Voice push-to-talk regression: Space key scrolls indefinitely and recording captures no audio (v2.1.83)

Description

Voice push-to-talk worked correctly in v0.2.81 but is broken in v2.1.83 on Windows 11.

Steps to Reproduce

1. Enable voice mode with /voice
2. Hold Space to start recording
3. Observe behavior

Expected Behavior (v0.2.81)

• Hold Space → 1-2 space characters typed during warmup → recording activates → speech is captured and transcribed

Actual Behavior (v2.1.83)

• Hold Space → spaces scroll continuously across the input bar for much longer before recording eventually activates
• When recording does activate, no audio is captured — transcription returns empty/nothing
• Microphone works fine in other applications
• /doctor reports no issues
• Multiple microphones detected and accessible (tested with built-in mic and Poly BT600 headset)

Environment

• OS: Windows 11 Pro 10.0.26200
• Shell: Git Bash
• Working version: 0.2.81
• Broken version: 2.1.83
• Auth: Claude.ai account (not API key)
• Audio devices: Multiple mics detected (High Definition Audio Device, Poly BT600)
• Node: installed via npm globally

Additional Context

• Custom keybindings (alt+k, meta+k, shift) were also tested as alternatives — none of them triggered recording at all. Only Space eventually enters recording mode, but captures nothing.
• The audio-capture.node native module in vendor/audio-capture/x64-win32/ may be relevant — it holds a file lock while Claude Code is running.

🤖 Generated with Claude Code