Description

Description

On DGX Station GB300, exporting NEMOCLAW_PROVIDER=build before NemoClaw v0.0.43
curl|bash correctly suppresses the "Run express install? [Y/n]" prompt but
does NOT skip the standard inference-provider menu (Inference options 1-8).
After Station detection, the wizard reaches [3/8] Configuring inference (NIM)
and prints the full menu, then waits at "Choose [1]:" for user input.

Per test case T6016435 spec: "installer must honour the chosen provider and
skip both the express prompt and the inference-provider menu". Express prompt
skip works (✓); provider-menu skip does NOT (✗). This breaks the
non-interactive provider-preselection contract that CI / IT scripts rely on.

Environment

Device:        DGX Station GB300 (galaxy-sku2-018, host 10.176.192.158)
OS:            Ubuntu 24.04.4 LTS
Architecture:  aarch64
Kernel:        6.17.0-1014-nvidia-64k
GPU:           NVIDIA GB300 (256703 MB) + NVIDIA RTX PRO 6000 Blackwell Max-Q (97887 MB)
NVIDIA driver: 610.39
Node.js:       v22.22.3 (auto-installed by curl|bash)
npm:           10.9.8
Docker:        29.5.0 (build 98f1464)
nvidia-ctk:    1.19.0
OpenShell CLI: 0.0.39
NemoClaw:      v0.0.43
OpenClaw:      N/A (wizard cut off at the inference menu — onboard not completed)

Steps to Reproduce

1. Start from a clean Station (no nemoclaw / openshell binaries, no ~/.nemoclaw)
2. export NEMOCLAW_PROVIDER=build
3. Run: curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
4. Type 'yes' to accept the third-party software notice
5. Observe Station detection at the platform-detection step
6. Observe [1/8] Preflight, [2/8] Gateway, [3/8] Configuring inference (NIM)

Expected Result

Per T6016435 spec:
   - "Run express install? [Y/n]" prompt is suppressed (no express offer)
   - Inference-provider menu (Inference options 1-8) is suppressed
   - Installer honours NEMOCLAW_PROVIDER=build and proceeds to model / API key
     setup for the "build" (= NVIDIA Endpoints) provider non-interactively

Actual Result

Express prompt suppression works (✓):
   - "Run express install? [Y/n]" never appears after Station detection
   - Wizard proceeds directly to [1/8] Preflight

Provider-menu suppression does NOT work (✗):
   - [3/8] Configuring inference (NIM) prints the full 8-option menu and
     waits for user input:

     [3/8] Configuring inference (NIM)
       ──────────────────────────────────────────────────
       Inference options:
         1) NVIDIA Endpoints
         2) OpenAI
         3) Other OpenAI-compatible endpoint
         4) Anthropic
         5) Other Anthropic-compatible endpoint
         6) Google Gemini
         7) Install Ollama (Linux)
         8) Model Router (experimental)
       Choose [1]:

The NEMOCLAW_PROVIDER=build env var is effectively ignored at this step.
Without NVIDIA_API_KEY also set in env, the wizard then blocks on
"NVIDIA API Key:" — meaning a CI run depending on env-preselection still hangs.

Logs

Reproduced 2026-05-15 ~06:05 UTC. Full transcript saved at
  /home/lab/day0-automation/20260511/station1-t6016435-provider-preset.txt
on local-lab.

Impact

Blocks DevTest task 580931 (v0.0.43 T6016435) from being marked Pass per spec.
Non-interactive provider pre-selection — a documented installer feature —
silently degrades: env var still suppresses express prompt, but the standard
provider menu reappears, defeating the "no user input needed" intent.
Customers using IT/CI playbooks that rely on NEMOCLAW_PROVIDER alone will
see scripts hang at the provider menu on v0.0.43.

Related but distinct: T6016434 (NEMOCLAW_NO_EXPRESS=1) and T6016436
(NEMOCLAW_NON_INTERACTIVE=1 + NVIDIA_API_KEY) both work correctly on v0.0.43.
Only the NEMOCLAW_PROVIDER pre-set path is broken.

Bug Details

[NVB#6179556]