a11y-wcag22

WCAG 2.2 Accessibility Plugin for Claude Code

Audit, fix, and advise on web accessibility conformance with WCAG 2.2 (levels A, AA, AAA) directly in Claude Code. Version 2.0.0 embeds 71 deterministic checks across 9 categories, runs evaluations in parallel across files, caches results incrementally, and tracks progress across audits.

Install

claude plugin add <git-repo-url>

After installing, run /a11y-audit in any project to start auditing.

Architecture

/a11y-audit or /a11y-fix          /a11y-test
        |                                |
        v                                v
  +-----------+                   +-----------+
  |   Skill   |  Entry point      |   Skill   |  Entry point
  |  Validates|  Dispatches       |  Validates|  Dispatches
  |  args     |  agent            |  infra    |  test agent
  +-----+-----+                   +-----+-----+
        |                               |
        v                               v
  +-----------+                   +-----------+
  | a11y-agent|  Orchestrator     |a11y-test- |  Test runner
  |  - Scope  |  Coordinates      |  agent    |  Fixtures +
  |  - Cache  |  all work         |           |  regression
  |  - Route  |                   +-----------+
  |  - Agg    |
  |  - Report |
  +-----+-----+
        |
        | parallel subagents
        v
  +-----+-----+-----+
  |     |     |     |
  v     v     v     v
[eval][eval][eval][browser]
  |
  v
a11y-evaluate skill (per file)
a11y-apply skill (fix mode)
a11y-browser skill (--browser)

The entry point skill validates arguments and dispatches the agent. The agent orchestrates: it scopes files, checks the cache, classifies content, and launches parallel subagents — one per batch of files, one for browser verification. Subagents call the internal skills (a11y-evaluate, a11y-apply, a11y-browser). Results are aggregated, merged with cached findings, and written to reports. This flow is the same whether you audit one file or fifty.

Components

Component	File	Role
Audit skill	skills/a11y-audit/SKILL.md	User-facing entry point for /a11y-audit — validates args, dispatches agent
Fix skill	skills/a11y-fix/SKILL.md	User-facing entry point for /a11y-fix — validates args, dispatches agent
Test skill	skills/a11y-test/SKILL.md	User-facing entry point for /a11y-test — validates test infra, dispatches test agent
Evaluate skill	skills/a11y-evaluate/SKILL.md	Internal: per-file rule evaluator (71 checks embedded)
Apply skill	skills/a11y-apply/SKILL.md	Internal: per-file fix applicator
Browser skill	skills/a11y-browser/SKILL.md	Internal: browser-based verification via Playwright MCP
Agent	agents/a11y-agent.md	Orchestrator: scoping, caching, classification, parallel dispatch, aggregation, reporting
Test agent	agents/a11y-test-agent.md	Test runner: golden-file and regression test execution

Commands

/a11y-audit -- Review code for accessibility issues

/a11y-audit src/components/LoginForm.vue     # audit a specific file
/a11y-audit src/components/                   # audit all supported files in a directory
/a11y-audit --changed                         # audit git-changed files (staged + unstaged)
/a11y-audit --level AAA                       # override conformance level (default: AA)
/a11y-audit --category "Forms and validation" # restrict to one category
/a11y-audit --no-cache                        # skip cache, force re-audit of all files
/a11y-audit                                   # defaults to changed files, level AA

Audit mode is read-only. No files are modified.

--no-cache flag

By default, /a11y-audit skips unchanged files using incremental caching (see Performance). Use --no-cache to bypass the cache and re-audit all scoped files. Useful when you've updated the checklist or want a clean baseline.

Output (every audit produces all three):

1. Inline summary -- printed to the conversation with pass/fail/manual counts and top priorities
2. JSON report -- docs/a11y-reports/YYYY-MM-DD-audit.json (machine-readable, used by /a11y-fix)
3. Markdown report -- docs/a11y-reports/YYYY-MM-DD-audit.md (human-readable, grouped by severity)

/a11y-fix -- Fix automatable issues

/a11y-fix                                               # fix all AUTO/PARTIAL findings from latest audit
/a11y-fix forms-and-validation-every-form-control-001   # fix a specific check by ID
/a11y-fix --category "Forms and validation"             # fix all fixable findings in a category

Requires a prior /a11y-audit. After applying fixes, the agent re-runs the audit on modified files and reports what resolved and what remains.