code-3d-explainer

Pass

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes several external command-line tools to perform its core functions. It uses google-chrome (or Chromium) in headless mode to render frames, ffmpeg and ffprobe to mux video/audio and inspect media properties, and ImageMagick (magick) to generate contact sheets and perform frame comparisons. These tools are invoked with argument arrays, minimizing command injection risks.
  • [EXTERNAL_DOWNLOADS]: The skill depends on external libraries and tools downloaded at runtime. The index.html file loads the three.js library from the jsDelivr CDN. Additionally, the skill utilizes bunx to execute the hyperframes package and bun add to download a local copy of three.js via the vendor_three.sh script.
  • [INDIRECT_PROMPT_INJECTION]: The skill features a multi-step fact-checking workflow defined in references/critic-prompt.md. This workflow instructs a subagent to fetch and analyze external source URLs provided in a fact sheet. This design creates a vulnerability surface where a malicious website cited as a source could attempt to inject instructions into the subagent's context, potentially compromising the integrity of the fact-checking process.
  • [DYNAMIC_EXECUTION]: The skill involves dynamic code generation and execution. It programmatically generates 3D scene logic in cell.js, which is then executed within a browser environment via three.js. The capture.ts script further uses the Chrome DevTools Protocol (CDP) to evaluate JavaScript expressions within the browser tabs to control rendering and extract metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 14, 2026, 09:50 PM