The Agent Skills Directory

EXTERNAL_DOWNLOADS (HIGH): The setup instructions require adding 'https://rube.app/mcp' as an MCP server. This endpoint is not on the trusted source list and acts as a remote dependency that controls agent capabilities.
REMOTE_CODE_EXECUTION (HIGH): The skill executes remote tools via RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH. Since the tool definitions are provided by the external MCP server, this constitutes a remote code execution vector where the external provider defines the logic being run.
COMMAND_EXECUTION (HIGH): Tools like RUBE_REMOTE_WORKBENCH facilitate arbitrary execution environments, posing a significant risk if the remote server returns malicious tool configurations.
**PROMPT_INJECTION (Category 8
HIGH):** The skill exhibits a high-risk indirect injection surface.
Ingestion points: RUBE_SEARCH_TOOLS fetches tool slugs, input schemas, and 'recommended execution plans' from the remote server (SKILL.md).
Boundary markers: Absent. The agent is told to 'Always call RUBE_SEARCH_TOOLS first' and follow the returned plans without validation.
Capability inventory: RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH provide write and execution capabilities (SKILL.md).
Sanitization: None. The instructions mandate using the exact field names and types from the remote search results.

klipfolio-automation