Repository navigation
re-onboarding a healthy sandbox fails twice with different errors, then the sandbox vanishes from list #10209
Description
Description
Re-running nemoclaw onboard against an existing, healthy sandbox with unchanged provider/model settings fails twice in a row with two different errors, and afterward the sandbox disappears from nemoclaw list while still being fully resolvable and healthy via nemoclaw {name} status.
Platform scope: Reproduced on Ubuntu 26.04 only; other platforms not tested for this exact scenario.
Regression: Unknown — earlier versions not tested for this specific scenario. A related but distinct symptom (interrupted same-name recreation demanding --recreate-sandbox in a closed loop) was previously reported and fixed as #9297; this report's trigger is a plain, uninterrupted re-onboard with no SIGINT/interruption involved at any point, so it is filed as a new, distinct trigger rather than a duplicate.
Related Bugs
#9297 — "onboard --resume is unrecoverable after an interrupted same-name recreation: it demands --recreate-sandbox and repeats the same demand when that flag is supplied." Closed/fixed. That report's trigger required deliberately interrupting (SIGINT) a same-name sandbox recreation mid-flight before resuming. This report's trigger involves no interruption at all — a plain nemoclaw onboard rerun against an already-healthy, unmodified sandbox — so while the error message mentions the same --recreate-sandbox flag, the underlying trigger path is different.
Environment
Device: Ubuntu 26.04 x86_64, KVM VM, single NVIDIA A100-SXM4-40GB (host GPU passthrough)
OS: Ubuntu 26.04
Architecture: x86_64
Node.js: v22.22.3
npm: 10.9.8
Docker: Docker CE
OpenShell CLI: 0.0.106
NemoClaw: v0.0.114
OpenClaw: 2026.7.1
Steps to Reproduce
- Onboard a sandbox with NVIDIA Cloud API and a specific model, default name (e.g.
my-assistant). Wait for it to complete normally and reach Ready. - Run
nemoclaw onboardagain. At the agent prompt choose OpenClaw[1]. At provider selection choose the same provider again (NVIDIA Endpoints[1]), re-enter the API key, accept the default (same) model, accept the default (existing) sandbox name, choose "Apply configuration". Accept the web-search default (no), accept the resource-profile default (no profile), confirm the policy-preset list as shown (Enter). - Observe the wizard's exit message and code.
- Run
nemoclaw onboarda third time (same conditions). It auto-resumes the saved session (skips agent/provider/model prompts), reaches the same policy-preset confirmation screen; confirm again (Enter). - Observe the wizard's exit message and code on this second attempt.
- Run
nemoclaw listand compare againstnemoclaw {name} statusfor the same sandbox name.
Expected Result
Re-onboarding an existing, unmodified, already-healthy sandbox with the same provider and model completes successfully. If it must fail for some reason, nemoclaw list and nemoclaw {name} status should agree on whether the sandbox exists and is the default.
Actual Result
Attempt 1 exits with: "OpenClaw onboarding for '{name}' is incomplete because its recorded OpenClaw runtime identity changed or is invalid. Resume or rerun onboarding."
Attempt 2 (auto-resumed session) exits with a different error: "A previous onboarding attempt recorded sandbox '{name}' with different build or policy inputs than this run requests. Pass --recreate-sandbox to rebuild it with the current settings. Onboarding did not finish." Neither plain re-onboard step in this flow ever asked for or required --recreate-sandbox, --fresh, or --resume — those flags only appear in the error's own remediation text.
After both failed attempts, nemoclaw list no longer shows the sandbox at all (a different, unrelated sandbox is the only one listed, and no sandbox carries the default * marker), while nemoclaw {name} status for the same sandbox name still resolves it directly and reports it Ready, at a specific resource version and revision.
Logs
Attempt 1 (interactive re-onboard, same provider/model, after policy-preset confirmation [8/8]):
OpenClaw onboarding for 'my-assistant' is incomplete because its recorded
OpenClaw runtime identity changed or is invalid. Resume or rerun onboarding.
Attempt 2 (nemoclaw onboard run again; wizard auto-resumed the saved session):
A previous onboarding attempt recorded sandbox 'my-assistant' with different
build or policy inputs than this run requests.
Pass --recreate-sandbox to rebuild it with the current settings.
Onboarding did not finish. Resume from the step that failed with:
nemoclaw onboard --resume
Completed steps are skipped; pass --fresh instead to start over.
Post-attempt registry check:
$ nemoclaw list
Sandboxes:
another-sandbox
...
(no sandbox named 'my-assistant' listed; no sandbox carries the default marker)
$ nemoclaw my-assistant status
Sandbox-scoped status for 'my-assistant':
Phase: Ready
Resource version: 37
Revision: 7
...
Activity
fix(onboard): finalize reused sandbox state (NVIDIA#10236)
fix(onboard): exclude the GPU proof from the sandbox create fingerpri…