The Agent Skills Directory

[COMMAND_EXECUTION]: The references/tool-builder.md file contains code examples that demonstrate handling a calculate tool by passing user or AI-provided strings directly to the Python eval() function. This is a highly dangerous pattern that permits arbitrary code execution on the host system if a malicious expression is provided. Copying this pattern into production code would create a critical vulnerability.
[REMOTE_CODE_EXECUTION]: The skill documentation highlights a built-in code_execution capability for agents, enabled via internal_tools().code_execution(True). This feature grants the agent the ability to execute code, which is a powerful function that requires strict sandboxing and security controls to prevent abuse or system compromise.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the inferencesh package from PyPI. It also suggests adding other skills via the belt-sh/cli and npx skills add commands. These actions involve downloading and executing code from external package registries, which should be verified for authenticity.

python-sdk