The Agent Skills Directory

Command Execution: The skill provides detailed instructions for using the sbx CLI to manage sandboxes. This includes commands for lifecycle management (run, create, stop, rm, prune), file operations (cp), and executing commands inside the sandboxed environment (exec). These commands are necessary for the skill's intended purpose of managing isolated microVMs.
Indirect Prompt Injection Consideration: The skill facilitates the creation of environments where AI agents interact with external data, such as local or cloned Git repositories. This creates a potential surface for indirect prompt injection if those repositories contain malicious instructions. The skill addresses this risk by emphasizing the use of the sbx architecture for isolation and providing guidance on using --clone for repository isolation.
Workspace Isolation Best Practices: The documentation provides clear guidance on choosing between bind mounts and --clone isolation. It also specifically warns against mounting sensitive files or credentials, recommending the use of secure credential stores instead. These patterns reflect security best practices for managing AI agent environments.
Isolated Privilege Management: The skill describes the use of sbx exec with flags like -u root to perform administrative tasks inside the sandbox. While this involves elevated privileges, the operations are confined within the isolated microVM, preventing impact on the host system.

docker-sandboxes-lifecycle