tao-port-huggingface-model

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The prompt explicitly tells the agent to "gather credentials" and to ask the user for image references/credentials (NGC/docker login), which would require the LLM to receive and potentially echo or embed secret values (e.g., in docker login/pull commands), creating an exfiltration risk.

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

• Potentially malicious external URL detected (high risk: 0.70). The skill explicitly pulls and runs TAO Toolkit container images from NVIDIA's registry (nvcr.io) during Phase 0 (e.g., docker pull "$TAO_PT_PUB_IMAGE" / references to nvcr.io and preparing tao-pytorch-base:latest), which downloads and executes remote container images at runtime and is a required external dependency.