gsearch

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

• Third-party content exposure detected (high risk: 0.85). Yes: the required gsearch follow <url> workflow navigates to a user-supplied URL and extracts the page’s readable text/JSON via CDP (Runtime.evaluate reading document.body.innerText / document.querySelector(...).innerText), so outsider-authored web content is ingested into the agent’s LLM context as the command output.