Description
On a fresh Windows ARM WSL2 install (no OpenShell present), running bash install.sh from the NemoClaw source tree completes [1/3] Node.js and [2/3] NemoClaw CLI successfully, then aborts at [3/3] Onboarding with:
[WARN] Host preflight found issues that will prevent onboarding right now.
- Install OpenShell: OpenShell is required before onboarding can create or manage a gateway.
Run the NemoClaw installer or `scripts/install-openshell.sh`.
This is circular advice: the user is currently running the NemoClaw installer (bash install.sh). The actual workaround is to run scripts/install-openshell.sh manually after install.sh completes — but the wording sends the user back to re-run the installer that just told them to re-run it.
Either:
- (a)
install.sh should run scripts/install-openshell.sh itself as part of step [2/3] or [3/3], OR
- (b) The error message should explicitly say "Run
scripts/install-openshell.sh" and drop the "Run the NemoClaw installer" suggestion (which is the loop the user is already in).
Environment
Device: Windows ARM reference host (Snapdragon X laptop)
OS: Ubuntu 24.04.4 LTS Noble Numbat inside WSL2
Architecture: aarch64 (Snapdragon X)
Node.js: v22.22.2
npm: 10.9.7
Docker: 29.1.3, build 29.1.3-0ubuntu3~24.04.2
OpenShell CLI: N/A (not installed at install.sh time)
NemoClaw: v0.1.0 (main HEAD cfa817b — installed by this run)
OpenClaw: N/A (onboard never completed)
Steps to Reproduce
- Fresh WSL2 Ubuntu-24.04 (no NemoClaw, no OpenShell, no nemoclaw artifacts in
~/.local/bin or ~/.nemoclaw).
- Clone NemoClaw main HEAD:
git clone --depth 1 https://github.com/NVIDIA/NemoClaw.git
- Run install.sh:
cd NemoClaw
NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1 NEMOCLAW_NON_INTERACTIVE=1 \
bash install.sh --yes-i-accept-third-party-software
- install.sh completes
[1/3] Node.js and [2/3] NemoClaw CLI successfully. At [3/3] Onboarding, observe the error.
Expected Result
Either:
- (a) install.sh runs
scripts/install-openshell.sh as part of its bootstrap so the user gets OpenShell installed in the same flow, OR
- (b) install.sh's "OpenShell missing" error says exactly which script to run, without suggesting the installer the user is already running.
In either case the install.sh flow should be self-sufficient on a fresh box (Node → NemoClaw CLI → OpenShell → onboard). Today it stops one step short and leaves a confusing breadcrumb.
Actual Result
install.sh aborts at [3/3] Onboarding with:
[3/3] Onboarding
──────────────────────────────────────────────────
Detected container runtime: docker
Running under WSL
[WARN] Host preflight found issues that will prevent onboarding right now.
- Install OpenShell: OpenShell is required before onboarding can create or manage a gateway.
Run the NemoClaw installer or `scripts/install-openshell.sh`.
- Install NVIDIA Container Toolkit and generate CDI device specs: …
…
nemoclaw onboard # or rerun with --no-gpu to skip GPU passthrough
[WARN] Skipping onboarding until the host prerequisites above are fixed.
[INFO] === Installation complete ===
NemoClaw (30s)
NemoClaw CLI is installed.
Onboarding did not run.
To finish setup, run:
$ source /home/lab/.bashrc
$ nemoclaw onboard
User experience:
nemoclaw onboard still fails because OpenShell is genuinely not installed.- User has to discover separately that the fix is
bash ~/NemoClaw-main/scripts/install-openshell.sh, not "re-run install.sh".
Logs
=== ~/ARM64_phase2.log excerpt — install.sh [3/3] Onboarding ===
[3/3] Onboarding
──────────────────────────────────────────────────
Detected container runtime: docker
Running under WSL
[WARN] Host preflight found issues that will prevent onboarding right now.
- Install OpenShell: OpenShell is required before onboarding can create or manage a gateway.
Run the NemoClaw installer or `scripts/install-openshell.sh`.
- Install NVIDIA Container Toolkit and generate CDI device specs: Docker is configured for CDI device injection (CDISpecDirs is set) but the `nvidia-container-toolkit` package (which provides `nvidia-ctk`) is not installed on the host. OpenShell's `gateway start --gpu` will fail with `unresolvable CDI devices nvidia.com/gpu=all` until the toolkit is installed and a CDI spec is generated.
