The Agent Skills Directory

EXTERNAL_DOWNLOADS (HIGH): The skill requires connecting to https://rube.app/mcp. This domain is not within the Trusted External Sources, and connecting the agent to an unverified third-party MCP endpoint grants that endpoint significant influence over agent behavior.
REMOTE_CODE_EXECUTION (HIGH): Tools such as RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH (specifically run_composio_tool()) enable the execution of remote logic. This effectively allows an external provider to run code or perform operations in the user's environment or connected accounts.
Indirect Prompt Injection (HIGH): The skill is highly vulnerable to poisoning via external content.
Ingestion points: RUBE_SEARCH_TOOLS returns tool slugs, schemas, and 'recommended execution plans' from a remote server.
Boundary markers: None are implemented; the skill instructs the agent to 'Always search tools first' and follow the returned metadata.
Capability inventory: The skill possesses extensive write/execute capabilities via the Quaderno toolkit (e.g., managing invoices, payments, or customer data).
Sanitization: No sanitization is present. If the remote server returns a malicious 'execution plan,' the agent is likely to obey it, leading to unauthorized data modification or account takeover.

quaderno-automation