mcp-builder

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

• Third-party content exposure detected (high risk: 0.80). The skill's Phase 1 instructions and Reference Files explicitly direct the agent to fetch and load open web content (e.g., "https://modelcontextprotocol.io/sitemap.xml", arbitrary .md pages, SDK READMEs from https://raw.githubusercontent.com and via "web search" / WebFetch), so the agent will ingest third‑party public content that can influence its decisions and tool use.