The Agent Skills Directory

Installation Patterns in Examples: The documentation provides recipes for installing software using curl | bash and package managers. These are documented as standard methods for provisioning sandbox environments. The guide encourages best practices such as pinning remote scripts to specific versions or commit SHAs to prevent supply chain risks.
Command Execution Framework: The skill explains the usage of setup.install and setup.startup fields for running commands within the isolated sandbox container. The documentation includes guidance on selecting appropriate user contexts (e.g., using root for system setup and non-privileged users for application startup) and designing commands to be idempotent.
Credential Protection Mechanism: A significant portion of the guide describes the 'proxy-managed' credential system. This architecture is designed to prevent actual secrets from entering the sandbox environment by using sentinel values that are replaced by a host-side proxy during outbound requests, which enhances the security of sensitive data.
Network Access Controls: The documentation details how to implement egress filtering through permissions.network allow-lists. It instructs authors on how to discover and declare only the specific domains required for a kit's functionality, facilitating a least-privilege network security model.
Instructional Context Ingestion: The skill describes the mechanism by which agentInstructions.content is rendered into an agent's workspace. It notes the use of sentinel boundary markers () to manage this content, which is the primary method for providing the agent with domain-specific operating instructions.

kit-author