research

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

• Third-party content exposure detected (high risk: 0.85). The required workflow calls the Tavily MCP tool (tools/call to https://mcp.tavily.com/mcp with tavily_research), which fetches public web content at runtime and returns it as readable text/structured content that is then parsed and emitted (and thus can be fed into the agent’s LLM context via the skill’s response).

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

• Potentially malicious external URL detected (high risk: 0.90). The script invokes npx -y mcp-remote https://mcp.tavily.com/mcp at runtime when no token is present, which downloads and executes remote npm package code (and also relies on the external API at https://mcp.tavily.com/mcp), so remote content is executed during runtime.