The Agent Skills Directory

Sandbox-Based Execution: The skill executes commands within an ephemeral microVM. This architecture provides strong isolation, ensuring that browser automation tasks do not affect the underlying host system.
Environment Provisioning: The skill installs the agent-browser utility during its initialization phase. This is an expected step for setting up the browser automation environment and occurs within the isolated sandbox.
Safe Command Invocation: Commands are passed to the sandbox using structured argument arrays. This is a secure way to handle command execution as it prevents shell-related injection vulnerabilities.

vercel-sandbox