axonaut-automation

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

• Third-party content exposure detected (high risk: 0.75). The required runtime workflow calls Rube MCP tools like RUBE_SEARCH_TOOLS/RUBE_MANAGE_CONNECTIONS, and the LLM ingests the returned tool schemas and execution plans (free-form text) from the external MCP server, which is an outsider source.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires connecting to and calling the MCP server at https://rube.app/mcp during runtime (via RUBE_SEARCH_TOOLS / RUBE_MANAGE_CONNECTIONS), which returns tool schemas and recommended execution plans that directly control agent prompts and execution, so this external endpoint is a runtime dependency that can control agent behavior.