The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the download of UI component definitions from the vendor's registry at https://codedvisuals.com/r/{name}.json via the shadcn CLI. This is standard functionality for distributed React component libraries.
[REMOTE_CODE_EXECUTION]: Installation of components using npx shadcn@latest add involves downloading and executing remote code within the project. Since these downloads originate from the verified vendor domain codedvisuals.com and are the primary purpose of the skill, this is classified as expected vendor behavior.
[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection as it documents components that interpolate external data into visual displays (e.g., title, steps, and prompt props in visuals like ai-chat.md or agent-flow.md).
Ingestion points: Data enters the agent's context through component props defined in the visuals/ directory.
Boundary markers: Explicit delimiters for untrusted input are not defined in the documentation.
Capability inventory: The skill allows the execution of npx shadcn for component installation.
Sanitization: Standard React rendering typically handles basic string escaping, though the skill instructions do not specify custom validation for interpolated data.

codedvisuals