Repository navigation
[DGX Spark][DGX Station][Onboard] Published Deep Agents base image fails NemoClaw v0.0.76 version check #6456
Description
Description
Fresh LangChain Deep Agents Code onboarding on DGX Spark selects a published sandbox base containing deepagents-code 0.1.12, while NemoClaw v0.0.76 requires 0.1.34. Sandbox image construction stops and onboarding exits nonzero.
Environment
Device: DGX Spark
OS: Ubuntu 24.04.4 LTS
Architecture: aarch64
Node.js: v22.23.1
npm: 10.9.8
Docker: Docker version 29.2.1, build a5c7197
OpenShell CLI: openshell 0.0.72
NemoClaw: nemoclaw v0.0.76
OpenClaw: N/A (Deep Agents onboarding did not complete)
Steps to Reproduce
-
Authenticate to the NGC container registry with
docker login nvcr.io. -
On a DGX Spark with no matching local Deep Agents base, run:
NEMOCLAW_GATEWAY_PORT=[unused-port] NEMOCLAW_PROVIDER=install-vllm NEMOCLAW_AGENT=langchain-deepagents-code NEMOCLAW_SANDBOX_NAME=[new-name] nemoclaw onboard --non-interactive --fresh --yes --yes-i-accept-third-party-software --agent langchain-deepagents-code --name [new-name]
-
Wait for the managed Qwen3.6-35B-A3B-NVFP4 vLLM endpoint to become healthy.
-
Observe the selected published base image and sandbox build result.
Expected Result
The published NemoClaw v0.0.76 Deep Agents base contains deepagents-code 0.1.34, the sandbox image builds, and onboarding completes.
Actual Result
Using LangChain Deep Agents Code base image: ghcr.io/nvidia/nemoclaw/langchain-deepagents-code-sandbox-base@sha256:4e03801ca735f33302835116bcd9d9e802269f7698cdb6d38c41c91934bddd93
ERROR: expected deepagents-code==0.1.34, found 0.1.12; dependency drift requires reviewing whether upstream now recognizes both managed aliases and removing this bridge when it does
Sandbox image build stopped after 20.6s
The onboarding process exited with status 1.
Logs
#20 0.174 ERROR: expected deepagents-code==0.1.34, found 0.1.12; dependency drift requires reviewing whether upstream now recognizes both managed aliases and removing this bridge when it does
ERROR: failed to build: failed to solve: process "/bin/sh -c python /tmp/patch-managed-deepagents-code.py" did not complete successfully: exit code: 1
Building the base from the exact NemoClaw v0.0.76 source produced deepagents-code 0.1.34, and onboarding succeeded when that locally built base was explicitly selected.
Activity
[-][DGX Spark][Onboard] Published Deep Agents base image fails NemoClaw v0.0.76 version check[/-][+][DGX Spark][DGX Station][Onboard] Published Deep Agents base image fails NemoClaw v0.0.76 version check[/+]hulynn commentedon Jul 8, 2026
[QA Revalidation] Also reproduced on DGX Station with NemoClaw v0.0.76.
Environment:
Tested the exact published image content from the original DGX Spark reproduction:
sha256:4e03801ca735f33302835116bcd9d9e802269f7698cdb6d38c41c91934bddd93
The image contains:
When this image was used as the Deep Agents sandbox base on DGX Station, onboarding failed with the same error:
ERROR: expected deepagents-code==0.1.34, found 0.1.12; dependency drift requires reviewing whether upstream now recognizes both managed aliases and removing this bridge when it does
The sandbox image build stopped after 21.0 seconds, sandbox creation failed, and onboarding returned exit code 1.
Conclusion:
This issue is not DGX Spark-specific. It is a published Deep Agents base-image artifact/version-validation issue. DGX Station also fails whenever it resolves or is forced to use this stale published image.
The earlier successful Station Express E2E used an existing local base image containing deepagents-code 0.1.34, so it did not exercise the affected published image.
fix(dcode): publish and validate sandbox base image (#6469)
fix(base-image): require release-matched sandbox bases (NVIDIA#6624)
fix(dcode): publish and validate sandbox base image (NVIDIA#6469)
fix(base-image): require release-matched sandbox bases (NVIDIA#6624)