shadcn

Warn

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses dynamic context injection in SKILL.md via the !npx shadcn@latest info --json`` directive. This pattern executes the specified shell command automatically when the skill is loaded by the agent, providing an execution vector at load time to gather project metadata.
  • [REMOTE_CODE_EXECUTION]: The skill instructions frequently use npx shadcn@latest, pnpm dlx, and bunx to download and execute code from the npm registry at runtime. While this is the intended functionality for managing shadcn/ui components, it introduces reliance on the integrity of the remote package and the npm supply chain. Additionally, the skill enables the installation of components from third-party GitHub repositories and arbitrary URLs, which involves downloading and integrating external code into the local project.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of UI components and blocks from external sources, including public GitHub repositories (owner/repo/item) and arbitrary URLs (e.g., https://api.npoint.io/abc123 or https://ui.shadcn.com/init). The add and apply commands fetch remote registry metadata and source files for integration into the user's project.
  • [PROMPT_INJECTION]: The skill incorporates an attack surface for indirect prompt injection by instructing the agent to fetch and process content from documentation and example URLs provided by the shadcn docs command.
  • Ingestion points: Documentation and examples URLs returned by npx shadcn@latest docs (e.g., from raw.githubusercontent.com or ui.shadcn.com).
  • Boundary markers: None. The skill does not instruct the agent to use specific delimiters or to ignore potential instructions embedded in the fetched content.
  • Capability inventory: The skill allows shell command execution via Bash (restricted to shadcn commands) and project file modifications (writing .tsx, .css, and .json files).
  • Sanitization: None. There is no evidence of filtering or validation for the remote documentation content before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 25, 2026, 12:44 PM
Security Audit — agent-trust-hub — shadcn