Voice server segfaults (signal 11) on WSL2 after updating to 1.0.100; /voice stuck on 'Fetching voice models' #5197

Summary

After updating to Copilot CLI 1.0.100 (which also pulled a new voice runtime), voice input no longer works on WSL2. The voice server process segfaults (signal 11) a few seconds after starting, so /voice stays on "Fetching voice models" and no model can be downloaded.

Environment

• Copilot CLI: 1.0.100 (linux-x64), latest at time of report
• OS: Ubuntu (glibc 2.39) on WSL2, kernel 6.18.33.2-microsoft-standard-WSL2
• CPU: x64 with AVX2, 8 cores, 16 GB RAM
• Audio: WSLg PulseAudio available (/mnt/wslg/PulseServer)
• Voice worked on earlier versions (e.g. 1.0.98) with the older runtime (Microsoft.AI.Foundry.Local.Core.so, ~645 MB)

Steps to reproduce

1. Update to 1.0.100; the update installs a new voice runtime (libfoundry_local.so, ~85 MB).
2. Start copilot and run /voice, choose Continue.
3. Status shows "Fetching voice models" indefinitely.

Actual behaviour

Process log:

[INFO]  voice server: boot pid=1708 endpoint=/run/user/1000/copilot-voice/<id>.sock
[INFO]  voice server: connected to spawned voice server pid=1708
[WARNING] [VoiceEngine] connection error: read ECONNRESET
[ERROR] [VoiceEngine] fatal: Engine connection closed.
[WARNING] [VoiceEngine] start failed: Engine connection closed.

WSL kernel log (repeated on every attempt, ~6-15s after the voice server starts):

copilot: MainThread: potentially unexpected fatal signal 11.
WSL (CaptureCrash): Capturing crash for pid: 1708, executable: .../@github/copilot-linux-x64/copilot, signal: 11

Already tried

• Removed stale /run/user/$UID/copilot-voice/* socket/pid files
• Deleted ~/.cache/copilot/foundry and ~/.github-copilot-cli/cache/models, then re-ran /voice to reinstall the runtime. The crash still occurs with the freshly downloaded runtime.
• ldd on libfoundry_local.so reports no missing libraries.

Expected behaviour

The voice server stays running, the model list loads, and dictation works.

Possibly related: #4024