serving-llms-on-instinct

Pass

Audited by Gen Agent Trust Hub on Aug 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run across several scripts (detect.py, validate.py, sync_recipes.py, evals.py) to execute local and remote (via SSH) shell commands for system discovery and Docker management. These operations are within the scope of its primary purpose of hardware-specific LLM serving.
  • [EXTERNAL_DOWNLOADS]: The skill downloads model recipes from github.com/vllm-project/recipes and queries Docker Hub (hub.docker.com) for the latest stable vLLM ROCm image tags. These are official, well-known sources relevant to the task.
  • [SAFE]: The skill demonstrates security best practices by recommending the use of .env files for secret management (e.g., HF_TOKEN) and includes a blacklist.json to prevent the agent from attempting to serve incompatible or risky model types. Verification steps and health checks are integrated into the deployment workflow to ensure endpoint stability.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 16, 2026, 01:20 AM
Security Audit — agent-trust-hub — serving-llms-on-instinct