cloudflare-agents

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly browses and scrapes arbitrary public URLs via the Browser Rendering/Puppeteer examples (e.g., templates/browser-agent.ts and references/advanced-features.md's browse the web section) and ingests external documents into RAG/Vectorize (templates/rag-agent.ts), and those untrusted page contents and documents are then passed to LLM calls (extractData, generateText/chat) for interpretation — creating a clear path for indirect prompt injection.

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

• Potentially malicious external URL detected (high risk: 0.90). The BrowserAgent calls page.goto(targetUrl) at runtime with user-supplied external URLs and then injects the fetched HTML into an LLM prompt (e.g., extractDataWithAI sends bodyContent to the model), so arbitrary external sites (any https://... URL passed as targetUrl) can directly control prompts used by the agent.