Description

[Description]
tags v0.0.2 release
After sandbox onboarded with default model nvidia/nemotron-3-super-120b-a12b, openclaw agent --agent main --local -m "hello" replies with xml tool_call without real response.

Note: this is because the inference of https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b is down. We better to provide a friendly error message when the endpoint is not reachable rather than display xml
[Environment]

1. nemoclaw onboard to finish the sandbox creation, choose default inference option and model.
2. nemoclaw my-assistant status shows sandbox Phase: ready
3. nemoclaw my-assistant connect
4. in the sandbox, run: sandbox@my-assistant:~$ openclaw agent --agent main --local -m "hello" --session-id reonboard-test

[Expected Result]
Real response from LLM (say hello back)

[Actual Result]

no real response but some tool_call xml

nemoclaw my-assistant connect
sandbox@my-assistant:~$ openclaw agent --agent main --local -m "hello" --session-id reonboard-test
(node:1087) [UNDICI-EHPA] Warning: EnvHttpProxyAgent is experimental, expect them to change at any time.
(Use node --trace-warnings ... to show where the warning was created)
(node:1103) [UNDICI-EHPA] Warning: EnvHttpProxyAgent is experimental, expect them to change at any time.
(Use node --trace-warnings ... to show where the warning was created)

OpenClaw 2026.3.11 (29dc654) — Ah, the fruit tree company!
<tool_call>
<function=read>
<parameter=file_path>
/sandbox/.openclaw/workspace/SOUL.md


</tool_call>
sandbox@my-assistant:~$

Bug Details

[NVB# 6038604]

[NVB#6038604]