build-house-from-plan

Warn

Audited by Gen Agent Trust Hub on Nov 15, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script scripts/render_house.py includes a function load_film_script that uses importlib.util to dynamically load and execute Python modules. The path to the script is retrieved from the script field in the user-provided plan.json file. This pattern allows for the execution of arbitrary local Python code if a malicious plan file is supplied.
  • [DYNAMIC_EXECUTION]: The script scripts/fetch_assets.py uses the exec() function to process a configuration line extracted from the local house_core.py file to determine the asset library path.
  • [EXTERNAL_DOWNLOADS]: The skill downloads 3D assets and metadata from external sources to populate the Blender scene.
  • scripts/fetch_assets.py fetches textures and HDRI environment maps from Poly Haven and ambientCG.
  • scripts/fetch_objaverse.py and scripts/find_objaverse.py download 3D models and search metadata from the Objaverse dataset repository on Hugging Face.
  • [COMMAND_EXECUTION]: The skill instructions in SKILL.md require the execution of various shell commands via uv run and blender to run the included Python scripts for plan validation, model building, and rendering.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes complex, untrusted plan.json files and 2D drawing data. A malicious input could attempt to influence the agent's behavior or exploit scripts that utilize data from the plan (such as the dynamic script loading functionality).
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 15, 2026, 03:07 AM