gemini-live-api-dev

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime path ingests outsider-authored free text via the Live API’s user-provided send_realtime_input(text=...) (or equivalent sendRealtimeInput({text: ...})), which is not authored by the operating user and is fed into the model context as the user’s message.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's fallback doc workflow explicitly instructs fetching https://ai.google.dev/gemini-api/docs/llms.txt (and then linked pages like https://ai.google.dev/gemini-api/docs/live-session.md.txt) at runtime to discover and load documentation that will be used as authoritative instructions for the agent, i.e., external content directly controlling agent behavior.