Cowork VM fails on Windows 11 Home: wrong DNS config, missing sessiondata.vhdx, no NAT

Environment

• OS: Windows 11 Home (Build 26200)
• Claude Desktop: v1.1.2685.0 (Microsoft Store)
• Architecture: x64

Summary

Cowork workspace fails to reach the Claude API on Windows 11 Home. The VM boots successfully (startup completes in ~11s, SDK installs fine, guest connection works), but the API is always reported as unreachable. Three bugs were identified:

Bug 1: Wrong DNS configuration in HNS endpoint

cowork-svc.exe configures the VM's DNS server as 192.168.0.1, but the actual WiFi DNS is 192.168.1.1. This was confirmed via Get-HnsEndpoint:

Name           : cowork-vm-eth0
IPAddress      : 172.16.0.21
GatewayAddress : 172.16.0.1
DNSServerList  : 192.168.0.1   <-- WRONG, should be 192.168.1.1

Even after deleting and letting cowork-svc recreate the HNS network, the DNS is always 192.168.0.1. The VM cannot resolve api.anthropic.com.

Bug 2: sessiondata.vhdx not created during workspace reinstall

When clicking "Reinstall the workspace", the bundle downloads rootfs.vhdx, vmlinuz, initrd, and copies smol-bin.vhdx from app resources, but never creates sessiondata.vhdx. The VM then fails with:

HCS operation failed: HRESULT 0x80070002
'cowork-vm': Attachment 'sessiondata.vhdx (Lun 2)' could not be found

We had to manually create a VHDX file to get past this error.

Bug 3: MSFT_NetNat WMI class missing on Windows 11 Home

The New-NetNat / Get-NetNat PowerShell cmdlets fail with "Invalid class". The NetNatCim.dll and NetNat.mof files are missing from C:\Windows\System32\wbem\. The WinNat service runs but has no NAT rules configured.

The HNS network is created with Type: NAT, but without the WMI provider, NAT rules may not be properly applied. This means the VM (on 172.16.0.0/24) has no source NAT to reach the internet through WiFi.

Bug 4: Network diagnostics use macOS commands on Windows

The log shows the app trying to run macOS commands for network diagnostics on Windows:

Failed to spawn /sbin/ifconfig: spawn /sbin/ifconfig ENOENT
Failed to spawn /usr/sbin/scutil: spawn /usr/sbin/scutil ENOENT

Reproduction steps

1. Install Claude Desktop from Microsoft Store on Windows 11 Home
2. Open Cowork
3. VM boots and connects but API is always "UNREACHABLE"

Log excerpt (key lines)

[VM:start] Startup complete, total time: 10669ms
[Keepalive] Starting (interval=2000ms)
[VM] API reachability: PROBABLY_UNREACHABLE
[VM] API reachability: UNREACHABLE
[VM:network] API is unreachable

What we tried (none fixed the API connectivity)

• Rebooting multiple times
• Reinstalling workspace (triggered sessiondata.vhdx bug)
• Enabling IP forwarding (IPEnableRouter = 1)
• Enabling per-interface forwarding on vEthernet and WiFi
• Installing WSL2 (hoping it would register NetNat WMI class — it didn't)
• Attempting ICS (Internet Connection Sharing) — changed adapter IP, broke VM connectivity
• Deleting/recreating HNS network — DNS always comes back as 192.168.0.1
• Running DISM /RestoreHealth

Full log

<details>
<summary>cowork_vm_node.log</summary>

See attached or available at: %APPDATA%\Claude\logs\cowork_vm_node.log

</details>