The Agent Skills Directory

[OBFUSCATION]: The file demo/index.html contains a large Base64-encoded string in the encodedHtml constant. This string decodes to a full HTML document, effectively hiding the demo's source code, structure, and embedded URLs from static analysis.
[DYNAMIC_EXECUTION]: The demo runner in demo/index.html dynamically executes code by decoding the encodedHtml string and injecting it into a sandboxed iframe using the srcdoc attribute. It also converts Base64-encoded asset data into Blob objects and objectURLs at runtime.
[EXTERNAL_DOWNLOADS]: The demo environment references resources from well-known external services. This includes loading fonts and styles from Google Fonts (fonts.googleapis.com) and Fontshare (api.fontshare.com), and design libraries like GSAP, Three.js, and Iconify from their respective CDNs.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided "Product Brief" data (as seen in demo/PROMPT.md) to generate complex HTML and JavaScript. This represents an attack surface where instructions embedded in external data could influence agent output. Evidence chain for demo/PROMPT.md:
Ingestion points: The Product brief section in the prompt template.
Boundary markers: None explicitly defined to isolate the brief data.
Capability inventory: Generates full HTML/JS documents with script execution.
Sanitization: No sanitization or validation of the brief content is mentioned.

dither-laser-dark-mode