deep-research

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

• Third-party content exposure detected (high risk: 0.85). The skill sends the user’s query to Gemini Deep Research, which at runtime performs web research and ingests scraped public web page text (outsider-authored free text) into the model’s context to synthesize the final report.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill calls the Gemini API at runtime (https://generativelanguage.googleapis.com/v1beta/interactions and the SSE variant ?alt=sse) to send built prompts and receive model outputs that directly determine the agent's behavior and results, making this an active external dependency that controls prompts.