The Agent Skills Directory

Indirect Prompt Injection (HIGH): The core workflow (Step 1 to Step 3) creates a high-severity injection surface. The agent is instructed to call RUBE_SEARCH_TOOLS and then use the resulting schemas to drive execution via RUBE_MULTI_EXECUTE_TOOL. Any malicious or poisoned data returned by the search tool would be followed by the agent.
Ingestion points: Data returned from RUBE_SEARCH_TOOLS (external server response).
Boundary markers: Absent; instructions explicitly command the agent to use 'exact field names and types from the search results.'
Capability inventory: RUBE_MULTI_EXECUTE_TOOL provides a direct execution path for tools defined by the external source.
Sanitization: Absent; there is no logic to validate or filter the remote tool schemas before use.
Remote Code Execution (HIGH): The skill requires the addition of an external, non-trusted MCP server (https://rube.app/mcp). This server defines the capabilities and logic available to the agent, effectively allowing remote control over the agent's available actions.
Data Exposure (MEDIUM): The use of a third-party gateway to manage 'Griptape' connections implies that sensitive task data and potentially authentication tokens are processed by an external service provider (rube.app), which is not listed as a trusted source.

griptape-automation