ptq

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

• Third-party content exposure detected (high risk: 0.75). High: the required workflow instructs reading and inspecting the downloaded model’s own files (e.g., modeling_*.py, README.md, config.json) and using trust_remote_code, so arbitrary outsider-authored free text from the model repository can be ingested into the agent/LLM context via the runtime inspection steps.

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

• Potentially malicious external URL detected (high risk: 0.90). The docs instruct cloning and installing the live HuggingFace transformers repo at runtime via "git clone --depth 1 https://github.com/huggingface/transformers.git", which fetches and installs remote code that would be executed/used as a runtime dependency.