tao-run-inference-service

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess to interact with container engines and platform tools (Docker, kubectl, Slurm). It prevents shell injection by using shlex.quote() on all user-controlled data (like model_path) and by generating unique identifiers (UUIDs) for job IDs, which are then used in command strings.
  • [SAFE]: Secrets such as HuggingFace tokens and Weights & Biases API keys are managed exclusively via host environment variables. The skill's internal logic and templates ensure these values are never hardcoded, logged, or requested via cleartext prompts.
  • [SAFE]: The skill manages service state through a local registry file in /tmp/. While this involves file system operations, the data stored is limited to service metadata required for routing requests and stopping services, which is consistent with the skill's operational requirements.
  • [SAFE]: Container images are resolved from a trusted vendor registry (nvcr.io). The resolution logic uses a manifest file (versions.yaml) to map architecture-specific keys to verified image URIs.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 05:48 PM
Security Audit — agent-trust-hub — tao-run-inference-service