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 used for a sandboxed demo preview. This is a standard packaging format for the platform and the decoded content is benign, consisting of HTML, CSS, and interactive JavaScript for visual demonstration.
[EXTERNAL_DOWNLOADS]: The skill configuration in demo/source.json identifies several runtime dependencies sourced from well-known and trusted Content Delivery Networks (CDNs), including Cloudflare and Google. These include GSAP for animations, Three.js for 3D rendering, and Iconify for iconography. These are standard libraries for modern web design and are sourced from established providers.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data (product briefs) to generate executable HTML and JavaScript code, which introduces an indirect injection surface.
Ingestion points: User-provided brief content in demo/PROMPT.md is processed by the agent to guide the generation of the final design.
Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from accidentally following instructions that might be embedded within the user brief.
Capability inventory: The skill allows the agent to generate full index.html files with interactive JavaScript and access to external assets.
Sanitization: The skill lacks explicit sanitization or escaping instructions for handling content from the brief before it is interpolated into the generated code.
[DYNAMIC_EXECUTION]: The demo loader in demo/index.html uses URL.createObjectURL and postMessage to dynamically load and inject assets into a sandboxed iframe. This architecture is designed to render a self-contained preview while maintaining security boundaries via a strict Content Security Policy (CSP).

editorial-tech