The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user requirements and generate high-privilege Kubernetes manifests and execution commands. This creates an attack surface where malicious or carefully crafted requirements could influence the agent to generate insecure configurations, such as pods with 'hostNetwork: true' or 'privileged: true' settings, which could compromise the underlying node.
[COMMAND_EXECUTION]: The skill guides the agent to use powerful command-line tools like 'kubectl' and 'helm' to modify the state of the cluster. This includes dangerous capabilities such as executing shell commands inside containers ('kubectl exec'), copying files to/from pods ('kubectl cp'), and creating debug sessions on host nodes ('kubectl debug node'), which require careful oversight to prevent unauthorized cluster access.
[REMOTE_CODE_EXECUTION]: Instructions within the reference files include fetching and executing remote scripts and manifests. This includes patterns where remote installation scripts from providers like Istio, Linkerd, and Submariner are piped directly to a shell ('curl | sh'). While these are well-known technology projects, the practice of executing remote content is a significant security vector.
[PRIVILEGE_ESCALATION]: The skill provides templates for defining and applying 'ClusterRole' and 'ClusterRoleBinding' resources. These resources grant administrative permissions across the entire cluster. If an agent is misled into applying an overly permissive role, it could lead to total cluster takeover.
[EXTERNAL_DOWNLOADS]: The skill references and downloads configurations from numerous external sources, including GitHub, 'istio.io', and 'linkerd.io'. Automated security scans identified the author's primary documentation link ('jeffallan.github.io') as being present on a malicious URL blacklist, suggesting the vendor's hosting environment may be untrusted.

kubernetes-specialist