Portable accessibility-audit skill bundle for Claude Code and Codex.
This repository packages a reusable agent skill for auditing web
projects with axe-core, optional Lighthouse checks, WCAG 2.1 AA
mapping, manual follow-up guidance, and configurable output modes.
a11y-audit/SKILL.md: core skill instructionsa11y-audit/references/: focused reference materiala11y-audit/scripts/: reusable helper scriptsa11y-audit/evals/evals.json: eval cases and recorded resultsa11y-audit/assets/: sample outputs and CI starter assetsa11y-audit/MANIFEST.yaml: bundle inventory and versioned hashesa11y-audit/CHANGELOG.md: bundle change historya11y-audit/HANDOFF.md: current validation status and next steps
| Mode | Output | Use case |
|---|---|---|
markdown |
Markdown report only | Human review, documentation |
markdown+json |
Markdown report + JSON data file | CI integration, dashboards, trend tracking |
markdown+issues |
Markdown report + issue tracker tickets | Active remediation workflow |
- Workspace-local
.a11y-audit/PROJECT_CONTEXT.mdconfiguration - Runtime URL reconciliation when the expected local port is wrong
- Explicit degraded-mode behavior when Lighthouse is unavailable
- Reusable helper scripts for first-run context creation and scanning
- Issue tracker support: GitHub, GitLab, Linear, Jira
Copy the a11y-audit/ folder into .claude/skills/ in the target
project.
Copy or symlink the a11y-audit/ folder into your Codex skills
directory.
Typical prompts:
Run an accessibility audit on this project.Audit this app for WCAG 2.1 AA issues and generate a markdown report.Use $a11y-audit to scan the homepage, about page, and contact page.
For project-specific setup, create .a11y-audit/PROJECT_CONTEXT.md in
the target workspace or use a11y-audit/scripts/bootstrap-context.js.
This repo includes:
- bundled eval definitions in
a11y-audit/evals/evals.json - sample output artifacts in
a11y-audit/assets/sample-output/ - a lightweight GitHub Actions validation workflow in
.github/workflows/validate-skill.yml
Validated in Claude Code (eval-1) and Codex against a real project for:
- core audit execution
- runtime URL reconciliation
- Lighthouse-unavailable degraded mode
- first-run context creation
- missing browser automation handling
Still pending as a direct end-to-end run:
- live authenticated issue tracker writes (
markdown+issuesmode)
MIT