huggingface-llm-trainer

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

• Third-party content exposure detected (medium risk: 0.30). scripts/dataset_inspector.py calls the Hugging Face Datasets Server API (e.g., /rows and then prints a human-readable “EXAMPLE DATA” preview) for a user-supplied dataset name/split, meaning outsider-authored dataset content is ingested into the runtime workflow before the agent selects a specific “item”.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill contains examples and runtime invocations that fetch and execute remote scripts (e.g., hf_jobs calling https://huggingface.co/datasets/mcp-tools/skills/raw/main/dataset_inspector.py and scripts that git-clone https://github.com/ggerganov/llama.cpp.git and build/run code), so remote content will be fetched at runtime and can execute code.