cloudflare-deploy

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

• Third-party content exposure detected (high risk: 0.85). High indirect prompt-injection risk: the Agents SDK’s AIChatAgent/Agent streams LLM context from user-provided chat messages and tool outputs (e.g., this.messages and onChatMessage), and those messages are outsider-authored free text at runtime.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill shows a runtime MCP integration that registers and fetches AI tools from the MCP server URL (env.MCP_SERVER_URL e.g. "https://mcp.example.com") — those fetched tools are passed into the model (this.mcp.getAITools(...) → tools) and therefore can directly control agent behavior/prompts at runtime.