web-search-tavily

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

• Third-party content exposure detected (high risk: 0.95). The skill’s runtime workflow calls Tavily endpoints like /search, /extract, /map, and /crawl, which fetch and extract arbitrary public web page content (outsider-authored) and then feed that extracted text/snippets back into the agent’s LLM context for summarization/research.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches arbitrary user-supplied URLs at runtime via Tavily Extract/Map/Crawl (e.g., "https://example.com/article1", "https://example.com", "https://docs.example.com"), and the retrieved page content is returned/injected into the agent's response context, which can directly control prompts or outputs.