Evals

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

• Third-party content exposure detected (high risk: 0.80). The skill's Data/DomainPatterns.yaml (research domain) and grader configs explicitly require tool_calls like "web_search" and "read_source", which means the agent is expected to fetch and evaluate external/public web sources (untrusted, user-generated content) as part of its grading/workflow and those external contents can influence scoring and subsequent decisions.