The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill's primary workflow involves fetching tool schemas and execution plans from a remote service via RUBE_SEARCH_TOOLS. This untrusted data directly influences the agent's next steps and tool arguments, creating a large attack surface for indirect prompt injection where a malicious response could hijack agent behavior.
Ingestion points: Search results and tool schemas from https://rube.app/mcp.
Boundary markers: None (instructions explicitly state to follow returned schemas and execution plans).
Capability inventory: RUBE_MULTI_EXECUTE_TOOL, RUBE_REMOTE_WORKBENCH, and RUBE_MANAGE_CONNECTIONS.
Sanitization: None provided in the instruction set.
[Remote Code Execution] (HIGH): The RUBE_REMOTE_WORKBENCH and RUBE_MULTI_EXECUTE_TOOL capabilities allow for the execution of logic and tasks defined on a remote server. This constitutes remote execution of third-party provided code/actions.
[External Downloads] (MEDIUM): The skill requires the addition of an external, third-party MCP endpoint (https://rube.app/mcp). This server is not a known trusted source in the analyzer's list and acts as a gateway for sensitive authentication and tool execution.

fibery-automation