design-system
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from various design sources to generate executable code and AI agent instructions.
- Ingestion points: External design sources including Figma nodes via MCP,
DESIGN.mdfiles, brand guides, specification documents, and code diffs from shipped pages as described inSKILL.mdandreferences/source-selection.md. - Boundary markers: Absent. The instructions lack delimiters or specific warnings to ignore instructions that might be embedded within the design metadata or specification text.
- Capability inventory: The skill writes source code for UI libraries, creates and modifies AI skill files (
SKILL.md), and dispatches local sub-agents for surveying tasks. - Sanitization: Absent. There is no evidence of filtering or sanitization of text extracted from design sources before it is interpolated into generated code or instructions.
- [COMMAND_EXECUTION]: The
evals/README.mdfile contains shell command patterns (bash and python3) intended for local execution to run the trigger evaluation suite using internal repository scripts.
Audit Metadata