The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates using the --yolo flag (alias for --dangerously-skip-permissions) when running the cmd tool in headless mode. This bypasses all security permission prompts, allowing the sub-agent to execute arbitrary terminal commands and modify the filesystem without human review.
[COMMAND_EXECUTION]: The documented command pattern cmd -p "<prompt>" does not include instructions for sanitizing the prompt content. If the string contains unescaped double quotes or other shell-sensitive characters, it could lead to command injection on the host system.
[PROMPT_INJECTION]: The skill establishes a sub-agent workflow that ingests untrusted data (e.g., from GitHub issues or local files) into the sub-agent's prompt. Without boundary markers or input sanitization, and given the sub-agent's elevated permissions (--yolo), this creates a significant risk of indirect prompt injection.
[EXTERNAL_DOWNLOADS]: The skill documents commands such as cmd taste learn and cmd skills for downloading personalization data and extensions from GitHub repositories, which represents an external code ingress point.

command-code