amc-setup-calibration-stack

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill asks the agent to solicit NGC and HuggingFace tokens (including "ask in chat" if no host prompt) and to embed them directly into shell commands (echo "<NGC_API_KEY>" | docker login ... and HF_TOKEN="..."), which requires the LLM to handle/output secrets verbatim and risks exfiltration.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill performs a runtime docker login to the NVIDIA NGC registry (nvcr.io) and then resolves/pulls images via docker compose, meaning remote container images from nvcr.io are fetched at runtime and will execute code in containers.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs running privileged commands (sudo usermod to add the user to the docker group and sudo chown -R on project/model directories), which request elevated rights and modify system user/group membership and filesystem ownership — actions that change the machine state and require sudo.