The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill recommends installing an external MCP server from an untrusted GitHub repository (github.com/nicholasoxford/shadcn-mcp). Installing unvetted MCP servers allows external code to operate with the AI agent's permissions, presenting a significant supply chain risk.
[REMOTE_CODE_EXECUTION] (HIGH): The skill provides instructions to run npx shadcn@latest add for components hosted on various third-party registries (e.g., @reui, @animate-ui, @diceui). This allows for the download and execution of remote code from unverified sources directly into the user's project codebase.
[COMMAND_EXECUTION] (HIGH): Ready-to-run shell commands for installation are provided throughout the skill. An automated agent executing these instructions could compromise the system if a registry hosts malicious component code.
[PROMPT_INJECTION] (HIGH): (Category 8: Indirect Prompt Injection) The skill exposes a large attack surface by ingesting and displaying data from external registries.
Ingestion points: Data enters via mcp__shadcn__search_items_in_registries and mcp__shadcn__view_items_in_registries in SKILL.md.
Boundary markers: Absent; external registry descriptions and examples are directly presented to the user and processed by the agent without isolation.
Capability inventory: The skill has the capability to suggest and facilitate shell command execution (npx shadcn add).
Sanitization: Absent; the skill does not include steps to validate or sanitize the integrity or content of components fetched from the internet before installation.

shadcn-component-discovery