llama-cpp

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

• Third-party content exposure detected (high risk: 0.90). This skill explicitly shows loading models from public third-party sources (e.g., the Quick Start "llama-server -hf bartowski/..." and multiple HuggingFace/GitHub links in SKILL.md) and the runtime examples call into model-provided artifacts (e.g., llama_model_chat_template / chat template usage and server loading of HF models), meaning untrusted, user-generated model content can be ingested and can change prompt formatting/behavior at runtime.

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

• Potentially malicious external URL detected (high risk: 0.80). The Quick Start shows the server loading a model from the Hugging Face hub at runtime (example: "llama-server -hf bartowski/Llama-3.3-70B-Instruct-GGUF:Q4_K_M", e.g. https://huggingface.co/bartowski/Llama-3.3-70B-Instruct-GGUF), and the skill explicitly reads chat templates from the loaded model (llama_model_chat_template), so remotely fetched model content can directly control prompt formatting/behavior.