The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The skill executes high-privilege commands using sudo to install binaries to /usr/local/bin and to modify the system /etc/hosts file. These operations are essential for the skill's primary purpose of local Kubernetes orchestration but represent a significant privilege level.
[EXTERNAL_DOWNLOADS] (LOW): Binaries are retrieved from storage.googleapis.com and github.com. These are trusted sources for Kubernetes-related tooling, resulting in a downgraded severity for the download activity itself.
[REMOTE_CODE_EXECUTION] (MEDIUM): The skill provides and executes local shell scripts (e.g., scripts/validate.sh) and complex multi-line examples via the allowed Bash tool. This involves executing provided logic that interacts with the system state.
[DYNAMIC_EXECUTION] (MEDIUM): The skill utilizes eval $(minikube docker-env), which dynamically reconfigures the current shell session's environment variables to point to the Minikube Docker daemon.
[INDIRECT_PROMPT_INJECTION] (LOW): The workflow ingests external data from the local file system, such as Kubernetes manifests and Dockerfiles, which could be manipulated by an attacker to influence the agent's behavior.
Ingestion points: Files in k8s/, ./frontend, and ./backend directories.
Boundary markers: Absent.
Capability inventory: Subprocess execution via Bash, kubectl, and docker tools.
Sanitization: Absent; the skill directly applies manifests and builds images from these paths.

minikube-setup