The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection pattern in SKILL.md to execute npx shadcn@latest info --json at load time. This command retrieves local project configuration (framework, aliases, Tailwind version) to inform the agent's context. The command is benign and consistent with the skill's purpose.\n- [REMOTE_CODE_EXECUTION]: The skill enables the agent to install source code from remote registries or arbitrary URLs (e.g., npoint.io) directly into the user's project using the shadcn CLI. While this allows ingestion of external code, the skill mitigates the risk by explicitly instructing the agent to use --dry-run, --view, and --diff flags to audit the code and by requiring a manual review of all added files before proceeding.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it retrieves and processes content from external registries (documentation and component source code) that could contain malicious instructions.\n
Ingestion points: External registry metadata, documentation, and source code fetched via npx shadcn docs and npx shadcn view (SKILL.md, cli.md).\n
Boundary markers: Absent; the instructions do not specify the use of delimiters when processing external tool output.\n
Capability inventory: Filesystem writes via shadcn add and arbitrary shell command execution via the Bash tool.\n
Sanitization: Relies on the user or agent to manually audit changes via CLI diffing tools before installation.\n- [EXTERNAL_DOWNLOADS]: The skill references and downloads data from ui.shadcn.com, which is a well-known service for UI components, and documentation suggests the ability to add components from arbitrary third-party URLs.

shadcn