serving-llms-on-instinct

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

• Potentially malicious external URL detected (high risk: 0.90). scripts/sync_recipes.py shallow-clones the remote repo at https://github.com/vllm-project/recipes.git and queries https://hub.docker.com/v2/repositories/vllm/vllm-openai-rocm/tags at runtime to obtain recipes and the Docker image tag which are then used to construct vLLM/docker command-line args and environment (i.e., remote content directly controls execution parameters).

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill instructs the user to run a privileged command ("sudo usermod -aG video,render $USER") which modifies system group membership and requires sudo, thus directly changing the machine's state.