mesh-gradient-dark-blue-clean
Warn
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The demo environment in
demo/index.htmlstores its application logic within a Base64-encoded string (encodedHtml). This string is decoded and loaded into an iframe'ssrcdocattribute at runtime. This technique of dynamic loading from a locally-defined string obscures the code's intent and logic from static analysis tools.- [PROMPT_INJECTION]: Thedemo/PROMPT.mdfile directs agents to emulate a design from an external URL (https://neuform.ai/pages/alert-interface/index.html). This creates a surface for indirect prompt injection. The ingestion point is the external reference; the prompt lacks boundary markers or sanitization instructions for the ingested content; the agent has code generation capabilities; and while the demo uses Content Security Policy (CSP), the prompt itself provides no specific safeguards against malicious instructions in the source design.- [EXTERNAL_DOWNLOADS]: The skill and its demo utilize external assets from established providers, including Tailwind CSS, Iconify, and Google Fonts. These resources are used for styling and rendering the design system components.
Audit Metadata