Description

[Issue Summary]
When nemoclaw onboard fails at step 4 (Ollama auth proxy unreachable), re-running curl | bash install.sh automatically resumes the failed session with --resume, skipping provider selection. The user has no way to choose a different provider without manually deleting ~/.nemoclaw/onboard-session.json.

[Environment]
Device: brev
OS: Ubuntu 22.04
Architecture: x86_64
Node.js: v22.22.2
npm: 10.9.7
Docker: Docker Engine 29.2.1
OpenShell CLI: openshell 0.0.32
NemoClaw: v0.0.24
OpenClaw: 2026.4.2
[Steps to Reproduce]
Run: curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
During onboard, select provider: Ollama (option 7)
Ollama installs and pulls model (e.g. nemotron-3-nano:30b)
Onboard fails at step 4 with: Local Ollama is responding on 127.0.0.1, but containers cannot reach the auth proxy at http://host.openshell.internal:11435
Re-run: curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
Observe: installer prints [INFO] Found an interrupted onboarding session — resuming it. and enters (resume mode)
All steps up to and including provider selection are skipped: [resume] Skipping provider_selection (ollama-local / nemotron-3-nano:30b)
Step 4 fails again with the same error
Re-run again — same loop, no way to pick a different provider

[Expected Behavior]
Re-running curl | bash after a failed onboard should either:

Ask the user whether to resume the previous session or start fresh, OR
Detect that the failure was in provider setup and allow re-selection of the provider step, OR
Provide a --fresh / --no-resume flag to force a new session
[Actual Behavior]
[INFO] Found an interrupted onboarding session — resuming it.

NemoClaw Onboarding
(resume mode)

[1/8] Preflight checks
[resume] Skipping preflight (cached)

[2/8] Starting OpenShell gateway
[resume] Skipping gateway (running)

[3/8] Configuring inference (NIM)
[resume] Skipping provider_selection (ollama-local / nemotron-3-nano:30b)

[4/8] Setting up inference provider
✓ Active gateway set to 'nemoclaw'
Local Ollama is responding on 127.0.0.1, but containers cannot reach the auth proxy
at http://host.openshell.internal:11435. Ensure the Ollama auth proxy is running.
The user is stuck in an infinite loop — every re-install auto-resumes the failed Ollama session.

[Logs]
Root cause in scripts/install.sh line 1195-1210 (v0.0.24):

if command_exists node && [[ -f "${HOME}/.nemoclaw/onboard-session.json" ]]; then
if node -e '
const data = JSON.parse(fs.readFileSync(file, "utf8"));
const resumable = data && data.resumable !== false;
const status = data && data.status;
process.exit(resumable && status && status !== "complete" ? 0 : 1);
' "${HOME}/.nemoclaw/onboard-session.json"; then
info "Found an interrupted onboarding session — resuming it."
onboard_cmd+=(--resume)
fi
fi
The check only considers resumable !== false && status !== "complete". A session that failed at step 4 has resumable=true, status=in_progress, so it always triggers auto-resume. There is no escape hatch.
[Impact]
P2 — Users who hit Ollama proxy failures (or any step 3-4 failure) are stuck in a resume loop. Workaround: rm ~/.nemoclaw/onboard-session.json then re-run installer.

[Suspected Root Cause / Notes]
install.sh auto-resume logic is too aggressive — it resumes any non-complete session without giving the user a choice. The --resume flag in onboard-command.ts skips all completed steps including provider_selection, so users cannot change their provider choice after a failure.

Possible fixes:

Add interactive prompt before auto-resume: "Resume previous session (Ollama) or start fresh? [R/f]"
Clear the session file if the failure was in provider setup (steps 3-4)
Add --fresh flag to nemoclaw onboard and install.sh to bypass resume
Mark session as resumable=false when the failure is in provider_selection or inference_setup

Bug Details

[NVB#6110488]