transformers-js

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly loads models and files from the public Hugging Face Hub (e.g., SKILL.md and references/CONFIGURATION.md show env.allowRemoteModels=true and links like https://huggingface.co/models?... and pipeline(...) calls that download model repos), so untrusted user-contributed model files/model cards from the open web can be fetched and directly affect model behavior and downstream actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The browser examples import and execute remote JavaScript at runtime from the CDN URL https://cdn.jsdelivr.net/npm/@huggingface/transformers (including the versioned variant https://cdn.jsdelivr.net/npm/@huggingface/transformers@3.8.1), which is a runtime-fetched dependency that will execute remote code as shown in the examples.