nano-banana

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

• Third-party content exposure detected (high risk: 0.75). The required runtime workflow can ingest outsider-authored free text via --deck mode: run_deck() loads a user-supplied deck_spec.json and uses slide fields like heading, visual, text_panel, and legacy prompt directly to build the Gemini request body (build_request_body() → call_api()), so any text authored by someone other than the operating user in that JSON becomes LLM context.