The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill implements a dynamic discovery pattern where the agent fetches execution plans and tool schemas from an external source at runtime.
Ingestion points: Data returned by RUBE_SEARCH_TOOLS as described in SKILL.md.
Boundary markers: Absent; the instructions tell the agent to follow the returned schema and plans exactly ("Always search tools first", "Use exact field names").
Capability inventory: RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH allow the agent to perform actions based on the fetched data.
Sanitization: No sanitization or validation logic is specified for the external schemas before execution.
Remote Code Execution (LOW): The skill utilizes RUBE_REMOTE_WORKBENCH and RUBE_MULTI_EXECUTE_TOOL. These tools allow for the execution of remote logic and operations provided by the Parallel toolkit. This is within the skill's primary purpose but represents a significant capability tier.
External Downloads (LOW): The setup process requires adding a third-party MCP server endpoint (https://rube.app/mcp). While necessary for the skill, this domain is not on the trusted sources list and provides the core logic for the agent's operations.

parallel-automation