blender-car-film

Warn

Audited by Gen Agent Trust Hub on Nov 1, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill implements a runtime execution environment to facilitate 'live' communication between the agent and Blender.
  • tools/bridge.py defines a socket listener on 127.0.0.1:9876 that receives Python code and executes it using exec(compile(code, "<live>", "exec")).
  • scripts/07_anim.py and scripts/10_fix.py utilize dynamic module loading via __import__('mathutils').
  • [COMMAND_EXECUTION]: Automated rendering and post-production workflows rely on the execution of external CLI tools.
  • tools/render_all.sh and tools/mix.sh execute ffmpeg with complex filter strings for video encoding and audio mixing.
  • tools/render_all.sh spawns background instances of blender.exe via shell commands.
  • tools/progress.ps1 executes PowerShell commands to monitor the file system for render progress.
  • [EXTERNAL_DOWNLOADS]: The skill fetches assets and generated content from well-known external services.
  • scripts/dl_ph.py downloads 3D models and textures from Poly Haven (api.polyhaven.com).
  • scripts/gen_sfx.py fetches AI-generated audio and music from ElevenLabs (api.elevenlabs.io).
  • [CREDENTIALS_UNSAFE]: The skill requires a sensitive API key for sound generation functionality.
  • scripts/gen_sfx.py accesses the ELEVENLABS_API_KEY environment variable directly to authenticate network requests.
  • [PRIVILEGE_ESCALATION]: The skill interacts with Windows kernel APIs to modify system behavior.
  • tools/stay_awake.ps1 utilizes P/Invoke to call SetThreadExecutionState from kernel32.dll, overriding system power management settings to prevent sleep.
  • [INDIRECT_PROMPT_INJECTION]: The 'live bridge' architecture introduces a vulnerability where the agent could be manipulated into executing malicious code through its automation scripts.
  • Ingestion points: The agent generates Python scripts in scripts/ which are then piped into the execution bridge.
  • Capability inventory: Includes full Python access within Blender, shell execution capabilities, and network access to external APIs.
  • Sanitization: There is no evidence of validation or sandboxing for the code processed by the bridge.py listener.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 1, 2026, 05:00 PM