The Agent Skills Directory

[SAFE]: The skill provides defensive configurations for Docker and GitHub Actions, including dropping Linux capabilities, enforcing non-root execution, and using read-only filesystems to prevent container escape and privilege escalation.
[EXTERNAL_DOWNLOADS]: References official base images from Microsoft's Container Registry (mcr.microsoft.com). It correctly advises pinning these images to specific SHA256 digests to mitigate supply chain risks associated with mutable tags.
[CREDENTIALS_UNSAFE]: Strongly advocates against hardcoding credentials, providing instructions on using Docker secrets (--mount=type=secret) and OIDC federation for cloud deployments to ensure secrets are never stored in image layers or environment variables.
[COMMAND_EXECUTION]: Outlines standard build and deployment workflows using dotnet, npm, and docker tools. It includes specific hardening for these commands, such as using no-new-privileges and pids_limit to prevent resource exhaustion and process exploitation.
[PROMPT_INJECTION]: The skill includes a scoping directive to ensure it is only used for DevOps-related tasks and not for application logic, serving as a functional constraint rather than a security bypass mechanism.

devops