Description

Version: nv2026.4.24

Environment: NVIDIA-hosted NemoClaw instance

Stopping and starting the hosted VM causes NemoClaw/OpenShell to create a completely new sandbox instead of reconnecting to the previous sandbox.
Because a new sandbox is created:
a new PVC/workspace volume is provisioned
previous workspace data is lost
I had files inside:
~/.openclaw/workspace/skills
These files existed before VM shutdown and disappeared after restart because the original sandbox no longer existed.

Expected behavior:Existing sandbox and workspace data should persist across VM stop/start cycles.

Actual behavior:After VM restart:

previous sandbox no longer exists

a completely new sandbox is created

workspace starts empty

previous skills/workspace data is unavailable

Important notes:

This appears different from the already-fixed registry persistence issue (#2276 / #2330).

In this case, the sandbox itself is recreated after VM restart.

Since a new sandbox is provisioned, a new PVC/workspace volume is also created.

Observed behavior suggests:

gateway/k3s/sandbox state may not persist across hosted VM stop/start cycles

onboarding/startup flow may recreate sandbox resources after restart

Reproduction Steps

Steps to reproduce:

• Create sandbox and add files inside ~/.openclaw/workspace/skills
• Stop the hosted VM instance completely
• Start the VM again, had started after 2 days.
• Observe that a new sandbox is create.
• Previous workspace files are gone in the sandbox, the data is present in vm but the data where the persistence should be
• /var/lib/docker/volumes/openshell-cluster-nemoclaw/_data/storage/pvc-..........._openshell_workspace-my-assistant

Like a new pvc is created and this is lost .
Expect is it should properly transfer .

Environment

Version: nv2026.4.24

Environment: NVIDIA-hosted NemoClaw instance

Debug Output

Logs

Checklist

• I confirmed this bug is reproducible
• I searched existing issues and this is not a duplicate