The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted user context from the 'Input Context' section (e.g., cluster names, IP addresses, and tool selections) and interpolating this data into high-privilege administrative command templates. * Ingestion points: User-provided context fields in SKILL.md. * Boundary markers: Absent; templates provided do not use delimiters to isolate untrusted data from instructions. * Capability inventory: Shell command execution via sudo, kubeadm, helm, and the aws CLI as specified in SKILL.md. * Sanitization: Absent; the instructions do not specify logic for filtering or escaping the interpolated input.
[COMMAND_EXECUTION]: The skill utilizes high-privilege commands such as sudo for modifying system configurations and the aws CLI for cluster state backups. While these are necessary for the skill's primary administrative purpose, they constitute a significant capability surface if misdirected by malicious input.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install various third-party configurations and utilities, such as the Cilium helm repository and the kube-hunter utility. These originate from well-known services and are acceptable within the scope of Kubernetes administration.
[SAFE]: Several reference files (e.g., architecture-patterns.md, code-organization.md) contain hundreds of sections of repetitive technical boilerplate text. While not inherently malicious, this documentation is deceptive regarding its technical depth and creates excessive noise in the agent's context window.

kubernetes