The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection in SKILL.md (!npx shadcn@latest info --json`) to execute a shell command automatically when the skill is loaded. This is used to populate project-specific variables but represents a silent execution of code in the user's environment.
[EXTERNAL_DOWNLOADS]: The add command documented in SKILL.md and cli.md permits downloading components from community registries (e.g., @magicui/shimmer-button) and arbitrary third-party URLs (e.g., https://api.npoint.io/abc123). This bypasses standard package manager safety checks.
[REMOTE_CODE_EXECUTION]: Code downloaded via the add command is written directly to the project's source tree. Subsequent imports or executions of these files by the agent or developer result in the execution of remote code. The instructions attempt to mitigate this by requiring the agent to use --dry-run and --diff for auditing before installation.
[DATA_EXFILTRATION]: In mcp.md, the registry configuration supports custom headers with environment variables (Authorization: Bearer ${MY_TOKEN}). While this is a standard authentication mechanism, it indicates the skill is designed to handle and transmit sensitive environment-based tokens to external registry endpoints.

shadcn