.11).
  4. Screen-reader names — with VoiceOver/NVDA, confirm each control announces a sensible name, role, and state. Icon-only buttons are the usual failure.
  5. Meaningful sequence — DOM order matches the intended reading order independent of CSS positioning.
  6. Error identification — trigger validation; confirm errors are announced, described in text (not color alone), and tied to their field.

Automated assist

Use the bundled scripts to verify what is mechanical. Treat automation as a floor, not a verdict — it catches roughly a third of issues.

  • Color contrast — node scripts/contrast.mjs <foreground> <background> for any color pair. Reports the ratio and pass/fail for AA/AAA normal, AA/AAA large, and UI/non-text. Example: node scripts/contrast.mjs "#777" "#ffffff" → ~4.48 (FAIL AA normal, PASS AA large).
  • Live page scan — node scripts/axe-scan.mjs <url-or-file> runs axe-core via puppeteer when puppeteer and @axe-core/puppeteer are installed; it groups violations by impact. If those deps are absent it prints install instructions and exits cleanly, so it never breaks an audit.

Scoring

For each finding, assign:

  • Severity — one of:
    • Critical — blocks a task entirely for some users (e.g. unlabeled submit button, keyboard trap, content unreachable).
    • Serious — major barrier with a hard workaround (e.g. body text below 4.5:1, missing form labels, no visible focus).
    • Moderate — degrades the experience but task is completable (e.g. vague link text, minor heading-order gaps).
    • Minor — polish or edge case (e.g. redundant alt text, slightly-small decorative target).
  • Criterion — the mapped WCAG success criterion number and name (e.g. 1.4.3 Contrast (Minimum)).
  • Remediation — a concrete fix. If the fix is code, name it and hand off to accessible-components.

Then compute a summary score: a count by severity and an overall verdict for AA:

  • Fail — any Critical or Serious finding open.
  • Conditional — only Moderate/Minor findings, or items that need manual confirmation.
  • Pass — no findings at or above Minor that affect conformance and all manual checks passed.

Output format

Produce a structured report:

# Accessibility audit — <target>
Mode: <code | screenshot | description | live>   Standard: WCAG 2.2 AA

## Summary
| Severity | Count |
|----------|-------|
| Critical | n |
| Serious  | n |
| Moderate | n |
| Minor    | n |
Verdict: Fail | Conditional | Pass

## Findings
### Critical
1. <title>
   - Criterion: 1.4.3 Contrast (Minimum)
   - Location: <selector / region / screenshot area>
   - Impact: <who is blocked and how>
   - Fix: <remediation; hand off to accessible-components if code>
### Serious
...

Group findings by severity (Critical first). End with the manual checks still outstanding, if any.

References

  • references/wcag-2.2-aa-checklist.md — full POUR checklist with criterion numbers and test methods.
  • scripts/contrast.mjs — contrast ratio calculator.
  • scripts/axe-scan.mjs — axe-core live scan (graceful if deps absent).
  • For stack-aware code fixes, follow ../ux-foundations/references/stack-detection.md and the accessible-components skill.