The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the agent in using the repl tool to execute Python commands via the host.agents and host.skills SDKs. These tools are used for administrative tasks such as creating profiles, updating system prompts, and managing skill visibility.
[COMMAND_EXECUTION]: Includes a specification for kernel.py and kernel.R sidecar scripts. These scripts are intended to provide reusable helper functions. The instructions detail a validation process that prevents the execution of arbitrary logic at the module level, restricting it to function definitions and literal assignments.
[EXTERNAL_DOWNLOADS]: Provides instructions for creating isolated conda environments and installing necessary packages using the manage_environments and manage_packages tools. This is used to support the specific requirements of custom agent profiles.
[SAFE]: The instructional hint regarding action concealment is identified as user interface guidance. The suggestion to 'not narrate' approval cards is intended to reduce output redundancy when the platform's native security dialogs are displayed to the user.

customize