Enable voice transcription in Linux/WSL builds

What version of Codex CLI is running?

codex-cli 0.105.0

What subscription do you have?

ChatGPT Business

Which model were you using?

gpt-5.3-codex

What platform is your computer?

Linux 6.6.87.2-microsoft-standard-WSL2 x86_64 x86_64

What terminal emulator and version are you using (if applicable)?

WSL

What issue are you seeing?

Voice transcription does not work when Codex runs on Windows 11 inside WSL (Linux target), even when the feature is explicitly enabled.

With this in config:

[features]
voice_transcription = true

Press-and-hold space never enters recording/transcription flow in WSL builds. The code path is compiled out on Linux (target_os = "linux"), so voice capture/transcription is unavailable in practice on WSL.

What steps can reproduce the bug?

1. Run Codex inside WSL on Windows 11.
2. Enable:

[features]
voice_transcription = true

1. Start Codex TUI and focus composer input.
2. Press and hold Space to trigger hold-to-talk transcription.

Repro context:

• Environment: Win11 + WSL (Codex built as Linux target)
• Relevant upstream change: PR #3381 (voice transcription)

What is the expected behavior?

When features.voice_transcription = true, hold-to-talk transcription should work in WSL/Linux builds the same way as other supported desktop targets:

• press/hold Space starts recording
• release Space stops recording
• transcription placeholder is replaced with transcribed text (or cleanly removed on failure)

Additional information

Root cause is compile-time OS gating, not the feature flag:

• Voice deps were restricted to cfg(not(target_os = "linux"))
• Voice module/event handling/composer paths were also gated by not(target_os = "linux")

Proposed fix branch/commit:

• Branch: DirkSchlossmacher/codex:fix/wsl-voice-transcription
• Commit: 4bbfa27cb2fcf0aa2564abf641ac05f018a60642

The fix switches those guards to feature = "voice-input" so Linux/WSL builds can use voice transcription when the feature is enabled.