updown-io-automation

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • EXTERNAL_DOWNLOADS (LOW): The skill instructions require the user to connect to an external MCP server at https://rube.app/mcp. This domain is not part of the predefined trusted source list. While it is a known integration platform (Composio), the reliance on an external endpoint for core functionality is noted.
  • PROMPT_INJECTION (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8c). It explicitly instructs the agent to 'Always search tools first' using RUBE_SEARCH_TOOLS and to follow the 'recommended execution plans' and 'input schemas' returned by the external server.
  • Ingestion points: Dynamic tool schemas and execution strategies fetched from https://rube.app/mcp via the RUBE_SEARCH_TOOLS tool.
  • Boundary markers: Absent. The skill does not provide delimiters or instructions to ignore malicious content within the fetched schemas.
  • Capability inventory: The agent has the capability to execute complex workflows and multi-step tool calls via RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH.
  • Sanitization: None. The instructions mandate strict compliance with the external search results ('Use exact field names and types from the search results').
  • COMMAND_EXECUTION (LOW): The skill uses RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH to perform operations on the Updown IO platform. While these are managed tool calls, they represent the execution of remote capabilities controlled by the MCP server's logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:32 PM