llama-cpp

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). SKILL.md explicitly instructs downloading and loading pre-quantized GGUF model files from public sources (e.g., "huggingface-cli download TheBloke/..." and links to HuggingFace) which are user/third‑party content that the skill then loads and uses for inference (./llama-cli, llama-server), so arbitrary models could embed instructions that materially change behavior.