How to use the Accessibility Audit agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. When the skill is installed and your request matches its description, the agent follows the method in that file.

This skill covers accessibility auditing. The agent first confirms the WCAG edition and level and the target pages or flows. It then runs automated scans to collect baseline violations and coverage gaps, and checks keyboard use, screen-reader behaviour, focus order and colour contrast by hand. Each barrier is mapped to a WCAG criterion with severity and user impact, remediation steps are proposed, and fixes are re-tested.

• Records the actual test coverage, not just a pass or fail.
• Keeps untested journeys and incomplete checks visible in the report.
• Treats missing scope or UI access as a limitation, not a passing audit.

How do you install the Accessibility Audit skill?

Download the ZIP for this skill from the page you are reading. Then choose one of two routes depending on where you use the agent.

For Claude Code, unzip the download into your personal skills folder ~/.claude/skills/ so it is available in every project. To limit it to one project, unzip it into that project's .claude/skills/ folder instead.

In the Claude apps, you can upload the skill under Settings → Capabilities → Skills rather than unzipping it manually.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills

How do you trigger the skill with a natural request?

You do not need a special command. Describe the audit you want in plain language and name the standard, the place to test and the flow. The skill's description mentions keyboard, screen-reader, focus and contrast barriers, so requests that mention those topics are the most likely to match.

Good requests name a staging URL or UI, a WCAG edition and level, and the journey to check. If you leave out the standard or the environment, the skill is designed to ask for clarification rather than guess.

• Audit keyboard use in our checkout staging flow against WCAG 2.2 AA.
• Find contrast and focus barriers on the sign-up pages and rank them by user impact.
• Plan verified remediation for screen-reader problems in our account settings journey.

What does a good audit output look like?

The skill's own example shows the expected approach. The agent records the exact staging URL and the states tested, then inspects tab order, visible focus, field labels, validation errors and completion without a pointer. Each observed barrier is linked to the applicable criterion and evidence.

Screen-reader and unvisited-state checks that were not performed are marked as pending. After the authorised change, the agent repeats the failed interaction. It will not call a whole product compliant based on one successful scan.

What do you need before running an audit?

The skill needs access to the target pages or UI, such as a URL, a staging environment or design artifacts. You also need to agree on the WCAG edition and level and list the target flows. Audit tools are referenced in the playbook, which includes automated scanners, but the exact tools are not named in SKILL.md.

The playbook's integration sketches assume project-local, lockfile-pinned tools, real component fixtures and an authorised test target. The skill does not run scripts and does not use the network itself.

• Access to the target pages, UI or design artifacts
• Agreed WCAG edition, level and list of target flows
• Audit tools for automated scanning, set up in your own project

What safety limits should you know about?

The skill expects an authorised test environment. Do not point it at sites or accounts you are not permitted to test. The playbook also warns that browser navigation may fetch remote subresources and that you should not use private records or real credentials.

Be aware that a DOM heuristic or a zero-violation automated scan does not establish WCAG or legal conformance. Keep manual and assistive-technology checks explicit in your reports.

• Test only environments you are authorised to test.
• Do not use private records or real credentials.
• Expect remote subresources to load during browser navigation.

Tips for getting useful results

Give the agent representative states, such as an error message, an empty form or a logged-in view, because these often hide barriers. Ask it to list what it did not test, so you know the limits of the audit.

Use the findings as a prioritised backlog. Ask for remediation steps, then re-run the failed interactions after each fix to confirm the barrier is gone.

• Name the exact states to test, including error and validation states.
• Ask for untested journeys to be listed explicitly.
• Re-test each fixed barrier before marking it closed.