The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system commands including curl, magick (ImageMagick), and shadcn@latest. It also uses agent-browser to control a headless browser for scraping.
[EXTERNAL_DOWNLOADS]: Fetches data, images, and code from user-provided URLs and third-party registries. Specifically, it suggests using third-party registries such as reui.io to add components to the project.
[REMOTE_CODE_EXECUTION]: The skill uses shadcn@latest add with the -o (overwrite) and -y (yes) flags to download and install components from remote registries. This allows remote content to overwrite local project files without manual verification.
[DYNAMIC_EXECUTION]: The skill uses agent-browser eval to execute arbitrary JavaScript code on remote, untrusted websites to extract computed styles and DOM information.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from the web and uses it to drive code generation and project configuration.
Ingestion points: External reference URLs and image files processed in capture.md via agent-browser and curl.
Boundary markers: Absent. The skill does not implement delimiters or instructions to ignore potential commands embedded in the scraped reference content.
Capability inventory: The skill can perform file writes to the project root (DESIGN.md) and component directories, and it can execute shell commands (shadcn, magick, agent-browser).
Sanitization: Absent. While it mentions stripping HTML tags for text copy, the results of agent-browser eval and pixel sampling are used directly to generate code and tokens.

pimp-my-shadcn