review-a11y

Accessibility Review - WCAG Conformance Audit

Advisory-only accessibility audit. Dispatches a QA - Web A11y Reviewer agent to evaluate web content against WCAG 2.2 Level AA, prioritize issues by real-world user impact, and recommend fixes. No changes are made.

Philosophy

Impact over compliance. A missing form label that blocks screen reader users is more important than a redundant ARIA role that harms nobody. The audit prioritizes barriers that prevent or degrade access for people with disabilities.

Diagnostic, not therapeutic. This skill identifies accessibility barriers and recommends fixes. It does not implement them. Ask for changes directly after reviewing the findings, or use /implement with the audit as context.

Workflow Overview

┌──────────────────────────────────────────────────────┐
│               ACCESSIBILITY REVIEW                   │
├──────────────────────────────────────────────────────┤
│  1. Detect web content                               │
│  2. Determine scope                                  │
│  3. Dispatch accessibility auditor                   │
│  4. Present audit report                             │
└──────────────────────────────────────────────────────┘

Workflow Details

1. Detect Web Content

Scan the project for web content files:

Extensions	Content type
.html, .htm	HTML documents
.jsx, .tsx	React/JSX components
.vue	Vue components
.svelte	Svelte components
.css, .scss, .less	Stylesheets
.ejs, .hbs, .pug, .njk	HTML templates

Exclude: node_modules/, vendor/, dist/, build/, .git/, and other generated/vendored directories.

If no web content is found: Report "No web content detected in this project. Accessibility audits apply to projects with HTML, CSS, or UI component files." and abort.

2. Determine Scope

Present detected web content to the user with file counts:

Detected web content:
  - HTML: 12 files
  - React (TSX): 28 files
  - CSS: 8 files

What should I audit?
  - Entire project (default)
  - Specific directory (e.g., src/components/)
  - Specific files

Accept the user's selection. Default: entire project.

3. Dispatch A11y Auditor

Assess scope size with Glob.

Small scope (roughly ≤20 web content files): Spawn a single QA - Web A11y Reviewer agent with the full scope.

Large scope (roughly >20 web content files): Partition by directory or component boundary. Spawn multiple QA - Web A11y Reviewer agents in parallel, each with a focused partition.

Prompt for each agent:

Audit the following web content for accessibility issues.
Scope: [partition or full scope]

Read all files in scope and perform your full audit:
1. Detect and run any automated accessibility tooling (axe-core, pa11y, etc.)
2. Manual inspection for issues automated tools miss (keyboard navigation,
   semantic correctness, dynamic content, ARIA usage, content quality, media)
3. Classify every issue by severity (CRITICAL / HIGH / LOW)

Produce your standard output format with summary, issues by severity,
and tooling recommendations.

4. Present Audit Report

Collect all agent responses. If multiple agents were dispatched, merge findings into a single report, deduplicating issues that span partitions.

Present a consolidated report:

## Accessibility Audit Report

Conformance target: WCAG 2.2 Level AA
Method: [automated + manual | manual only]
Scope: [what was audited]
Issues found: N (X critical, Y high, Z low)

### CRITICAL
[merged critical findings from all agents]

### HIGH
[merged high findings]

### LOW
[merged low findings]

### Tooling Recommendations
[if applicable — recommendations for adding automated accessibility testing]

### Suggested Next Steps
[Based on findings:
- If critical/high issues found: recommend fixing them, noting which
  would be handled by HTML changes vs CSS changes vs JS changes
- If no significant issues: "No significant accessibility barriers found"
- If no automated tooling exists: recommend adding axe-core or pa11y]

The report is your synthesis. If multiple agents were dispatched, look for patterns across partitions (e.g., missing labels is a project-wide habit, not an isolated incident). Don't just concatenate agent outputs.

Agent Coordination

• All auditor agents run in parallel (they are read-only and independent)
• Wait for all agents to complete before presenting the consolidated report
• If an agent fails or times out, note the failure in the report and continue with other results

Abort Conditions

Abort:

• No web content files detected in scope
• Not a git repository

Do NOT abort:

• No automated accessibility tooling available (proceed with manual audit)
• A single agent fails in a multi-agent dispatch (report failure, continue)
• Few issues found (report "no significant issues" — that's a valid outcome)