gzh-search

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

• Third-party content exposure detected (medium risk: 0.30). Required runtime workflow takes user-provided keywords to call the doubaoya.com API and then the agent ingests the returned JSON fields (e.g., article titles/account names/publish times) to render a table and generate “选题洞察,” meaning outsider-authored text (from sources returned by the search) is read at runtime via the API response.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill calls https://doubaoya.com/api/apis/gongzhonghao/search-article/call at runtime and its README explicitly instructs relaying any top-level "notice" field from the API response verbatim to users, so remote response content can directly control agent output.