setup-design-system
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates on an attack surface that involves processing untrusted external data and code, such as AI-generated components and design tool exports.
- Ingestion points: External content is introduced via component source files (typically in
src/components/ui/) and design token definitions in JSON, YAML, or CSS formats. - Boundary markers: The skill provides explicit instructions to treat imported markup and AI-generated code as data, emphasizing that only the governed token spec and the skill's own rules are authoritative to prevent instruction-like comments from being treated as binding.
- Capability inventory: The skill employs several high-capability tools, including
write_file,create_file, andexecute_command(used for running linters, tests, and build processes), which could potentially be abused if a malicious injection overrides agent instructions. - Sanitization: The skill implements a robust 'GenUI governance pipeline' that requires automated token conformance checks, accessibility audits (axe-devtools), and code scanning to detect and reject dangerous patterns like
eval(),new Function(), or unvalidated dynamic styling.
Audit Metadata