The Agent Skills Directory

Dynamic Context Injection: The skill utilizes a dynamic command syntax (!npx shadcn@latest info) to inject project-specific configuration and component lists directly into the agent's context when the skill is loaded. This is used to adapt recommendations to the user's current environment.
Indirect Prompt Injection Surface: The skill instructions suggest fetching component documentation and code examples from URLs resolved by the CLI. This content is used to inform the agent's code generation, creating a potential vector where instructions embedded in external documentation could influence agent behavior. The skill mitigates this by advising users to review all added components and files.
Command Execution & Remote Installation: The skill relies heavily on the shadcn CLI to manage the project, including adding components from both official and community registries. This involves downloading and executing code within the user's project, which is the primary intended function of the tool.
Registry Secret Management: Documentation in mcp.md describes support for private registries using environment variable substitution for authorization tokens (e.g., ${MY_TOKEN}). This follows standard practices for secret management in development tools.

shadcn