Screen readers and dictation
You can run a workflow without looking at the dashboard: dictate a prompt into the prompt editor, run it, and have a screen reader tell you how the run is going and what the AI answered. This page covers what ctrlyoke does for screen readers and how to use your operating system's dictation with it.
What a screen reader announces
While a run is going, the dashboard tells your screen reader what changes, wherever your focus is. It doesn't depend on what the output pane is showing, so scrolling the output never triggers an announcement and nothing is missed because it was off screen.
Screen reader announcements in Settings → General → Accessibility (ui.accessibility.outputAnnouncements) chooses how much you hear:
| Value | What is announced |
|---|---|
| Step changes and finished responses | A step starting, completing, failing, being cancelled or ending because no turn started; the workflow completing, stopping or pausing; a question the AI asks you; the harness waiting for input; and each response once it finishes. This is the default. |
| Also read responses as they stream | All of the above, plus the response a paragraph at a time while it arrives, at most once every few seconds. |
| None | Nothing. |
A finished response is read as a short summary of what the step did, such as "Changed 3 files, ran 2 commands", followed by the AI's last message. Thinking, tool calls and diffs are never read out. Code blocks are read as "Code block, 12 lines, omitted", tables as "Table with 4 rows", links by their text, web addresses by their site name, and full file paths by the file name.
Announcements are polite: they wait until your screen reader finishes what it is saying. Each dashboard you have open announces its own run, so a phone and a desktop browser watching the same run both speak.
"Waiting for input" is announced only while Waiting for input notifications are on, which they are by default.
Read the last response
Press Alt+L (Option+L on macOS) anywhere in the dashboard to hear the most recent finished response again. It works in every announcement mode, including None. The output toolbar also has a Read last response button, hidden until it gets keyboard focus.
Moving through the output
The output pane is a region named "Workflow output". Inside it:
- Every step header is a level 3 heading, named for the step, its harness and how it went, for example "Step 3, Fix tests, Claude Code, finished". Use your screen reader's heading navigation (
Hin NVDA and JAWS, the rotor in VoiceOver) to jump from step to step. - Every message from the AI is an article named for its step, such as "Response, Step 3", so you can move between messages past the tool calls in between.
In a long run the output pane keeps only the steps near where you are in the page, so to reach an early step, jump with the minimap or scroll there first. When a step header is pinned to the top of the pane, the pinned copy is not a second heading.
The terminal
Turn on Terminal screen reader mode in Settings → General → Accessibility (ui.accessibility.terminalScreenReaderMode) so your screen reader can read the Terminal tab and follow new output. Browsers can't tell when a screen reader is running, so ctrlyoke can't turn this on for you.
Dictating prompts
The prompt editor accepts text from your operating system's dictation like any other text field, so you don't need anything from ctrlyoke to speak a prompt:
| Platform | How to dictate |
|---|---|
| Windows | Press Win+H for voice typing, or use Voice Access. |
| macOS | Press the Dictation shortcut set in System Settings → Keyboard. |
| iOS | Tap the microphone on the keyboard in the browser dashboard. |
| Android | Tap the microphone on Gboard in the browser dashboard. |
Then run the prompt with Ctrl+Enter, or queue it with Ctrl+Shift+Enter.