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
…
nemoclaw onboard # or rerun with --no-gpu to skip GPU passthrough
[WARN] Skipping onboarding until the host prerequisites above are fixed.
[INFO] === Installation complete ===
NemoClaw (30s)
NemoClaw CLI is installed.
Onboarding did not run.
To finish setup, run:
$ source /home/lab/.bashrc
$ nemoclaw onboard
=== workaround (manually run install-openshell.sh) ===
$ cd ~/NemoClaw-main
$ bash scripts/install-openshell.sh
[install] Detected Linux (aarch64)
[install] Installing OpenShell from release 'v0.0.39'...
[install] Verifying SHA-256 checksum...
[install] Installed openshell to /home/lab/.local/bin/openshell (user-local path)
[install] openshell 0.0.39 installed
# Now nemoclaw onboard works (after also setting NEMOCLAW_SANDBOX_GPU=0 + Ollama systemd override; see NVB#6199809, NVB#6199755).
Related
Possibly a regression of NVB#6000851 / GH#708 (Closed Fixed 2026-04-16) which described "nemoclaw onboard will automatically install openshell in the PATH ~/.local/bin/openshell" — that bug suggested auto-install was working. Today on Windows ARM reference host main HEAD cfa817b, install.sh's onboard step does not auto-install OpenShell and instead aborts with a circular-advice error.
Discovered during ARM64 validation for PR #3925 QA but is not PR-introduced — reproduces on main HEAD cfa817b.
This is the 5th bug in the Windows ARM reference host-install-experience series I'm filing today:
- NVB#6198894 (PR-induced regression on Spark+Station ARM64)
- NVB#6199735 (rebuild destroys --no-gpu sandbox)
- NVB#6199755 (openshell-docker-gateway idle shutdown)
- NVB#6199809 (preflight false-positive Snapdragon iGPU as NVIDIA GPU)
- this one
NVB#6199869
Description
On a fresh Windows ARM WSL2 install (no OpenShell present), running
bash install.shfrom the NemoClaw source tree completes[1/3] Node.jsand[2/3] NemoClaw CLIsuccessfully, then aborts at[3/3] Onboardingwith:This is circular advice: the user is currently running the NemoClaw installer (
bash install.sh). The actual workaround is to runscripts/install-openshell.shmanually after install.sh completes — but the wording sends the user back to re-run the installer that just told them to re-run it.Either:
install.shshould runscripts/install-openshell.shitself as part of step[2/3]or[3/3], ORscripts/install-openshell.sh" and drop the "Run the NemoClaw installer" suggestion (which is the loop the user is already in).Environment
Steps to Reproduce
~/.local/binor~/.nemoclaw).cd NemoClaw NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1 NEMOCLAW_NON_INTERACTIVE=1 \ bash install.sh --yes-i-accept-third-party-software[1/3] Node.jsand[2/3] NemoClaw CLIsuccessfully. At[3/3] Onboarding, observe the error.Expected Result
Either:
scripts/install-openshell.shas part of its bootstrap so the user gets OpenShell installed in the same flow, ORIn either case the install.sh flow should be self-sufficient on a fresh box (Node → NemoClaw CLI → OpenShell → onboard). Today it stops one step short and leaves a confusing breadcrumb.
Actual Result
install.sh aborts at
[3/3] Onboardingwith:User experience:
nemoclaw onboardstill fails because OpenShell is genuinely not installed.bash ~/NemoClaw-main/scripts/install-openshell.sh, not "re-run install.sh".Logs
Related
Possibly a regression of NVB#6000851 / GH#708 (Closed Fixed 2026-04-16) which described "nemoclaw onboard will automatically install openshell in the PATH
~/.local/bin/openshell" — that bug suggested auto-install was working. Today on Windows ARM reference host main HEADcfa817b, install.sh's onboard step does not auto-install OpenShell and instead aborts with a circular-advice error.Discovered during ARM64 validation for PR #3925 QA but is not PR-introduced — reproduces on
mainHEADcfa817b.This is the 5th bug in the Windows ARM reference host-install-experience series I'm filing today:
NVB#6199869