The Agent Skills Directory

[OBFUSCATION]: The demo/index.html file stores the demo's HTML content in a Base64-encoded string, which is decoded at runtime. This hides the script logic and page structure from direct text-based analysis.\n- [DYNAMIC_EXECUTION]: The demo environment uses srcdoc to inject dynamically decoded HTML into an iframe and URL.createObjectURL to manage media assets. This represents runtime assembly and execution of code.\n- [REMOTE_CODE_EXECUTION]: The skill depends on external JavaScript libraries for animations and icons. These are sourced from well-known services including Cloudflare's GSAP library and the Iconify CDN.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external design references and Neuform platform data to generate new code. This creates an attack surface where instructions embedded in reference designs could influence the agent's output. Every Cat 8 finding MUST document all four:\n
Ingestion points: The skill accepts a reference design from demo/index.html or external Neuform URLs in demo/PROMPT.md.\n
Boundary markers: The instructions do not define explicit delimiters to separate reference data from the system prompt.\n
Capability inventory: The agent generates full HTML, CSS, and JavaScript documents.\n
Sanitization: No input validation or filtering of external content is specified in the skill instructions.

gooey-blob-system