What is the screen reader testing skill?
Screen reader testing is a skill in wshobson/agents for validating web apps with VoiceOver, NVDA, JAWS, TalkBack and Narrator. It sets a minimum test matrix, explains browse and focus modes, and gives test patterns for modals, live regions and tab interfaces, with reference files per screen reader.
Screen reader testing is a skill in the accessibility-compliance plugin of wshobson/agents. It is a practical guide to checking a web application with real assistive technology rather than relying on automated scans alone.
The main SKILL.md holds the testing strategy and three common scenarios. Setup, commands and checklists for each screen reader sit in separate reference files the agent reads before testing.
What changes in your agent's output
- Test plans name specific screen reader and browser pairs, starting with NVDA and Firefox, VoiceOver and Safari, and VoiceOver on iOS Safari.
- Findings distinguish browse mode, focus or forms mode and application mode, since each behaves differently.
- Modals, live regions and tab widgets are checked against known-good markup and keyboard behavior.
- Dynamic content announcements and error states are tested, not only the happy path.
- The agent logs role, name and state of an element to approximate what assistive technology sees.
Who it is for
Best for
- Developers who must confirm ARIA work actually announces correctly.
- QA planning a manual accessibility pass across several platforms.
- Teams debugging a report that a widget is unusable with a screen reader.
Not for
- Running a screen reader for you. An AI agent cannot listen to speech output, so a person must test.
- Automated contrast or markup scanning, which belongs to an audit skill.
- Native mobile app testing beyond the browser coverage it lists.
Use it with UiChemy
After UiChemy produces a WordPress page, run the same screen reader checks on the live page. UiChemy Figma to WordPress converts designs to editable pages, but accessibility still needs a human pass. Click Get Started on uichemy.com.
How to install Screen Reader Testing Skill: VoiceOver, NVDA and JAWS Guide
gh skill install wshobson/agents screen-reader-testingThe repository describes itself as one Markdown source adapted to each tool by generators, so folder paths differ per tool.Install only this skill
Run
gh skill install wshobson/agents screen-reader-testingand pick your coding tool when prompted. The README says it reads skills from GitHub without cloning and installs them without the plugin's agents, commands or hooks.Check the requirement
The README says gh skill requires GitHub CLI 2.90 or later. Update the CLI first if the command is not found.
Or install the plugin
In Claude Code, run
/plugin marketplace add wshobson/agentsand then install the plugin that holds this skill, accessibility-compliance, from the claude-code-workflows marketplace.Other tools
The README lists Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot and Pi, each with its own route. Use the table in the README for yours.
Use the skill
Ask your agent to build a screen reader test plan for a page or widget and to run through the modal, live region and tab scenarios.
Check that it loaded
Ask the agent to list its skills and look for screen-reader-testing, or request a test plan and check that it names screen reader and browser pairs.
Key features
- Screen reader tableLists VoiceOver, NVDA, JAWS, TalkBack and Narrator with platform and browser pairings, and approximate usage shares of about 15, 31, 40, 10 and 4 percent.
- Coverage tiersMinimum coverage is NVDA with Firefox, VoiceOver with Safari on macOS and VoiceOver with Safari on iOS. Wider coverage adds JAWS, TalkBack and Narrator.
- Reading modesBrowse or virtual mode reads content, focus or forms mode interacts with controls, and application mode serves custom widgets. The skill says to test in more than one.
- Per-reader referencesSeparate files cover VoiceOver setup and commands, iOS touch gestures, NVDA modes and test scripts, and JAWS with TalkBack platform commands.
- Modal dialog scenarioShows dialog markup with a labelled title and description, plus focus code that stores the opener, moves focus in, traps Tab and returns focus on Escape.
- Live region scenarioCovers polite status regions, assertive alerts, progress bars with value attributes and log regions that announce additions only.
- Tab interface scenarioProvides tablist markup with selected state and controls, and arrow, Home and End key handling that moves focus and activates the tab.
- Debugging helperA short script logs role, accessible name, expanded, selected, checked and disabled state and visibility for an element.
- Do and do not listTest with real screen readers, use semantic HTML first, check keyboard first, cover error states, and do not assume one screen reader is enough.
Frequently asked questions
What is screen reader testing?
It is a skill in the accessibility-compliance plugin of wshobson/agents. It guides testing of web apps with VoiceOver, NVDA, JAWS, TalkBack and Narrator, sets a minimum coverage matrix, explains reading modes and supplies tested markup patterns for modals, live regions and tabs.
How do I install the screen reader testing skill?
Run gh skill install wshobson/agents screen-reader-testing and pick your coding tool. The README says gh skill needs GitHub CLI 2.90 or later and installs the skill without cloning or the plugin's agents. The accessibility-compliance plugin is the other route.
Which screen readers should I test with first?
The skill's minimum coverage is NVDA with Firefox on Windows, VoiceOver with Safari on macOS and VoiceOver with Safari on iOS. Wider coverage adds JAWS with Chrome, TalkBack with Chrome on Android and Narrator with Edge. The usage shares it lists are approximate.
Can an AI agent test with a screen reader for me?
The skill is a guide for how to test. An agent can write test plans, markup and scripts, and can log accessible names, but listening to real speech output on real devices needs a person. The skill itself says to test with actual screen readers, not simulators.
Does it work with Claude Code, Cursor and Codex?
The repository README says its collection works across Claude Code, Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot and Pi, adapted per tool by generators. Each tool has its own install route, so follow the README table for yours.
How is it different from an accessibility audit skill?
Audit skills such as Accessibility Review score code against WCAG criteria. This skill focuses on how to run manual checks with assistive technology. Accessibility Testing is a quick WCAG reference, and the two complement each other in a full test plan.
What are screen reader testing alternatives?
For a tool-driven audit, A11y Ally combines axe-core, pa11y and Lighthouse. For announcements specifically, ARIA Live Regions goes deeper than the live region examples here. Neither replaces testing with a real screen reader.


