The Agent Skills Directory

[SAFE]: The skill enforces a secure security posture by explicitly prohibiting high-risk behaviors such as the use of sudo, OS-level package managers (apt, brew, yum), and remote script execution patterns like curl | bash.- [SAFE]: It directs the agent to use platform-managed tools (manage_packages, inspect_packages) for all environment modifications, preventing the use of unsafe in-cell commands like %pip or !pip which are forbidden.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data in the form of notebook runtime errors (e.g., ImportError). This ingestion surface is secured by restricting the agent to a limited set of capabilities (manage_packages, notebook_execute) and providing clear boundary markers, such as requiring user intervention for system-level dependencies.- [DYNAMIC_EXECUTION]: The skill manages runtime package installation for Python and R environments. It mitigates risk by requiring isolated environments for non-standard specifications and relying on trusted repositories like PyPI, CRAN, and Conda-forge.

env-management