a11y-tester-basic

A lightweight, embeddable accessibility testing skill. Runs axe-core against a URL via Playwright MCP and enriches each violated component with MagentaA11y acceptance criteria. Designed to drop into any agent stack — only two MCP servers required, no other ARC skills.

For the richer JIRA-ready, persona-mapped, WCAG-enriched workflow, use a11y-tester with the full ARC skill set.

What it does

• Navigates to a URL with Playwright (Chromium)
• Injects axe-core 4.8.4 from CDN and runs it
• Filters to critical + serious violations only (with full unfiltered count for context)
• Looks up each violated element type in MagentaA11y for component-level acceptance criteria
• Returns a single structured Markdown report — no preamble, no commentary

See SKILL.md for the exact report shape, severity rules, and constraints.

Triggers (natural language)

The skill loads automatically when the host agent sees prompts like:

• "test this page for accessibility"
• "a11y test https://example.com"
• "run a11y test"
• "accessibility test"
• "check for WCAG violations"
• "audit this URL"
• "is this page accessible"
• "find accessibility issues"

Requirements

Two MCP servers must be available in your IDE:

# Playwright MCP — browser automation + axe-core injection
claude mcp add playwright npx @playwright/mcp@latest

# MagentaA11y MCP — component acceptance criteria
claude mcp add magentaa11y --scope user -- npx -y magentaa11y-mcp

Cursor, VS Code, Codex, and Windsurf equivalents are configured automatically by arc-skills-deploy. See SKILL.md for manual JSON config.

Output

A single Markdown block: summary table, violations table, MagentaA11y acceptance criteria per element type, and the untruncated raw violations JSON. The output is intentionally terse and parseable — built for orchestrator consumption.

Using inside an agent orchestration layer

This skill is invokable two ways:

1. Natural-language trigger — the host agent recognises one of the trigger phrases above and loads the skill.
2. Direct delegation from an orchestrator or agent team — another skill or agent invokes a11y-tester-basic by name, passes a URL, and consumes the structured report.

Wiring into an agent team

When composing an agent team or higher-level workflow, reference this skill by its skill name and the URL to test:

Delegate to skill: a11y-tester-basic
Input: url = "https://example.com/checkout"
Expect: structured Markdown report (see SKILL.md "Step 4: Output")
Parse: `## Violations` table for triage, `## Raw Violations (JSON)` for programmatic handling

Because the skill emits no preamble or prose commentary, the orchestrator can splice the report directly into a larger document, feed ## Raw Violations (JSON) into a remediation step, or fan out per-rule fixes.

Example orchestration step

An orchestrator's plan might look like this:

1. a11y-tester-basic → URL X → capture critical/serious violations
2. (your remediation skill) → consume Raw Violations JSON → propose fixes
3. a11y-tester-basic → re-test URL X → confirm violation count dropped

For a real, full-stack example see a11y-audit-fix-agent-orchestrator, which sequences analysis → remediation → validation across multiple skills.

Why "basic"?

a11y-tester-basic deliberately does not delegate to other skills, format JIRA tickets, or map issues to user personas. It is a single, self-contained tester — the smallest reliable unit you can compose. Reach for a11y-tester when you need the richer workflow.