What breaks most often in Windows voice-to-Codex setups?

WiaiKit launches tomorrow as a small setup and recovery kit for a Windows workflow where OpenWhispr turns speech into text and the result is used with Codex.

I kept the first version focused on the setup friction I personally hit: npm not being available in PowerShell, Electron install issues, native module rebuilds, Visual Studio Build Tools confusion, and Push-to-Talk fallback mode.

If you have tried a similar voice-driven coding workflow on Windows, what broke first for you? Node/npm, audio capture, key listeners, model setup, or something else?