source-evidence-dataset-development

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

• Third-party content exposure detected (low risk: 0.10). The workflow retrieves and consumes web pages, URLs, and scientific literature via search and retrieval tools, exposing the agent to low-risk outsider-authored free text.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill references an external GitHub repository URL via npx skills use https://github.com/tiangong-ai/skills to dynamically fetch and load a runtime agent skill into the context during execution.