The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent on how to create 'command' type MCP servers using the uip CLI. This functionality allows the execution of arbitrary shell commands and arguments (e.g., via npx or uvx) on the host system. While this is a documented feature of the UiPath Agent Gateway, it represents a high-privilege capability that could be exploited if malicious commands are provided.
[REMOTE_CODE_EXECUTION]: The skill describes using uip agenthub mcp refresh-tools to start Orchestrator jobs that execute 'coded' MCP servers. These servers are Python packages published to a UiPath Orchestrator folder. As these resources originate from a trusted vendor platform, this remote execution pattern is considered part of the intended system functionality.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests data from external sources and uses this data to construct CLI commands.
Ingestion points: Data enters through uip agenthub mcp list, mcp get, mcp-tools candidates, and mcp-tools list which fetch resource names and metadata from Orchestrator.
Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore instructions embedded in the ingested data.
Capability inventory: The skill can create or update servers, bind resources, and trigger Orchestrator jobs via refresh-tools.
Sanitization: Absent; other than a regex check for slugs, there is no validation of the content of metadata or description fields before they are used in commands.
[DYNAMIC_EXECUTION]: The skill facilitates dynamic execution by enabling the registration and execution of external MCP servers (coded, command, remote, and swagger) at runtime. This allows for the dynamic addition of tools and capabilities to the agent's environment, which requires careful management of the sources and configurations used for these servers.

uipath-mcp-servers