claude-design-system-hooks

Pass

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install several Node.js packages under the @claude-design/ scope and to clone a repository from GitHub (BharathKumarSuresh/claude-design-system-hooks). These are typical procedures for using a developer tool.
  • [EXTERNAL_DOWNLOADS]: The skill mentions a plugin ecosystem and a marketplace that allows the installation of community-provided tools (e.g., @community/lottie-animator), which involves downloading and running third-party code.
  • [PROMPT_INJECTION]: The core functionality of the skill is to generate UI code based on natural language descriptions, which constitutes an indirect prompt injection surface.
  • Ingestion points: Natural language input provided via the prompt and description fields in CLI commands and programmatic API calls.
  • Boundary markers: No explicit delimiters or instruction-ignore markers are mentioned in the provided examples.
  • Capability inventory: The skill has the capability to write generated code directly to the local filesystem using save() methods.
  • Sanitization: No specific sanitization, validation, or sandboxing of the AI-generated code is documented before it is saved to disk.
  • [COMMAND_EXECUTION]: The skill utilizes a CLI tool named claude-design to perform operations such as configuring API keys, generating components, and extracting design tokens from files.
Audit Metadata
Risk Level
SAFE
Analyzed
May 16, 2026, 03:49 PM