The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and fetches manifests and installation scripts from several well-known infrastructure and CNCF projects.
Evidence: references/gitops.md includes commands to install ArgoCD and Flux manifests; references/service-mesh.md downloads installation scripts from official Istio and Linkerd domains; references/multi-cluster.md fetches the clusterctl binary from Kubernetes Sigs and installation scripts from Submariner.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools to manage and interact with cloud infrastructure.
Evidence: Numerous commands using kubectl and helm are present throughout all reference files for deploying workloads, modifying configurations, and inspecting cluster state.
[PRIVILEGE_ESCALATION]: The skill includes instructions for performing tasks that require elevated system or cluster permissions.
Evidence: references/multi-cluster.md demonstrates using sudo for binary installation; references/troubleshooting.md provides patterns for debugging nodes using privileged containers and chroot /host.
[CREDENTIALS_UNSAFE]: The documentation contains examples of sensitive data handling, including dummy credentials and methods to retrieve secrets.
Evidence: references/configuration.md includes example Secret manifests with placeholder keys like sk-1234567890abcdef and generic passwords; references/troubleshooting.md provides commands for extracting and base64-decoding credentials from the cluster.
[METADATA_POISONING]: There is a discrepancy in the metadata regarding the author's identity.
Evidence: The author field in the SKILL.md frontmatter points to a GitHub profile (Jeffallan) that does not match the vendor context (alismx) associated with the skill.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to instructions embedded in external data retrieved from the cluster.
Ingestion points: The agent is instructed to read kubectl logs, kubectl describe output, and user-provided YAML manifests.
Boundary markers: The instructions do not define explicit delimiters or warnings to ignore instructions embedded in cluster data.
Capability inventory: The skill utilizes powerful tools (kubectl, helm) capable of modifying cluster state, executing code in containers, and managing network access.
Sanitization: No explicit sanitization or validation of retrieved cluster data is implemented in the provided instructions.

kubernetes-specialist