make-pixel-splats

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The render.mjs script utilizes dynamic import() to load animation 'plans' based on command-line arguments provided by the user. This is a core feature that allows for extensible scene generation.
  • Evidence: render.mjs imports a path resolved from process.argv.
  • [EXTERNAL_DOWNLOADS]: The setup process requires the user to download and install several dependencies.
  • Evidence: SKILL.md and package.json reference @codepawlhq/crucible and playwright-core via npm.
  • Evidence: The Python-based compositor howto_compose.py is invoked using uv run with the pillow and numpy packages.
  • [COMMAND_EXECUTION]: The documentation provides specific shell commands for the user to execute for video processing.
  • Evidence: SKILL.md includes ffmpeg commands for assembling PNG frames into an MP4 film.
  • [INDIRECT_PROMPT_INJECTION]: The skill represents a platform for executing arbitrary scene-building logic provided by the user or external scripts. This creates a surface for potential code execution if the plans or builders are sourced from untrusted parties.
  • Ingestion points: Command-line arguments for plan paths in render.mjs; user-defined builders in medieval2.js and scenes.js.
  • Boundary markers: Not present; the framework executes imported modules directly.
  • Capability inventory: File system writes (writeFile, mkdir, save), file deletion (unlink), and process spawning via browser automation (chromium.launch).
  • Sanitization: Absent; the framework expects valid executable JavaScript/Python code as input.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 02:21 AM