marquee-loop
Warn
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: MEDIUMOBFUSCATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [OBFUSCATION]: The file
demo/index.htmlcontains a large Base64-encoded string assigned toencodedHtml. This string encodes a full HTML document including scripts and styles, which conceals the demo's implementation from plain-text analysis. - [DYNAMIC_EXECUTION]: The skill uses a script in
demo/index.htmlto decode theencodedHtmlstring at runtime usingatob()andTextDecoder(). The resulting HTML is then dynamically injected into a sandboxed iframe via thesrcdocproperty. - [INDIRECT_PROMPT_INJECTION]: The
demo/PROMPT.mdfile defines a workflow that ingests external data, such as product briefs and content anchors, to populate the generated UI. This represents an attack surface where instructions embedded in processed data could influence agent behavior, though the skill's capabilities are limited to UI generation. - [EXTERNAL_DOWNLOADS]: The demo configuration in
demo/source.jsonand the decoded HTML content reference external dependencies from well-known services, including Cloudflare (GSAP), Supabase (image storage), Tailwind (CSS), and Iconify. These are standard resources for web development.
Audit Metadata