hystruct-automation

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (MEDIUM): The skill requires the addition of an unverified external MCP server endpoint (https://rube.app/mcp) to the agent's configuration. This domain is not recognized as a trusted source.
  • [REMOTE_CODE_EXECUTION] (HIGH): The skill includes instructions for using RUBE_REMOTE_WORKBENCH with run_composio_tool(). This allows the remote service to execute complex operations and logic within the agent's environment, effectively granting remote execution capabilities to the unverified provider.
  • [COMMAND_EXECUTION] (HIGH): Implements RUBE_MULTI_EXECUTE_TOOL, which dynamically executes tools based on slugs and schemas returned by the external service. This provides a direct path for executing arbitrary actions defined by the remote endpoint.
  • [PROMPT_INJECTION] (HIGH): The skill is highly vulnerable to Indirect Prompt Injection (Category 8).
  • Ingestion points: Data entering the agent context via RUBE_SEARCH_TOOLS and tool outputs from the Hystruct integration.
  • Boundary markers: Absent. The skill provides no instructions to the agent to treat external tool definitions or outputs as untrusted.
  • Capability inventory: Significant write and execute capabilities including RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH.
  • Sanitization: Absent. The skill explicitly instructs the agent to follow the "recommended execution plans" provided by the external search results without validation.
  • [DATA_EXPOSURE] (MEDIUM): Use of RUBE_MANAGE_CONNECTIONS involves handling authentication links and potentially sensitive connection statuses (ACTIVE/INACTIVE) for external toolkits.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 01:25 PM