The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !npx shadcn@latest info --json`` syntax in SKILL.md to automatically fetch project metadata when the skill is loaded. This provides necessary project context such as framework, Tailwind version, and import aliases.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external registries and documentation URLs, creating a surface for potentially malicious instructions embedded in third-party content.
Ingestion points: Registry items added via the CLI and content fetched from URLs returned by the npx shadcn@latest docs command.
Boundary markers: None; the skill relies on instructions for the agent to manually review and verify all added files against defined styling and composition rules.
Capability inventory: Shell command execution (shadcn CLI), file system writes, and network fetching.
Sanitization: Absent; the skill defines strict composition rules and instructs the agent to audit code using CLI flags before integration.
[EXTERNAL_DOWNLOADS]: The skill supports installing components and presets from remote URLs using the shadcn CLI.
Evidence: npx shadcn@latest add [URL] and --preset [URL] examples in cli.md.
Mitigation: The skill instructions mandate the use of audit tools like --dry-run, --diff, and --view to inspect code from external sources before applying changes to the project.

shadcn