skills/nvidia/skills/hsb-setup/Gen Agent Trust Hub

hsb-setup

Pass

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Performs privileged system configuration on remote devkits via SSH heredoc blocks.
  • Evidence: Uses sudo nmcli for network management and sudo sysctl for receive buffer configuration.
  • Evidence: Modifies system persistence by creating jetson_clocks.service in /etc/systemd/system/.
  • Evidence: Automates user group management using sudo usermod -aG docker.
  • [EXTERNAL_DOWNLOADS]: Downloads source code and system packages from remote sources.
  • Evidence: Clones the project repository from GitHub (defaulting to the official NVIDIA Holoscan organization).
  • Evidence: Installs multiple build dependencies and libraries using apt-get install.
  • [CREDENTIALS_UNSAFE]: Accesses and guides the deployment of SSH keys to facilitate automated remote access.
  • Evidence: Scans the local ~/.ssh/ directory for existing public keys (id_ed25519.pub, etc.).
  • Evidence: Provides user-facing commands to append local keys to the remote devkit's ~/.ssh/authorized_keys file.
  • [PROMPT_INJECTION]: Vulnerable to indirect prompt injection through external repository content.
  • Ingestion points: Reads docs/user_guide/setup.md and docs/user_guide/build.md from the cloned repository.
  • Boundary markers: Not used when interpolating or studying repository content.
  • Capability inventory: The skill has extensive capabilities including Bash execution, sudo privileges on the devkit, and file write access.
  • Sanitization: No explicit validation or sanitization is performed on the content retrieved from the repository.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 2, 2026, 05:03 PM
Security Audit — agent-trust-hub — hsb-setup