Simulate a screen reader with AI
Walk through HTML or JSX as a sequence of spoken labels. Compare reading, tab, heading and form navigation before trying the same task with assistive software.
To simulate a screen reader with AI, give the agent the markup for a small task, such as choosing a delivery method. Three buttons can each say “Choose” visually and still make sense beside three cards. Heard in isolation, the same word may tell a shopper very little. The screen-reader-lab skill produces a text walkthrough of the names and roles the markup is likely to expose.
Try the task twice. Read the page in document order, then move between focusable controls only. A heading that explains the next choice on the first pass might never feature on the second. Use that observation to plan a short test with a real screen reader.
What it does
- Follow the likely reading order for a checkout section.
- Inspect the accessible name of a field with a visible label.
- Compare the announced states of an expanded and collapsed control.
- Review heading navigation without reading every paragraph.
- Trace a keyboard route between controls and identify ambiguous stops.
Install and use
- Install the skill in your agent and provide HTML or JSX. The simulation itself needs no speech engine.
- Choose a mode and ask for the name, role and state at each step.
- Compare the result with the task, then test the rendered interface using the browser and assistive software your audience uses.
FAQ
Source
Made by Community Access (Taylor Arndt). Original skill: screen-reader-lab, published on GitHub under the MIT licence.
View on GitHub