ml-experiment

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

• Third-party content exposure detected (high risk: 1.00). The skill's SKILL.md explicitly calls the arxiv_to_prompt tool and in the "Reproduce a Paper" workflow says "Fetch paper using arxiv_to_prompt" and "Extract key method details" so the agent will ingest public arXiv papers (third-party content) and act on their instructions to implement experiments, enabling indirect prompt injection.