The Agent Skills Directory

Isolated Command Execution: The skill allows for running arbitrary commands within an ephemeral Firecracker microVM. This is the primary function of the sandbox, providing a secure boundary between untrusted code and the host system.\n- Network Egress Policies: It details the use of network security policies, including SNI filtering and 'deny-all' configurations. These features are essential for restricting the network access of potentially unsafe code.\n- Trusted Resource Utilization: The skill integrates the official @vercel/sandbox SDK and links to verified documentation on vercel.com and GitHub. As these resources are provided by the vendor, they are considered safe for implementation.\n- Indirect Prompt Injection Surface: As the environment is designed for untrusted code, it constitutes a potential attack surface. The skill manages this through documented ingestion points (writeFiles, source initialization) and boundary markers (updateNetworkPolicy) as noted in SKILL.md.

vercel-sandbox