The Agent Skills Directory

[OBFUSCATION]: The file demo/index.html contains a large Base64-encoded string within the encodedHtml variable. This string decodes to a complete HTML document including a Content Security Policy, asset management scripts, and a WebGL-based background implementation. While used for sandboxing the demo, the use of Base64 to hide executable content is a significant obfuscation pattern.
[EXTERNAL_DOWNLOADS]: The skill downloads and references assets from well-known technology providers and CDNs, including Google Fonts (fonts.googleapis.com), Fontshare (api.fontshare.com), Iconify (code.iconify.design), and Tailwind CSS (cdn.tailwindcss.com). These sources are generally trusted but represent external dependencies for the skill's runtime environment.
[INDIRECT_PROMPT_INJECTION]: In demo/PROMPT.md, the agent is instructed to use external design references from neuform.ai and the provided demo/index.html as high-fidelity targets for generating new UI. This ingestion of external/complex content as a behavioral reference creates a surface for potential indirect prompt injection attacks where instructions could be hidden in the reference materials.
Ingestion points: demo/PROMPT.md references external URLs and the provided demo HTML.
Boundary markers: Absent. The instructions do not caution the agent to disregard potential embedded commands in the reference content.
Capability inventory: UI generation (HTML/CSS/JS); no dangerous system, shell, or file-write capabilities were identified in the associated scripts.
Sanitization: None. The agent is directed to match the fidelity of the references directly, increasing the likelihood of processing malicious instructions if present.

solar-duotone-bold