holoscan-install-wheel

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The runtime workflow fetches free-form example Python scripts and YAML from outsider-authored public sources (GitHub raw URLs) via curl -fsSL ... -o /tmp/..., then reads the YAML text and executes the downloaded .py files with python3, so outsider text can enter the LLM context indirectly through the agent’s handling of those contents.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly downloads and executes remote Python scripts from https://raw.githubusercontent.com/nvidia-holoscan/holoscan-sdk/v${SDK_VER}/examples (curl ... -o ... then python3) and requires fetching the installation doc at https://docs.nvidia.com/holoscan/sdk-user-guide/sdk_installation.html at runtime to determine which wheel to install, so external content both controls runtime behavior and is executed.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The prompt explicitly tells the user/agent to run a sudo command (sudo /opt/nvidia/holoscan/examples/download_example_data) and to install system-level Debian packages/data under /opt, and also directs downloading and executing remote scripts — actions that request elevated privileges and can modify system state.