modal-llm-serving

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

• Third-party content exposure detected (high risk: 0.70). The skill explicitly pulls and loads models from the public Hugging Face Hub (see SKILL.md guidance to pin model revisions and scripts/qwen3_throughput.py which installs huggingface-hub, sets HF_XET_HIGH_PERFORMANCE, and uses MODEL_NAME/MODEL_REVISION and a Hugging Face cache Volume), meaning untrusted user-published model artifacts are ingested and executed by the runtime and could influence generated outputs and subsequent behavior.