The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses official base images from Microsoft Container Registry (MCR) and Google Container Registry (GCR) in its Dockerfile templates. These are recognized as well-known and trusted sources.
[REMOTE_CODE_EXECUTION]: The GitHub Actions deployment template utilizes official actions from the azure/ and actions/ organizations, which are maintained by Microsoft and GitHub respectively.
[DATA_EXFILTRATION]: The skill explicitly instructs the agent to avoid requesting or outputting secrets, such as tokens or keys. It provides comprehensive guidance on using Azure Key Vault and Managed Identity for secure, passwordless secret management.
[INDIRECT_PROMPT_INJECTION]: The AKS Automatic Readiness sub-skill processes external Kubernetes manifests, creating an injection surface. This is mitigated by the following evidence chain: Ingestion occurs in azure-kubernetes-automatic-readiness/SKILL.md via local manifest files; while explicit boundary markers are absent, the skill's capabilities are restricted to read-only assessment; sanitization is performed by requiring the agent to present all proposed changes as human-readable diffs for manual user authorization before any file operations occur.
[PRIVILEGE_ESCALATION]: The skill promotes the principle of least privilege by enforcing the use of non-root users (runAsNonRoot: true), disabling privilege escalation, and dropping all container capabilities in its deployment templates.

azure-kubernetes