The Agent Skills Directory

[Isolated Code Execution]: The skill focuses on using Vercel Sandbox to run code in ephemeral Firecracker microVMs. This provides strong isolation between the host environment and untrusted code, which is an effective mitigation against many forms of remote code execution risks.
[Network Policy Management]: It includes examples for configuring outbound network policies using SNI filtering and CIDR rules. This allows developers to restrict the domains a sandbox can communicate with, reducing the risk of data exfiltration during the execution of untrusted scripts.
[Credential Brokering]: The guidance describes a pattern for injecting secrets via network policy transforms. This technique allows the sandbox to interact with external APIs without ever exposing the raw credentials to the code running inside the microVM.
[Vulnerability Detection]: The skill includes automation (chainTo) that detects the use of known-vulnerable libraries like vm2 or unsafe shell execution patterns in the agent's context and suggests transitioning to a secure sandboxed environment.

vercel-sandbox