The Agent Skills Directory

[OBFUSCATION]: The file demo/index.html contains a large Base64-encoded string assigned to the variable encodedHtml. This string decodes to a complete HTML document including internal scripts and styles. While utilized for sandboxing a demo preview, this encoding technique effectively obscures the code's structure and intent from static analysis tools.
Evidence: const encodedHtml = "PCFET0NUWVBFIGh0bWw+..." in demo/index.html.
[INDIRECT_PROMPT_INJECTION]: The prompts provided in demo/PROMPT.md instruct the AI agent to visit and recreate a design based on an external URL (https://neuform.ai/pages/kiwi-floral-artisan-botanical-design-1/index.html). This introduces an attack surface where a compromised or malicious external page could provide instructions that the agent might mistakenly follow while attempting to analyze the design.
Ingestion points: Direct instruction to visit and use the visual/interaction reference from the external Neuform design page.
Boundary markers: None present; the instructions do not include delimiters or warnings to ignore embedded commands in the reference content.
Capability inventory: The skill enables the generation of responsive HTML/CSS and utilizes multiple JavaScript libraries for animation and layout.
Sanitization: No sanitization or validation of the content retrieved from the external URL is described or implemented.
[EXTERNAL_DOWNLOADS]: The demo loader and metadata reference several external runtime libraries and assets. These dependencies are fetched from well-known and reputable services including Cloudflare (for GSAP), Google Fonts, and the Tailwind CDN.
Evidence: demo/source.json maps local runtime paths to original URLs on cdn.tailwindcss.com, cdnjs.cloudflare.com, and fonts.googleapis.com.

css-border-gradient