/voice mode: Poor Korean recognition and --send flag not auto-submitting

Hi, I've been using Claude Code Max and I'm truly grateful for this amazing tool. I'd like to report an issue I encountered with the /voice feature.

Description

Two issues with the /voice mode:

1. Poor Korean (non-English) speech recognition

When using /voice mode and speaking in Korean, the recognition accuracy is significantly lower compared to English. The transcription frequently contains errors or fails to capture spoken words correctly.

2. --send flag does not auto-submit

When starting voice mode with /voice --send, the transcribed text is not automatically submitted after recording ends (releasing Space). The user still has to manually press Enter to send the message, which defeats the purpose of the --send flag.

Steps to Reproduce

1. Run /voice --send in Claude Code CLI
2. Hold Space and speak in Korean
3. Release Space

Expected: Korean speech is accurately transcribed and automatically submitted
Actual: Transcription has poor accuracy for Korean, and the message is not auto-submitted

Environment

• OS: macOS (Darwin 24.6.0)
• Shell: zsh