mcp-integration

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow can ingest outsider-authored free text via MCP tool results from configured external servers (e.g., SSE/HTTP/stdio servers) where the server returns arbitrary natural-language content that the agent then places into its LLM context.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes stdio examples that spawn npx to fetch and run remote npm packages at runtime (e.g., "npx -y @modelcontextprotocol/server-filesystem" which will pull code from the npm registry, e.g. https://registry.npmjs.org/@modelcontextprotocol/server-filesystem), meaning remote code is fetched and executed as part of runtime—this meets the criteria for a runtime external dependency that executes remote code.