mcp-client

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

• Third-party content exposure detected (high risk: 0.70). The skill’s runtime path mcp-client.py call/list/emit sends requests to an MCP server specified by the operating user (--url or --stdio) and then ingests the server’s JSON-RPC responses (including content[].text and tool descriptions/schemas) into the script output/markdown, so if that MCP server is outsider-authored it can inject free text into the agent’s LLM context via the returned text.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill can contact remote MCP servers at runtime (e.g., the HTTP example https://mcp.example.com/github) which may return prompts/resources that directly control agent behavior, and the stdio example runs "npx -y @modelcontextprotocol/server-github" which fetches and executes remote code—both are runtime external dependencies that can control prompts or execute code.