The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs high-impact cleanup operations that affect the entire local Minikube environment rather than just the skill-specific clusters.
Evidence: The script setup-clusters.sh executes minikube delete --all, which destroys all existing Minikube profiles on the user's system.
Evidence: Multiple scripts use pkill -f to terminate processes like kubectl port-forward and devspace dev globally by name.
[COMMAND_EXECUTION]: User-provided strings are interpolated directly into shell commands, presenting a command injection surface.
Evidence: In SKILL.md, the agent is instructed to run git worktree add /tmp/ark-worktree-<branchN-slug> <BRANCHN>, where <BRANCHN> is taken directly from the user's request.
Evidence: The instruction to run bash .claude/skills/multi-branch-cluster-setup/scripts/setup-clusters.sh <branch1> <branch2> ... passes raw user input as shell arguments.
[PROMPT_INJECTION]: The instructions explicitly command the agent to bypass human-in-the-loop safety checks for high-risk operations.
Evidence: SKILL.md contains the instruction: "If the user's message contains branch names... extract all of them directly — do NOT ask for confirmation."

multi-branch-cluster-setup