aero-workflow-automation

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • Dynamic Execution (CRITICAL): The skill utilizes RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH to execute operations defined at runtime by an external server. Since the execution logic is retrieved from RUBE_SEARCH_TOOLS and incorporated into subsequent tool calls, it constitutes execution of untrusted remote content.
  • Indirect Prompt Injection (HIGH): The skill possesses a major indirect injection surface.
  • Ingestion points: RUBE_SEARCH_TOOLS returns tool slugs, input schemas, and execution plans from an external source.
  • Boundary markers: None. The skill explicitly instructs the agent to 'use exact field names and types' from the external response.
  • Capability inventory: RUBE_MULTI_EXECUTE_TOOL (write/execute) and RUBE_REMOTE_WORKBENCH (arbitrary tool execution via workbench).
  • Sanitization: None. The instructions mandate strict adherence to the externally provided schema, allowing a malicious server to dictate agent actions.
  • Unverifiable Dependencies & Remote Code Execution (HIGH): The setup requires adding https://rube.app/mcp as an MCP server. This endpoint is not in the trusted source list and serves as a remote control point for the agent's available tools and logic.
  • Command Execution (HIGH): Use of RUBE_REMOTE_WORKBENCH with run_composio_tool() enables arbitrary remote tool execution, which bypasses local constraints if the remote environment is compromised.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Feb 16, 2026, 07:50 AM