tao-run-on-docker

Warn

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill documents a process for relocating the Docker data root that involves multiple commands requiring administrative privileges. This includes stopping and starting system services, modifying configuration files in protected directories (/etc/docker/daemon.json), and performing file operations on system-level paths using sudo.
  • [COMMAND_EXECUTION]: The skill invokes a setup script (setup-nvidia-gpu-host.sh) from a relative path to verify or install the NVIDIA GPU host runtime environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides commands to read data from external container environments, creating a potential surface for indirect prompt injection.
  • Ingestion points: The agent is instructed to use docker logs and docker inspect, which retrieve output and metadata from running containers.
  • Boundary markers: There are no instructions for using delimiters or boundary markers when processing the output of these commands.
  • Capability inventory: The skill enables shell command execution, system service management, and file system modifications.
  • Sanitization: No sanitization or validation steps are defined for data ingested from container logs or inspection results.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 2, 2026, 04:54 PM
Security Audit — agent-trust-hub — tao-run-on-docker