byob

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

• Third-party content exposure detected (high risk: 0.90). This skill explicitly reads user-provided data files and public HuggingFace datasets (hf:// URIs) in "Step 2
• Data" and loads prompt/template files (e.g., prompt.jinja2, system_prompt, response_field) which are then used to generate prompts, scorers, and runtime behavior, so untrusted third-party content can directly influence prompt construction and tool actions.