The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated scanners have blacklisted the author's documentation domain (jeffallan.github.io). Additionally, the skill instructions include patterns for downloading and executing remote scripts from various sources. While some sources are well-known, others point to the flagged author domain.
[REMOTE_CODE_EXECUTION]: Instructions in references/service-mesh.md, references/multi-cluster.md, and references/gitops.md direct the user to pipe remote curl/wget outputs directly into bash or kubectl. Examples include Istio, Linkerd, and Submariner installation scripts.
[PRIVILEGE_ESCALATION]: The skill provides detailed instructions for bypassing standard security controls in references/troubleshooting.md and references/storage.md. This includes:
Creating privileged pods with securityContext.privileged: true.
Using kubectl debug node to gain host-level access via chroot /host.
Mounting sensitive host paths using hostPath volumes.
Defining ClusterRoles with broad permissions in references/custom-operators.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, specifically container logs and events via kubectl logs and kubectl describe.
Ingestion points: kubectl logs in references/troubleshooting.md.
Boundary markers: None identified; logs are streamed directly to the terminal/context.
Capability inventory: Full file system access via manifests, shell execution via kubectl exec, and cluster configuration management.
Sanitization: No evidence of log sanitization or escaping before the agent processes the content.
[CREDENTIALS_UNSAFE]: references/configuration.md and references/workloads.md contain illustrative but realistic credential patterns, including API keys (sk-123...), JWT secrets, and private SSH/TLS keys. While used as placeholders, these templates could lead to accidental exposure if users copy them verbatim.
[COMMAND_EXECUTION]: The skill extensively uses kubectl and helm commands to manage cluster state, which is a powerful capability that could be abused if the agent is influenced by malicious log data or manifests.

kubernetes-specialist