The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill is highly vulnerable to Indirect Prompt Injection. It ingests external, untrusted content from the Kubernetes cluster and possesses high-privilege capabilities.
Ingestion points: kubectl-mcp-server call get_pod_logs, kubectl-mcp-server resources (cluster status, pod events).
Boundary markers: Absent. There are no instructions to the agent to treat cluster data as untrusted or to ignore embedded commands.
Capability inventory: High-impact write operations including scale_deployment and install_helm_chart via kubectl-mcp-server call.
Sanitization: Absent. No mention of filtering or escaping output from the cluster before it enters the agent's context.
[COMMAND_EXECUTION] (HIGH): The skill provides a direct interface for the agent to execute arbitrary tools and commands on a connected system. The call and serve commands allow the agent to perform complex operations with side effects on production infrastructure.
[DATA_EXFILTRATION] (MEDIUM): The skill exposes sensitive information from kubeconfig and allows starting a network-accessible server via kubectl-mcp-server serve --transport streamable-http --port 8000, which could be used to expose cluster data externally.
[PERSISTENCE] (LOW): The skill instructs the user to modify shell configuration files (~/.bashrc, ~/.zshrc) to add aliases, which is a method of maintaining a persistent presence in the user's shell environment.

k8s-cli