canai-prepare
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill includes boilerplate code that references libraries from well-known services such as Tailwind CSS (via cdn.tailwindcss.com), Alpine.js (via cdn.jsdelivr.net), and Lucide (via unpkg.com). These references are intended for local preview of generated HTML files.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided design assets, such as screenshots and mockups, to generate HTML code. This ingestion of untrusted data (ingestion point: image/mockup files) represents an attack surface. The skill uses navigation comments like 'Type / Short Label' as boundary markers and instructions focus on structural mapping. Capabilities include file system access via Bash, Read, Write, and Edit tools. No explicit sanitization logic for input designs is described, but the primary purpose is code generation rather than instruction execution.
Audit Metadata