[BUG] Cowork: "Download failed" is a false error — VM service fails to start, no download is ever attempted (Windows, 1.25927.0)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Cowork's Linux workspace stopped starting on Aug 5, 2026, one day after the app updated to 1.25927.0. Every bash call returns Download failed. Check your internet connection and try again.

No download is failing, because no download is ever attempted. I renamed %APPDATA%\Claude\vm_bundles out of the way and restarted the app. It never recreated the folder, never attempted a download, and continued logging a bundle path and version hash for a directory that no longer exists. No [downloadVM], [Bundle:status], or [cache] lines appear at all.

The real error surfaces separately in the same log: VM service not running. The service failed to start. Failure time is a fixed ~3.02s across 30+ attempts, which is a local timeout, not a network transfer.

Root cause: the VM service fails to start, and the failure is reported as a network download error. Two defects in the same subsystem prevent recovery: the auto-reinstall skip flag persists across a full OS reboot, and the repair path explicitly preserves the compressed cache it should be replacing, so it re-extracts the same archives rather than re-downloading.

What Should Happen?

The workspace should start so bash commands run. Failing that, the error should name the actual failure — VM service startup — rather than reporting a network download error for a step where no download was attempted.

Error Messages/Logs

--- Attempt that did read the bundle: all files validated, still "Download failed" ---
2026-08-06 09:56:31 [info] [cache] rootfs.vhdx checksum validated
2026-08-06 09:56:31 [info] [cache] vmlinuz checksum validated
2026-08-06 09:56:31 [info] [cache] initrd checksum validated
2026-08-06 09:56:31 [info] All files ready in C:\...\Claude\vm_bundles\claudevm.bundle
2026-08-06 09:56:31 [error] [VM:steps] download_and_sdk_prepare failed (19477ms): Download failed. Check your internet connection and try again.
2026-08-06 09:56:31 [info] [VM:start] Skipping auto-reinstall (already attempted once)

--- The real error ---
2026-08-06 10:51:45 [warn] [vm-client] getNetworkDrives unavailable; keeping prior snapshot: Error: VM service not running. The service failed to start.

--- After renaming vm_bundles away: still reports the missing bundle, no download attempted ---
2026-08-06 10:50:53 [info] [VM:start] Beginning startup, bundlePath=C:\...\Claude\vm_bundles\claudevm.bundle
2026-08-06 10:50:53 [info] [VM:start] Bundle version: 6d1538ba6fecc4e5c5583993c4b30bb1875f0f5a
2026-08-06 10:50:53 [info] [VM:steps] download_and_sdk_prepare started
2026-08-06 10:50:56 [error] [VM:steps] download_and_sdk_prepare failed (3027ms): Download failed. Check your internet connection and try again.
2026-08-06 10:50:56 [info] [VM:start] Skipping auto-reinstall (already attempted once)

--- Repair path preserves the cache it should replace ---
2026-08-06 09:56:08 [info] [deleteVMBundle] Reinstall files deleted (sessiondata.img and compressed cache preserved)

--- Fixed ~3.02s timeout, sample across attempts ---
3017ms  3022ms  3023ms  3025ms  3027ms  3030ms  3034ms  3039ms  3262ms

--- Stack ---
at Tt (C:\Program Files\WindowsApps\Claude_1.25927.0.0_x64__pzs8sxrjxfjjc\app\resources\app.asar\.vite\build\index.chunk-mUkWEEwF.js:10:1546)
at async Object.startVM (...\index.chunk-4MCuNsI1.js:78:42752)
at async ...\index.chunk-CeSX0peY.js:1:609340
at async Session.<anonymous> (node:electron/js2c/browser_init:2:119861)

Steps to Reproduce

Windows, Claude desktop updated to 1.25927.0.
Open any Cowork session, ask Claude to run any bash command.
Returns Download failed. Check your internet connection and try again. immediately.
Check %APPDATA%\Claude\logs\cowork_vm_node.log — the bundle files pass checksum validation before the failure, and the real error is VM service not running.
Rename %APPDATA%\Claude\vm_bundles away and restart the app. It does not recreate the folder or attempt a download; it logs the same missing bundle path and version hash and fails at the same ~3.02s.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

The build prior to 1.25927.0; worked through Aug 4, 2026

Claude Code Version

Claude desktop app (Cowork) 1.25927.0.0 — not the CLI

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Related but distinct: #55649, #56145, #57968, #58368 (Windows, sandbox never worked) and #68030 (macOS). This report is a regression on a previously working install at 1.25927.0, with logs showing the "Download failed" string is a false error for a VM-service startup failure — a diagnosis not present in those reports.

Ruled out, all verified: tethered to a phone hotspot (identical failure), 500GB free disk, no VPN, all antivirus fully disabled, no Windows Updates installed in the prior six days (Get-HotFix), VirtualMachinePlatform Enabled, HypervisorPlatform was Disabled and enabling it plus a full reboot changed nothing, bcdedit hypervisorlaunchtype Auto, bundle folder renamed away entirely.

vmms does not exist on this machine (no full Hyper-V role); vmcompute and hvhost are both Running.

Install: MSIX, SignatureKind: Developer, C:\Program Files\WindowsApps\Claude_1.25927.0.0_x64__pzs8sxrjxfjjc. Bundle version 6d1538ba6fecc4e5c5583993c4b30bb1875f0f5a.

Only the Cowork bash sandbox is affected. Chat, file tools, web fetch, MCP connectors and Claude in Chrome all work normally. This is the Claude desktop app in Cowork mode, version 1.25927.0.0, confirmed latest at time of filing. Not the Claude Code CLI.