GitHub/accessibility-scanner: finds accessibility gaps and attempts to fix them (github.com)

GitHub’s accessibility-scanner (github/accessibility-scanner) is an AI-powered GitHub Action that scans websites, files, repositories and dynamic content for accessibility barriers, files actionable GitHub issues, and can assign those issues to Copilot to propose fixes. It runs as a workflow (.github/workflows/) and requires a Personal Access Token (PAT) with write scopes (actions, contents, issues, pull-requests; metadata: read) — note GitHub Actions’ default GITHUB_TOKEN cannot be used. Inputs include urls, target repository, token, cache_key and optional auth (login_url/username/password or a JSON auth_context for complex flows). You can trigger scans manually or via CI, cache results across runs, and opt out of Copilot assignment. For AI/ML practitioners and dev teams this is significant because it automates end-to-end accessibility discovery and remediation at scale, integrating programmatic scanning with Copilot-generated code suggestions and human review. Technical implications: it leverages Playwright-like authenticated sessions via auth_context, supports repo- and file-scoped Copilot instructions, and creates traceable issues/PRs so fixes enter normal review pipelines. Caveats: the project is in beta, Copilot’s fixes aren’t guaranteed to be fully accessible, and all proposed changes should be reviewed before merging. The action is open-source (MIT) and intended to speed accessibility workflows while keeping human oversight.