make-pixel-splats

Warn

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The rendering script render.mjs dynamically imports JavaScript modules from paths provided as command-line arguments.
  • Evidence: The line const { frames } = await import(pathToFileURL(resolve(planPath)).href); allows the skill to execute arbitrary code from any local file if the user or agent specifies an untrusted path.
  • [DATA_EXFILTRATION]: The local HTTP server implemented in render.mjs is vulnerable to path traversal due to insufficient input validation.
  • Evidence: The server decodes the URL pathname and joins it with projectRoot using path.join without verifying that the resulting path remains within the intended directory. This could allow an attacker or malicious web content to read sensitive files from the host system using ../ sequences.
  • [PRIVILEGE_ESCALATION]: The browser automation configuration disables the Chromium sandbox to facilitate GPU access.
  • Evidence: render.mjs includes the --disable-gpu-sandbox flag in the browser launch arguments. While common for hardware-accelerated rendering in headless environments, this reduces the security isolation of the browser process.
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of a vendor-scoped package to function.
  • Evidence: package.json lists @codepawlhq/crucible as a dependency. This is a vendor-owned resource authored by the skill's creator (codepawl) and is used as the core rendering engine.
  • [COMMAND_EXECUTION]: The skill instructions and logic involve executing multiple system commands.
  • Evidence: The setup requires npm install, and the rendering pipeline uses node to run the renderer and ffmpeg to assemble video files from generated frames.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 16, 2026, 02:30 AM