Accessibility: Users with motor disabilities have no reliable input method

The Problem

Claude Code currently has no reliable input path for users with motor disabilities. macOS has world-class dictation built right into the OS — instead of reinventing the wheel, this should be leveraged. Even as a stopgap until the built-in voice service is more reliable. Because when the voice service doesn't work, we can't work.

The built-in voice mode is a good start, but it's not there yet — it cuts off mid-sentence, misses words, and doesn't auto-enable on startup despite being configured to (#34012).

A straightforward solution — allowing a custom STT endpoint — was raised in #7407, which is now locked with no resolution. That issue is from the 7,000s. We're now past 36,000.

Why I'm Listing All of These

I compiled this list because the scale of this problem is invisible. Individual issues get filed, auto-flagged as duplicates, auto-closed, and buried. Nobody ever sees the full picture. The duplicate-detection bot is doing its job — but the effect is that 150+ voices get silenced one at a time, and the underlying problem never gets the visibility it deserves. This is what falling through the cracks looks like.

The Scale of This Problem

This isn't one person complaining. I searched the issue tracker and found 200+ open and closed issues related to voice input being broken, unreliable, or inaccessible. Here's a sample, categorized:

Voice doesn't work on startup (voiceEnabled ignored):
#30908, #30968, #33805, #33841, #33860, #33939, #33954, #34311, #34401, #34464, #34618, #34676, #34980, #35566, #35644, #35648, #35980, #33378, #33354

Spacebar / push-to-talk broken:
#31341, #33737, #33795, #34019, #34550, #34611, #31110, #35156, #33605, #31783

Connection failures / stream errors / transcription failures:
#30279, #30293, #30904, #31082, #31101, #31223, #32960, #33421, #33720, #34094, #34174, #34414, #33929

Voice not recognized as a command / fails silently:
#33591, #33798, #33892, #33909, #34096, #34221, #34222

Audio capture / native module broken (Windows, Linux, WSL):
#31065, #33261, #33331, #33504, #33816, #33941, #34056, #34136, #30915, #33023, #33701, #32249

OS-level dictation / external voice tools broken by Claude Code:
#1852, #15307, #20570, #30929, #16233, #20476

Custom STT endpoint / bring-your-own-voice:
#2116, #7407 (locked), #25954, #29399, #36052, #9980, #34053

Language / transcription quality:
#30868, #30890, #33061, #33158, #33283, #33299, #33333, #33505, #33528, #33597, #33697, #33782, #34016, #34028, #31599, #34101

Voice UX (toggle, auto-submit, keybinding, device selection):
#30392, #30894, #32771, #33025, #33240, #33393, #33787, #33826, #33831, #33972, #34067, #34305, #34376, #35392, #36041, #34170

VS Code / remote / mobile voice support:
#26113, #29399, #32597, #34743, #34907, #35948

Accessibility-specific issues:
#1852, #11002, #15509, #23874, #25785, #35257, #35273, #15854, #31477, #25095

That's 150+ unique issues listed above, and I stopped counting. Each one represents a person who tried to use voice input, hit a wall, took the time to file a report, and in most cases got an auto-close.

What We're Asking For

One of:

• Let OS-level dictation work in the terminal input
• Make the built-in voice mode reliable
• Allow a custom STT endpoint so users can bring their own (a single env var)

Any one of these would unblock the accessibility community. Right now, none of them work.

Why This Matters

This isn't a feature request. It's an accessibility gap that excludes users with motor disabilities from using the product they're paying for. The tools to fix it are simple and well-understood. 150+ issues and a locked predecessor bug (#7407) say this community has been patient long enough.