The Agent Skills Directory

[OBFUSCATION]: The file demo/index.html contains a very large Base64-encoded string (encodedHtml) that stores the entire HTML, CSS, and JavaScript payload for the design demo. Hiding executable logic in this manner is a common technique for bypassing static analysis.
[DYNAMIC_EXECUTION]: The demo environment in demo/index.html decodes the Base64 payload at runtime and injects it into a sandboxed iframe using the srcdoc attribute. It also dynamically generates Blob objects and Object URLs for media assets received via postMessage.
[EXTERNAL_DOWNLOADS]: The skill fetches several runtime dependencies from well-known services. These are identified in demo/source.json and loaded via relative paths in demo/index.html that resolve to external CDN sources:
GSAP library from cdnjs.cloudflare.com.
Iconify components from code.iconify.design.
Tailwind CSS from cdn.tailwindcss.com.
Google Fonts from fonts.googleapis.com and fonts.gstatic.com.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to generate complex HTML/CSS based on user prompts and design guidelines in SKILL.md.
Ingestion points: User instructions processed via demo/PROMPT.md.
Boundary markers: None identified to separate instructions from untrusted data.
Capability inventory: Generation and execution of HTML and JavaScript via a browser-based sandboxed environment.
Sanitization: While strict Content Security Policies (CSP) are applied to the demo iframe, the underlying mechanism for interpolating user-controlled data into generated templates lacks explicit sanitization, posing a risk of indirect prompt injection.

high-contrast-skeuomorphic-clean