The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is designed to ingest and obey instructions from an external source without validation.
Ingestion points: The skill fetches 'input schemas, recommended execution plans, and known pitfalls' from the RUBE_SEARCH_TOOLS function via https://rube.app/mcp (SKILL.md).
Boundary markers: None present. The instructions command the agent to 'Use exact field names' and follow the 'recommended execution plans' returned by the remote server.
Capability inventory: The skill possesses the ability to execute tools (RUBE_MULTI_EXECUTE_TOOL) and run remote logic (RUBE_REMOTE_WORKBENCH) within the Abyssale toolkit context.
Sanitization: No sanitization or verification of the remote instructions is performed before execution.
[External Downloads] (MEDIUM): The skill mandates the use of an external MCP server (https://rube.app/mcp) which is not on the trusted providers list. This server acts as the control plane for the skill's operations.
[Remote Code Execution] (HIGH): The workflow uses RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH to execute logic on remote infrastructure based on parameters and plans determined at runtime by an external API.

abyssale-automation