The Agent Skills Directory

Command and Dynamic Execution: The skill provides APIs such as sandbox.exec and interpreter.runCode for executing shell commands and scripts within a containerized sandbox environment. While these represent powerful execution capabilities, the instructions include specific safety measures, such as using argument lists (argv) instead of raw strings to avoid shell injection vulnerabilities, and utilizing timeouts for remote processes.
Data Isolation and Input Handling: The skill acts as a gateway for the agent to process data via sandbox outputs, which creates a potential surface for indirect prompt injection if external data is ingested. To mitigate data exposure risks, the skill explicitly advises against placing sensitive credentials or secrets within the sandbox environment, recommending instead that live credentials stay in the parent Worker using outbound handlers.
Resource References: The skill references official documentation and repositories hosted on Cloudflare's infrastructure. These resources are provided for SDK orientation, lifecycle management, and implementation examples, and align with the intended use of the vendor's official tools.

sandbox-next