solar-duotone-bold
Warn
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: The file
demo/index.htmlcontains a large Base64-encoded string (encodedHtml) that hides the entire structure and logic of the reference design design demo.- [DYNAMIC_EXECUTION]: The skill dynamically decodes theencodedHtmlvariable usingatob()and injects it into an iframe'ssrcdocattribute, creating a runtime execution path that bypasses static analysis.- [EXTERNAL_DOWNLOADS]: The skill references and loads scripts and media from external sources, includingcdn.tailwindcss.com,code.iconify.design, and a Supabase storage bucket (hoirqrkdgbmvpwutwuwj.supabase.co).- [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of external design data. - Ingestion points:
demo/PROMPT.mddirects the agent to treat thedemo/index.htmlfile and external Neuform.ai URLs as high-fidelity references for visual and interaction design. - Boundary markers: Absent. No instructions are provided to the agent to ignore or sanitize potentially malicious instructions embedded in the reference design source.
- Capability inventory: The skill utilizes network fetching (
fetch) and dynamic code execution (srcdocand Base64 decoding). - Sanitization: Absent. The skill does not validate or sanitize the reference HTML content or fetched metadata before processing.
Audit Metadata