webgl-laser

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEOBFUSCATIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [OBFUSCATION]: The skill uses Base64 encoding in the demo/index.html file to store the demo's HTML content within a string (encodedHtml). This content is decoded at runtime to populate a sandboxed iframe. This is a common pattern for self-contained design demos.
  • [DYNAMIC_EXECUTION]: The skill performs dynamic execution in two contexts:
  • WebGL shaders are compiled at runtime from string literals in SKILL.md using the gl.compileShader and gl.linkProgram APIs.
  • The demo/index.html file decodes a Base64 string and injects it into an iframe using the srcdoc attribute. It also dynamically injects script URLs into the sandboxed environment.
  • [EXTERNAL_DOWNLOADS]: The demo environment references and fetches resources from several external providers:
  • Visual assets are loaded from Supabase storage (hoirqrkdgbmvpwutwuwj.supabase.co).
  • Typography and fonts are sourced from Google Fonts (fonts.googleapis.com, fonts.gstatic.com) and Fontshare (api.fontshare.com).
  • Runtime scripts are referenced from TailWind CSS and Iconify CDNs.
  • All identified external domains are well-known services and consistent with the vendor's documented infrastructure.
  • [INDIRECT_PROMPT_INJECTION]: The skill includes instructions to ingest brand colors from the host page's CSS variables (--brand-accent).
  • Ingestion points: The initWebGLLaser function in SKILL.md reads CSS variables using getComputedStyle.
  • Boundary markers: The skill includes clear scope definitions and visual targets (e.g., pointer-events: none) to restrict how the generated code interacts with the page.
  • Capability inventory: The skill's primary function is visual rendering on a canvas, with no file system write or sensitive network capabilities.
  • Sanitization: The code includes a hexToRgb01 utility that performs basic parsing and validation of the expected hex color format.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 01:32 AM
Security Audit — agent-trust-hub — webgl-laser