ai-specialist

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

• Third-party content exposure detected (high risk: 0.90). The mcp-builder SKILL.md (Phase 1) explicitly tells the agent to use WebFetch / web search to load external resources — e.g., https://modelcontextprotocol.io/llms-full.txt and raw GitHub READMEs — and to read those documents to drive tool selection and implementation, which exposes the agent to untrusted third-party content that can materially influence next actions.

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

• Potentially malicious external URL detected (high risk: 1.00). The mcp-builder skill explicitly instructs using WebFetch at runtime to load external specifications and SDK READMEs (https://modelcontextprotocol.io/llms-full.txt, https://raw.githubusercontent.com/modelcontextprotocol/python-sdk/main/README.md, https://raw.githubusercontent.com/modelcontextprotocol/typescript-sdk/main/README.md), and those fetched documents are required and used to guide/drive agent prompts and implementation decisions.