The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill requires connection to an external MCP server located at https://rube.app/mcp. This domain is not within the trusted source list. The server acts as a remote dependency that defines the available tools and logic for the agent.
[INDIRECT PROMPT INJECTION] (HIGH): This skill exhibits a significant attack surface for indirect injection.
Ingestion points: Data enters via RUBE_SEARCH_TOOLS, which provides "tool slugs, input schemas, recommended execution plans, and known pitfalls" from the external server.
Boundary markers: None. There are no instructions to sanitize or validate the schemas/plans returned by the server.
Capability inventory: The skill possesses high-privilege capabilities including RUBE_MULTI_EXECUTE_TOOL (execution of actions on Raisely) and RUBE_REMOTE_WORKBENCH (remote code environment).
Sanitization: None detected. The agent is instructed to "Always call RUBE_SEARCH_TOOLS first" and use the exact results for execution.
[REMOTE_CODE_EXECUTION] (HIGH): The inclusion of RUBE_REMOTE_WORKBENCH with the function run_composio_tool() suggests the ability to execute code or complex logic in a remote environment managed by the untrusted rube.app endpoint.
[COMMAND_EXECUTION] (HIGH): Through RUBE_MULTI_EXECUTE_TOOL, the agent can be manipulated into executing arbitrary tool commands if the external schema provider returns malicious tool definitions.

raisely-automation