roblox-vfx

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Several Python scripts utilize the subprocess module to perform various tasks:
  • scripts/audition.py and scripts/compose.py invoke other Python scripts within the toolkit.
  • references/corpus/tools/ utilities (contact.py, strip.py, measure.py, palette.py) execute ffmpeg and ffprobe for media processing.
  • The skill instructions guide the user to execute blender in headless mode to run mesh generation scripts like sim/mesh/make_meshes.py.
  • [DYNAMIC_EXECUTION]: The toolkit generates executable code in multiple ways:
  • scripts/emit_luau.py and scripts/emit_scenes.py generate Luau ModuleScripts (VFXKit.lua, VFXScenes.lua) from JSON definitions.
  • scripts/payload.py assembles Luau execution payloads by splicing files and performing placeholder substitutions.
  • SKILL.md instructs users to run generated payloads in Roblox Studio using loadstring() after fetching them from a local server.
  • The browser simulator in sim/index.html uses eval() to load the snap.js helper.
  • [EXTERNAL_DOWNLOADS]: The skill interacts with external domains for resources:
  • scripts/classify_lib.py and scripts/fetch_tex_hires.py download asset thumbnails from the official Roblox domain (thumbnails.roblox.com).
  • The simulator fetches the three.js library from cdnjs.cloudflare.com and web fonts from fonts.googleapis.com.
  • [INDIRECT_PROMPT_INJECTION]: The 'harvester' workflow processes VFX data (JSON recipes) extracted from external Roblox Studio packs. While the toolkit includes filename sanitization and structured parsing, the ingestion of untrusted external content that is then used to generate logic for an AI agent represents a potential (though limited) indirect prompt injection surface.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 10:21 PM