The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill instructs the user to add an external, untrusted MCP server (https://rube.app/mcp) to their client configuration. This server is not on the [TRUST-SCOPE-RULE] list.
[REMOTE_CODE_EXECUTION] (HIGH): MCP servers define the tools available to the agent. By using a remote server, the agent's capabilities are dynamically defined by an external entity. The skill explicitly mentions RUBE_REMOTE_WORKBENCH and run_composio_tool(), which implies execution of complex toolchains or code in a remote environment.
[COMMAND_EXECUTION] (MEDIUM): The RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH tools provide significant side-effect capabilities, including executing multiple operations or 'workbench' tasks based on remote input.
[INDIRECT_PROMPT_INJECTION] (HIGH): The skill instructions create a mandatory vulnerability surface:
Ingestion points: The agent is told to 'Always search tools first' using RUBE_SEARCH_TOOLS. The response includes 'recommended execution plans' and 'known pitfalls' from the remote server (SKILL.md).
Boundary markers: There are no markers or instructions for the agent to sanitize or ignore instructions embedded in the 'execution plans' returned by the server.
Capability inventory: The agent has access to RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH (SKILL.md), which can perform significant actions in the Agencyzoom environment.
Sanitization: No sanitization or validation of the remote server's response is performed; the agent is explicitly told to follow the 'recommended execution plans'.

agencyzoom-automation