[All Platforms][Nemoclaw] nemoclaw destroy kills dashboard port forward (18789) even when another sandbox is still running

[zNeill]

Description

[Description]
When two sandboxes exist and the 18789 port forward is pointing to the second (most recently onboarded) sandbox, destroying the first sandbox incorrectly stops the port forward. The 18789 forward belongs to the second sandbox, not the one being destroyed — the destroy should not touch it. After destroy, the remaining sandbox loses its dashboard access.

[Environment]
Device: ubuntu 22.04
Node.js: v22.22.2
npm: 10.9.7
Docker: Docker Engine 28.2.2
OpenShell CLI: 0.0.25
NemoClaw: v0.0.10-9-gc4767b63 (source install, PR #1522)
OpenClaw: 2026.4.2 (d74a122)

[Steps to Reproduce]

1. nemoclaw onboard → create sandbox "sandbox-1"
2. nemoclaw onboard → create sandbox "sandbox-2"
3. nemoclaw list (confirm both sandboxes present)
4. nemoclaw sandbox-1 destroy → confirm with 'y'
5. lsof -i :18789 (port is now unoccupied)

[Expected Result]
At step 5, port 18789 should still be occupied. The remaining sandbox (sandbox-2) is still running and needs dashboard access.

[Actual Result]
$ lsof -i :8080
COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME
docker-pr 491794 root 7u IPv4 1020988 0t0 TCP *:http-alt (LISTEN)

$ lsof -i :18789
(empty — port not occupied)

Port 18789 (dashboard forward) is gone even though sandbox "sandbox-2" is still running.

Bug Details

Field	Value
Priority	Unprioritized
Action	Dev - Open - To fix
Disposition	Open issue
Module	Machine Learning - NemoClaw
Keyword	NemoClaw, NEMOCLAW_GH_SYNC_APPROVAL, NemoClaw-SWQA-RelBlckr-Recommended

---

[NVB# 6063387]

[wscurran]

✨
Possibly related open PRs:

• #1522 feat(compat): bump openclaw from 2026.3.11 to 2026.4.2

---

Possibly related open PRs:

• #1522 feat(compat): bump openclaw from 2026.3.11 to 2026.4.2

[brandonpelfrey]

I retested this on current main and was unable to reproduce it.

I validated the scenario with two sandboxes, explicitly forwarded port 18789 to the second sandbox, then destroyed the first sandbox. The forward remained attached to the second sandbox and lsof -i :18789 still showed the
listener after the destroy.

This appears to already be fixed on current main. The likely fixing change was the destroy-path hardening in PR #1624 (fix(cli): clean up messaging providers and services on sandbox destroy), which added coverage to ensure
NemoClaw only tears down the dashboard forward/gateway when the destroyed sandbox was the last remaining sandbox.

If QA can still reproduce this on a newer build, please share the exact NemoClaw commit/version and destroy/onboard sequence used.