connect

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow delegates to Composio MCP over session.mcp.url and executes actions in external apps based on the user’s instructions, so outsider-authored free text (the user’s request content) can drive tool execution paths without requiring a trusted, pre-selected item selection step.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill binds the agent to an external MCP server at session.mcp.url (used at runtime in the code snippet) which lets that remote endpoint route and execute tool actions for the agent, so the agent's behavior can be controlled by that external URL (session.mcp.url).