event-driven

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow uses the read_url tool to fetch full text of news/announcements from arbitrary URLs/pages, and that fetched outsider-authored free text is then ingested by the LLM for scoring via the {news_content} prompt (public web content → LLM context), creating an indirect prompt-injection risk.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill uses the built-in read_url tool at runtime to fetch arbitrary external news URLs whose fetched content is then injected verbatim into the LLM scoring prompt (i.e., external news URLs fetched via read_url), so runtime-fetched URLs directly control the model's instructions.