mcp-client

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md, the required runtime workflow for the “mcp-client” skill makes POST requests to a user-configured remote MCP server endpoint and ingests free text contained in the MCP responses (including SSE data: lines parsed as JSON and returned result content blocks from tools/call, prompts/get, and resources/read).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill makes runtime POSTs to a remote MCP endpoint (referenced as SERVER_URL) and explicitly fetches prompts/resources (e.g., prompts/get, resources/read) that can directly control agent instructions, so SERVER_URL is a required external dependency used at runtime.