agency-carousel-growth-engine

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested via the Playwright website scraping step (analyze-web.js), which reads arbitrary user-supplied target URLs and extracts their page text (features/pricing/testimonials/competitors) into analysis.json, then that extracted prose is used to generate prompts for the LLM/Gemini.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill uses Playwright at runtime to scrape arbitrary target website URLs (the user-supplied target site, e.g., https://) and injects that fetched site content into the Gemini slide-generation prompts, meaning remote site content directly controls prompt instructions during execution.