swift-mlx-lm

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly downloads and loads models, tokenizers, and config files from public hubs (e.g., "HubClient.default" / Hugging Face as shown in SKILL.md and references/model-container.md), and those untrusted third‑party model artifacts and config.json/generation_config.json are parsed and used at runtime to drive generation, tool-call parsing, and behavior—so external content can influence tool use and actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill uses HubClient.default to download model assets from the Hugging Face hub (e.g., https://huggingface.co for IDs like mlx-community/Qwen3-4B-4bit) at runtime, and those fetched config/tokenizer files and model weights can directly control chat templates, EOS tokens and tool-call formats that affect prompts/behavior.