gemini-tts

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

• Third-party content exposure detected (high risk: 0.70). The runtime LLM prompt is built from segments.json (or --text) via buildPrompt()/segments.map(s => s.text) and sent to Gemini in generateAudio() as contents: [{ parts: [{ text: prompt }] }]; if those segment texts come from outsiders (e.g., downloaded/forwarded content or public web text), that free-form outsider text is injected into the LLM context.

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

• Potentially malicious external URL detected (high risk: 0.80). The script auto-installs and imports the remote npm package "@google/genai" at runtime via "npm install --no-save @google/genai" (fetching from the npm registry, e.g. https://registry.npmjs.org/@google/genai), which downloads and executes third‑party code required for generating the TTS prompts/audio.