Skip to content

Sandbox image build includes nemoclaw-blueprint/.venv from local tree (breaks build / risks leaking secrets) #774

@mercl-lau

Description

@mercl-lau

Description

When onboarding builds the sandbox image from a local NemoClaw checkout, the staging step copies the whole nemoclaw-blueprint tree into the Docker build context (cp -r in bin/lib/onboard.js). If a developer (or any local workflow) has created a .venv directory under nemoclaw-blueprint—for example after running Python tooling, pytest, nv run, nv sync—that path is copied along with the rest of the blueprint and ends up in the image via COPY nemoclaw-blueprint/ in the Dockerfile.

That is problematic for two reasons:

Build failures — packaging can choke on or mishandle .venv in ways that don’t affect users who only use the released installer without a local .env.
Security — .venv often holds secrets; it must not be baked into image layers.
End users who never have a local .env under nemoclaw-blueprint may never see this, but the build path should be hygienic regardless.

Reproduction Steps

Repro (typical dev):

  1. Clone NemoClaw and work under nemoclaw-blueprint with Python (e.g. uv sync that create nemoclaw-blueprint/.venv).
    Run onboarding so it builds the sandbox image from the local repo (e.g. nemoclaw onboard after curl … | bash or equivalent using this tree).
    Observe build/packaging errors involving .venv, or inspect image/context and see .env included.

Expected: Local-only artifacts (at minimum .env, and ideally .venv / pycache / .pytest_cache) are excluded from the Docker build context when staging nemoclaw-blueprint.

Image

Environment

openClaw: 2026.3.11 (29dc654)
OpenShell: 0.0.14
macOS 26.1 (Apple M4, arm64)
Policy presets applied: pypi, npm
Inference: NVIDIA Cloud API (nvidia/nemotron-3-super-120b-a12b)

Debug Output

Logs

Checklist

  • I confirmed this bug is reproducible
  • I searched existing issues and this is not a duplicate

Metadata

Metadata

Assignees

No one assigned

    Labels

    DockerSupport for Docker containerizationGetting StartedUse this label to identify setup, installation, or onboarding issues.NV QABugs found by the NVIDIA QA TeambugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions