The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The script scripts/shadcn_add.py uses subprocess.run to execute external commands. While it avoids shell=True and uses a list for arguments, it still allows the agent to execute arbitrary CLI subcommands available through the shadcn tool.\n- EXTERNAL_DOWNLOADS (MEDIUM): The skill is designed to download and execute code from the npm registry via npx shadcn@latest add. This exposes the system to potential supply chain vulnerabilities if the registry or the package is compromised.\n- REMOTE_CODE_EXECUTION (MEDIUM): Executing code via npx with the @latest tag pulls and runs remote scripts that are not pinned to a specific version, representing a dynamic execution risk.\n- PROMPT_INJECTION (LOW): The file references/canvas-design-system.md contains a 'Final Verification' section that uses a simulated user feedback pattern ('User already said...') to coerce the AI into maintaining a specific quality standard or persona.

ui-design-system