built-in spoken read-back of Claude's replies (accessibility / hands-free)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I have dyslexia, and reading long terminal output is slow and tiring for me.
Claude Code's replies are often long and code-heavy, so I currently highlight
each response by hand and run macOS read-back at ~3x speed just to get through
it. This breaks my flow on every single turn.

/voice solves input (I can talk to Claude), but there's no way to listen to
its replies. So only half my workflow is hands-free. I need Claude Code to
read its responses back to me automatically, the way I can already speak to
it.

Proposed Solution

A built-in "speak replies" mode that:

1. Automatically reads each assistant reply aloud when the turn finishes.
2. Lets me pause and resume with a hotkey — resuming from the exact word it

stopped on, not restarting.

1. Has adjustable speech rate and voice (fast readers need ~3x).
2. Can skip or just announce code blocks instead of reading symbols aloud,

which is unusable at speed.

1. Has a quick mute / "stop speaking" key and a per-session toggle.
2. Works cross-platform (macOS say, plus a sensible engine on

Linux/Windows).

Ideally it pairs with the existing /voice dictation so input + output
together give a fully hands-free loop. Ship it off by default, opt-in via a
setting.

Alternative Solutions

• Manually highlighting each reply and using macOS system read-back at 3x —

works but tedious and breaks flow every turn.

• /voice — only does dictation (input), not read-back.
• I built a working prototype: a Stop hook extracts the final reply, strips

markdown, skips fenced code blocks, and pipes it to macOS say -v <voice> -r
<rate>. A key-bound script does true pause/resume via kill -STOP / kill
-CONT on the say process. It works well, but it requires scripting and
isn't something most users could set up.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

1. I start a Claude Code session and turn on "speak replies."
2. I use /voice to dictate: "Refactor the auth middleware to use JWT."
3. Claude finishes its reply, which automatically begins reading aloud at my

preferred ~3x rate. Code blocks are skipped, so I hear the explanation, not
raw syntax.

1. Partway through, I want to re-read a file — I tap a hotkey to pause the

voice mid-sentence.

1. I tap it again and the voice resumes from the exact word it paused on.
2. I never touched the mouse, never highlighted text, and stayed in flow the

whole time.

Additional Context

_No response_