GitHub Actions
MarketplaceChoose a repository in your own GitHub account. We save the selected pages into its managed workflow and start the first Action immediately; runs and reports stay with you.
Open-source accessibility testing
Audit the pages that matter.
Remove barriers before release.
Start with one website and choose from the unique pages we find, or enter up to 20 public pages yourself. GitHub Actions runs and reports stay in your repository; local plugin reports stay on your device.
WCAG 2.2
Keyboard
Reflow
Evidence
Create an audit
GitHub ActionsSet up your accessibility audit
Setup progress
- 1Add pages
- 2Choose pages
- 3Repository
- 4Review
Checkpoint reached
Your workflow file is ready
The workflow is configured for your pages.
- 1Copy the workflow
Use the button below to copy the generated file.
- 2Add or replace one file
Commit the configured file at
.github/workflows/accessibility-audit.yml. - 3Run the audit
Open Actions → Accessibility audit → Run workflow. There are no URL fields to complete.
- 4Get the evidence
Download the combined report from the run summary.
Preview the workflow file
One run. One evidence pack.
- Accessible HTML
- Excel workbook
- JSON data
- CSV
- SARIF
- Screenshots
- ZIP archive
Use it where you work
One audit engine. Six ways to run it.
Choose a ready-to-use option or install the open-source developer package in your preferred coding assistant. Every card states what is available today.
Claude Desktop
DownloadUpload the packaged custom-plugin ZIP in Customize → Plugins. Keep the ZIP compressed.
Download the current ZIPClaude Code
DownloadDownload and extract the ready local-agent package, then load that directory from the project you want to audit.
Cursor
DownloadDownload and extract the ready local-agent package, add it to Cursor’s local plugin directory, then reload Cursor.
Codex CLI
DownloadDownload and extract the portable Agent Plugin package, register that directory as a local marketplace, then open the CLI plugin browser.
GitHub Copilot
DownloadsChoose the packaged Copilot CLI or VS Code payload. Public marketplace listing remains a separate submission step.
Installation note: Claude Desktop accepts its ZIP directly. The local-agent and Copilot archives must be extracted and installed using the linked client instructions. They require Node.js 22 or later and a Playwright-supported browser. The Codex IDE extension does not currently support plugins. Only audit websites you are authorised to test.
A shorter setup
From a page list to useful evidence.
Approve the exact pages, choose a repository, and let the builder save and start the managed workflow for you.
Choose your pages
Discover unique exact-origin pages from one URL, or enter up to 20 public pages yourself.
Choose where it runs
Connect GitHub and select a repository you control, or create a starter repository first.
Run and review
Connected setup starts the Action immediately; the manual fallback gives you the workflow to add and run yourself.
Honest by design
All 55 WCAG 2.2 Level A and AA criteria stay accountable.
Every report pairs selective automated evidence with one criterion-specific human procedure and an evidence prompt. Untested, partial, or inconclusive work stays visible and never becomes an automatic pass or certification claim.
Evidence-led browser checks
Chromium runs by default, with optional Firefox and WebKit coverage. Checks include semantics, names, contrast signals, structure, component states, links, 200% text resize, 320px reflow, and configurable keyboard, form, focus, and live-region journeys. Unresolved tool results become review work; bounded checks disclose partial coverage.
Review-ready, trackable outputs
An executive summary and prioritised actions lead the accessible HTML report. Standard A/AA results stay separate from supplemental and optional AAA advice. Consolidated remediation items retain their page, viewport, state, and selector occurrences across Excel, JSON, CSV, SARIF, screenshots, and ZIP exports.
Human decisions remain explicit
No automated tool can certify WCAG conformance. Screen-reader experience, meaning, usability, and exceptions still require qualified review.
Built in public
Inspect it. Improve it. Use it.
The Action and its test methodology are open source under the MIT licence.