The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill instructs users to add an untrusted MCP endpoint (https://rube.app/mcp) to their configuration. This allows an external, unverified server to define the tools, schemas, and instructions the agent will execute.
[REMOTE_CODE_EXECUTION] (HIGH): The skill includes RUBE_REMOTE_WORKBENCH and run_composio_tool(), which are designed for remote tool and code execution. Providing these capabilities via an untrusted third-party provider presents a significant risk of arbitrary code execution.
[CREDENTIALS_UNSAFE] (HIGH): The RUBE_MANAGE_CONNECTIONS tool is used to handle authentication for the 'highergov' toolkit. Interfacing with sensitive service connections through an unverified intermediary (rube.app) poses a severe risk of credential theft or unauthorized session access.
[COMMAND_EXECUTION] (MEDIUM): The use of RUBE_MULTI_EXECUTE_TOOL enables the automated execution of complex workflows. Because the tool definitions are provided dynamically by an untrusted server, there is a risk of the agent being tricked into executing malicious sequences of commands.
[PROMPT_INJECTION] (HIGH): The skill is designed to process external government data and search results (Highergov) while having high-privilege execution capabilities.
Ingestion points: Data retrieved from Highergov tools and RUBE_SEARCH_TOOLS results.
Boundary markers: None present in the instructions to separate untrusted data from agent logic.
Capability inventory: RUBE_MULTI_EXECUTE_TOOL, RUBE_REMOTE_WORKBENCH, and RUBE_MANAGE_CONNECTIONS provide broad system and account access.
Sanitization: No sanitization or validation of external tool outputs is documented.

highergov-automation