deep-research

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

• Third-party content exposure detected (high risk: 0.85). The skill submits the user-provided query to Gemini Deep Research and then ingests the model’s completed interaction.outputs[].text (free-form Markdown) into the agent context via formatOutput()/stdout, which can include web-grounded content sourced from public pages fetched at runtime (outsider-authored text).