agent-as-a-router-coding

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). Several links point to assets hosted by individual/unvetted accounts (a GitHub repo that the instructions clone and pip-install, Hugging Face model/dataset under a personal username, and small personal domains), which can deliver arbitrary code or malicious artifacts and are therefore potentially suspicious.

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

• Third-party content exposure detected (high risk: 0.70). The required runtime workflow includes downloading Hugging Face assets/datasets (public web content) via scripts/download_hf_assets.py and then running evaluation with --hf-dataset-dir, so outsider-authored dataset text can be ingested into the agent’s LLM prompt context through the task prompt field.