The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references official documentation and code examples from Cloudflare's developer portal (developers.cloudflare.com) and GitHub repositories (github.com/cloudflare). These references to a well-known service are handled safely according to established security guidelines.
[COMMAND_EXECUTION]: The skill documents the use of the sandbox.exec API for executing commands within isolated Linux containers. This is the core functionality of the service described and includes explicit instructions for handling process handles and avoiding implicit shell execution.
[INDIRECT_PROMPT_INJECTION]: The skill describes an environment for building applications that may process untrusted data (e.g., agent harnesses). It provides architectural guidance on secret management (using Worker environment variables rather than sandbox environment) and outbound traffic control to mitigate potential exfiltration risks inherent in agent development.

sandbox-next