Generates a numbered, step-by-step manual test script covering keyboard navigation and screen-reader (NVDA / VoiceOver) verification for a given component or page, with every test case mapped to its WCAG 2.2 Success Criterion.
When invoked
Inputs the agent requires:
- Component or page name - e.g. "ConfirmDialog", "CheckoutForm", "NavMenu".
- List of interactive widgets present - e.g. modal dialog, combobox, menu button, tabs, text inputs, toggle buttons.
- ARIA roles claimed (optional) - if available from source or design spec.
- Target browser / AT version - defaults to NVDA + Firefox (Windows) and VoiceOver + Safari (macOS).
Steps
- Inventory the widgets. Read the component source (or design spec) with
Read / Grep and record every focusable element, its role, its expected accessible name, and the states it exposes (aria-expanded, aria-pressed, aria-checked, and so on). - Assemble the sheet. Apply
widget-a11y-test-matrix to that inventory: it owns the per-archetype keystroke rows, the expected NVDA and VoiceOver announcements, the WCAG 2.2 success criterion per row, the placeholder substitution rules, and the output template. - Emit the script with the Result column left blank for the human tester, and the NVDA build, VoiceOver / macOS version, and paired browser recorded in the header.
Hand-off targets
- Code-level fixes after a failing test case -
accessibility-code-critic (qa-accessibility) reads the source, identifies the violating line, and proposes a fix. - Focus trap implementation -
wcag-focus-trap. - ARIA role and keyboard pattern reference -
aria-authoring-patterns. - Keyboard navigation rules -
wcag-keyboard-navigation. - Automated pre-check before the manual run -
axe-a11y, to clear structural defects before a tester sits down. - Screen-reader script authoring as a standalone skill -
screen-reader-test-author.
