Cowork: orphaned VM process blocks all resume attempts on Windows
GitHub Issue: Cowork orphaned VM process blocks all resume attempts on Windows
Repo: https://github.com/anthropics/claude-code/issues/new
---
Title: Cowork: orphaned VM process blocks all resume attempts on Windows
Body:
Bug Description
Cowork sessions on Windows (Claude Desktop) become permanently stuck when the VM connection drops. The orphaned process cannot be killed or replaced, blocking all future resume attempts until the user manually kills processes from an external terminal.
This is a recurring issue — it happens frequently enough to disrupt workflow. When Claude Desktop hangs during cowork, there is no UI mechanism to force-stop the session, and the orphaned process persists across VM restarts.
Environment
- OS: Windows 10/11 (MINGW64_NT-10.0-26200)
- Claude Desktop: v1.1.4498.0 (Windows Store)
- Claude Code SDK: v2.1.51 (inside VM)
Steps to Reproduce
- Start a cowork session in Claude Desktop
- Session hangs mid-conversation (connection drops)
- Close and reopen Claude Desktop
- Try to resume or start new cowork — blocked with:
RPC error -1: process with name "sleepy-blissful-hamilton" already running
Root Cause (from logs)
The VM connection drops abruptly:
[error] failed to write stdin: Error: failed to write data: An established connection was aborted by the software in your host machine.
The app tries to kill the orphaned process but fails because the daemon is disconnected:
[error] kill failed with error: Error: sdk-daemon not connected
Subsequent resume attempts fail because the VM still thinks the old process is running:
[error] Spawn failed: RPC error -1: process with name "sleepy-blissful-hamilton" already running (id: b6bc4bdf-e270-4bd6-9fa4-346a6ceab016)
The new process can't be killed either (it was never actually created):
[error] kill failed with error: RPC error -1: process 7a3cbe9f-271e-4ddd-a3f9-f5fcf29f5809 not found
VM shutdown also times out:
[error] Failed to run onQuitCleanup(cowork-vm-shutdown): Error: Request timed out
Full Error Timeline (cowork_vm_node.log)
2026-03-03 23:53:27 [error] failed to write stdin: Error: failed to write data: An established connection was aborted by the software in your host machine.
(repeated 5x)
2026-03-03 23:53:28 [error] kill failed with error: Error: sdk-daemon not connected
(repeated 2x)
2026-03-03 23:54:41 [warn] Ping failed: Keep-alive ping timed out
2026-03-03 23:54:56 [error] Spawn failed: RPC error -1: process with name "sleepy-blissful-hamilton" already running (id: b6bc4bdf-e270-4bd6-9fa4-346a6ceab016)
2026-03-03 23:54:57 [error] kill failed with error: RPC error -1: process d7216f61-d5b9-43dc-87c1-bbba3b80d730 not found
(repeated 2x)
2026-03-03 23:56:19 [error] Spawn failed: RPC error -1: process with name "sleepy-blissful-hamilton" already running (id: b6bc4bdf-e270-4bd6-9fa4-346a6ceab016)
(second resume attempt also fails)
2026-03-03 23:56:19 [error] kill failed with error: RPC error -1: process 7a3cbe9f-271e-4ddd-a3f9-f5fcf29f5809 not found
2026-03-04 00:00:57 [error] Failed to run onQuitCleanup(cowork-vm-shutdown): Error: Request timed out
main.log corroboration
2026-03-03 23:53:27 [error] Process spawn error: failed to write data: An established connection was aborted by the software in your host machine.
2026-03-03 23:54:26 [error] Sentry caught: { type: 'Error' } (12x)
2026-03-03 23:54:56 [error] Process spawn error: RPC error -1: process with name "sleepy-blissful-hamilton" already running
2026-03-03 23:56:19 [error] Process spawn error: RPC error -1: process with name "sleepy-blissful-hamilton" already running
2026-03-04 00:00:57 [error] Failed to run onQuitCleanup(cowork-vm-shutdown): Error: Request timed out
Expected Behavior
- When the VM connection drops, the orphaned process should be automatically cleaned up
- Resume attempts should force-kill stale processes before spawning new ones
- Claude Desktop should provide a UI button to force-stop a hung cowork session
- VM shutdown should not time out — it should force-terminate processes
Current Workaround
Must open a separate terminal and manually kill the process:
pkill -f "claude cowork"
This is not discoverable for non-technical users and loses all in-progress work.
Impact
- Frequency: Happens repeatedly on Windows
- Severity: Completely blocks cowork usage until manual intervention
- Data loss: In-progress conversation and work is lost with no recovery path
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