website-to-hyperframes

Pass

Audited by Gen Agent Trust Hub on Aug 18, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from captured websites to drive content generation and code creation, creating a surface for indirect prompt injection.\n
  • Ingestion points: The agent reads site headlines, descriptions, and GLSL shader code from files like extracted/visible-text.txt and extracted/shaders.json (documented in references/step-1-capture.md).\n
  • Boundary markers: No delimiters or instructions to ignore embedded commands within the captured data are specified.\n
  • Capability inventory: The skill spawns sub-agents to write HTML/JS code and executes shell commands via npx (documented in references/step-6-build.md and SKILL.md).\n
  • Sanitization: There is no evidence of filtering or escaping captured website content before it is interpolated into prompts or generated compositions.\n- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill utilizes external tools and fetches libraries at runtime.\n
  • Shell commands: Executes npx hyperframes for site capture, linting, validation, and rendering.\n
  • Remote libraries: Loads animation and font libraries from cdn.jsdelivr.net and fonts.googleapis.com (documented in references/techniques.md).\n- [DYNAMIC_EXECUTION]: Generates and executes dynamic content by creating HTML/JS compositions based on input from external websites and extracted design tokens.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 18, 2026, 04:10 PM
Security Audit — agent-trust-hub — website-to-hyperframes