viral-short-video

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

• Third-party content exposure detected (medium risk: 0.30). The runtime path generate_viral_script.py ingests outsider-provided free text via CLI parameters like --topic/--product/--target into the LLM prompt construction (build_script_prompt()), which then drives generated viral_script.json and is subsequently used for storyboard generation.

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

• Potentially malicious external URL detected (high risk: 0.80). The setup script downloads YouTube videos at runtime (e.g. https://www.youtube.com/watch?v=zIvOY4JTkbs) and those fetched videos are later converted to keyframes and injected into Gemini Vision / content-generation prompts, meaning external content directly influences the agent's prompt inputs and outputs.