Vouch Request: Fix SSH handshake secret persistence across restarts

[mak1jk]

I'd like to fix the SSH handshake secret persistence issue that causes sandbox connectivity to break after container restarts.

Root Cause

deploy/docker/cluster-entrypoint.sh generates a new random SSH_HANDSHAKE_SECRET on every start, but the Sandbox CRD retains the old secret in its pod spec. After restart, the gateway's secret doesn't match the sandbox's, causing handshake verification failed errors.

Fix

Persist the generated secret to /var/lib/rancher/k3s/server/ssh-secret.dat and reuse it on subsequent starts. The env var override is preserved for manual overrides. This ensures the gateway always injects the same secret, matching what the Sandbox CRD already has.

Verification

I diagnosed this locally on Ubuntu/WSL2 with NemoClaw v0.1.0 + OpenShell v0.0.11:

1. Traced the NSSH1 handshake flow from openshell ssh-proxy through the gateway to the sandbox
2. Compared OPENSHELL_SSH_HANDSHAKE_SECRET env vars between openshell-0 and the sandbox pod after restart — confirmed they diverge
3. Applied the fix to the container's entrypoint, verified secrets match across multiple restart cycles
4. Confirmed openshell sandbox connect works after docker restart

I can explain every line of the change and how it interacts with the HelmChart template injection, K3s manifest lifecycle, and Sandbox CRD pod spec.

Related

• Sandbox Not Available Anymore After Machine Restart NemoClaw#486 "Sandbox Not Available Anymore After Machine Restart"
• Branch: fix/persist-ssh-handshake-secret

[drew]

Thanks for opening this discussion. We have a ticket for this here, #487, and am I actively working on the item here, #488. We should have this fixed early next week.