Description

[Description]
After successfully running openshell inference set --model <new> --provider <p> on a Running NemoClaw sandbox, nemoclaw list and nemoclaw <name> status continue to show the old (onboard-time) model and provider indefinitely. The gateway itself has the new model in effect — openshell inference get confirms the change — but the NemoClaw CLI's display is stale.

[Environment]
Device: x86_64 Ubuntu 22.04 host (reproduced on a viking-evt1-108 test box);
OS: Ubuntu 22.04 (sandbox container base); host is Ubuntu 22.04 x86_64
Architecture: x86_64
Node.js: v22.x (shipped inside OpenClaw image)
npm: bundled
Docker: Docker Engine (containerd/k3s runtime inside openshell-cluster-nemoclaw container)
OpenShell CLI: openshell 0.0.32
NemoClaw: v0.0.23
OpenClaw: 2026.4.2 (d74a122)
Provider: ollama-local (type=openai), upstream Ollama 0.20.3 with nemotron-3-nano:30b
Sandbox image: built with NEMOCLAW_INFERENCE_BASE_URL=https://inference.local/v1 (baked via Docker ARG)

[Steps to Reproduce]

1. nemoclaw onboard a sandbox (e.g. name tel) with provider nvidia-prod and model nvidia/nemotron-3-super-120b-a12b.

2. After onboard completes, confirm baseline display:

    $ nemoclaw list
    Sandboxes:
      tel *
        model: nvidia/nemotron-3-super-120b-a12b  provider: nvidia-prod  CPU  policies: ...
   

3. Change the model at runtime:

    $ openshell inference set -g nemoclaw --model moonshotai/kimi-k2.5 --provider nvidia-prod
    Gateway inference configured:
      Route: inference.local
      Provider: nvidia-prod
      Model: moonshotai/kimi-k2.5
      Version: 3
   

4. Confirm the gateway now has the new model:

    $ openshell inference get
    Gateway inference:
      Provider: nvidia-prod
      Model: moonshotai/kimi-k2.5
   

5. Re-run nemoclaw list (and nemoclaw tel status).

[Expected Behavior]
nemoclaw list / status should reflect the model and provider currently in effect at the gateway — i.e. match what openshell inference get returns. Alternatively, if the display is
intentionally "as-onboarded", NemoClaw should either document that explicitly or annotate the row to indicate the live model differs (e.g. model: <onboarded> (live: <current>)), so
users don't mistake the stale row for current reality.

[Actual Behavior]
Both commands still show the original onboard-time model:

      $ nemoclaw list
      Sandboxes:
        tel *
          model: nvidia/nemotron-3-super-120b-a12b  provider: nvidia-prod  CPU  policies: ...

[Logs]
User session excerpt:

      $ openshell inference set -g nemoclaw -m moonshotai/kimi-k2.5 -p nvidia-prod
      error: unexpected argument '-m' found
      Usage: openshell inference set [OPTIONS] --provider <PROVIDER> --model <MODEL>

      $ openshell inference set -g nemoclaw --model moonshotai/kimi-k2.5 --provider nvidia-prod
      Gateway inference configured:
        Route: inference.local
        Provider: nvidia-prod
        Model: moonshotai/kimi-k2.5
        Version: 3

Bug Details

[NVB#6108117]