The Agent Skills Directory

[OBFUSCATION]: The file demo/index.html uses Base64 encoding to embed the primary demo HTML document within a loader script.
Evidence: const encodedHtml = "PCFET0NUWVBFIGh0bWw+..." in demo/index.html.
Analysis: Deobfuscation reveals a legitimate HTML template for the 'Kiwi Floral' design demo. The technique is used for packaging the demo content rather than concealing malicious intent.
[EXTERNAL_DOWNLOADS]: The skill and its demo components reference and download content from several external domains.
Images are hosted on a Supabase storage bucket (hoirqrkdgbmvpwutwuwj.supabase.co) and referenced via direct URLs in SKILL.md.
The demo in demo/index.html and demo/source.json utilizes well-known technology CDNs for runtime dependencies, including cdnjs.cloudflare.com (for GSAP), cdn.tailwindcss.com (for Tailwind CSS), and code.iconify.design (for icons).
[INDIRECT_PROMPT_INJECTION]: The skill's workflow creates an attack surface for indirect prompt injection by instructing the agent to process data from an external website.
Ingestion points: The agent is directed to visit and parse search results from https://www.aura.build/assets based on user-provided tags in SKILL.md.
Boundary markers: The instructions do not define specific delimiters for the external content.
Capability inventory: The agent's capabilities include selecting content and returning multiple URLs to the user based on external data.
Sanitization: There is no mention of explicit sanitization or validation of the content retrieved from the remote site.

aura-asset-images