The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides a command to download an installation script from the official cube-js GitHub repository (github.com/cube-js/cube). This is the standard method for installing the required vendor CLI tool.
[REMOTE_CODE_EXECUTION]: The installation instructions include piping a remote script to the shell (curl | sh). While this is a remote execution pattern, it is directed to the official repository of the skill author (cube-js) for the purpose of setup.
[COMMAND_EXECUTION]: The skill facilitates the execution of terminal commands through the cube CLI for managing sessions, dashboards, and multi-tenant security configurations.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for potential instruction injection by interpolating user or agent-provided identifiers into shell commands. 1. Ingestion points: Placeholders such as , , and are used in CLI command templates. 2. Boundary markers: No explicit delimiters or warnings to the agent to ignore embedded instructions in the data are present. 3. Capability inventory: Terminal command execution via the cube CLI. 4. Sanitization: The skill includes high-level instructions to the agent to avoid printing sensitive tokens and to verify row-level security models manually.

cube-embed