cloud-architect

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

• Third-party content exposure detected (high risk: 0.70). SKILL.md specifies an MCP server (“fetch”) for web content retrieval, which at runtime commonly pulls arbitrary public pages (outsider-authored free text) that the agent may read into its LLM context via the web-fetch workflow.

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

• Potentially malicious external URL detected (high risk: 0.90). The MCP Servers section includes runtime commands and endpoints that fetch and execute remote packages/services (e.g., "npx @upstash/context7-mcp" and "uvx awslabs.aws-pricing-mcp-server") and the API URL https://knowledge-mcp.global.api.aws, indicating high-confidence external dependencies that run remote code or supply runtime content to the agent.