The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the terminal tool to interact with the Box CLI. It includes robust safeguards such as verifying the absolute path of the command runner, preventing the use of sudo, and forbidding modifications to the system PATH or global npm configurations.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install official packages from the @box npm scope and PyPI. These are well-known, trusted dependencies from the service provider (Box, Inc.) and do not pose a security risk as implemented.
[DATA_EXFILTRATION]: While the skill handles sensitive file data and OAuth credentials, it contains strict instructions to prevent data leakage. It explicitly forbids requesting client secrets in chat, writing them to configuration files, or logging sensitive information. It also guides the agent to use the narrowest possible OAuth scopes.
[PROMPT_INJECTION]: No malicious prompt injection patterns or attempts to bypass safety filters were detected. The skill's instructions are focused on operational logic, identity verification, and maintaining a secure interaction boundary.
[REMOTE_CODE_EXECUTION]: No patterns of executing arbitrary remote code (e.g., piping URLs to bash) were identified. All script execution is confined to the local installation of verified Box tools.

box