[Bug] Voice mode recording completes but transcription returns no output

Bug Description
Voice mode: recording indicator works but transcription produces no output

Environment:

• macOS (Darwin 25.3.0)
• iTerm2
• Claude Code (latest version)
• Authenticated via Claude.ai account
• SoX installed via Homebrew

Steps to reproduce:

1. Enable voice mode with /voice
2. Hold Space — recording indicator appears and animates left to right
3. Speak clearly for 2-3 seconds
4. Release Space

Expected: Spoken text is transcribed into the input field.

Actual: Recording indicator disappears and nothing happens — no transcription, no error message.

Additional info:

• SoX works correctly outside Claude Code (rec test.wav / play test.wav captures and plays back audio fine)
• Microphone permissions are granted for iTerm2 (verified in System Settings > Privacy & Security > Microphone)
• Other audio recording apps work correctly with the same mic

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.83
• Feedback ID: 89aaa171-7d1b-430f-a93e-03398d89fc88

Errors

[{"error":"ZodError\n    at X89 (/$bunfs/root/src/entrypoints/cli.js:4164:253)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-25T16:02:01.678Z"},{"error":"ZodError\n    at X89 (/$bunfs/root/src/entrypoints/cli.js:4164:253)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-25T16:02:02.165Z"},{"error":"Error: console.error: [bun] Warning: ws.WebSocket 'unexpected-response' event is not implemented in bun\n\n    at <anonymous> (/$bunfs/root/src/entrypoints/cli.js:730:9126)\n    at <anonymous> (/$bunfs/root/src/entrypoints/cli.js:676:23263)\n    at value (native:171:22)\n    at warn (native:241:9)\n    at emitWarning (ws:43:42)\n    at #onOrOnce (ws:170:18)\n    at JcT (/$bunfs/root/src/entrypoints/cli.js:6768:7410)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-25T16:02:05.435Z"}]