codex-visual

Pass

Audited by Gen Agent Trust Hub on Nov 20, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: Several scripts, including codex_generate.py, breathe.py, and frames_to_anim.py, utilize subprocess.run to call legitimate external tools like the codex CLI and ffmpeg, or to execute internal self-test modules. This functionality is essential to the skill's primary purpose of asset generation and preview building.
  • [EXTERNAL_DOWNLOADS]: The embed.py utility downloads machine learning model weights from Hugging Face (huggingface.co). These downloads are verified against hardcoded SHA256 hashes defined in the script and are necessary for the skill's perceptual identity auditing features.
  • [DYNAMIC_EXECUTION]: The skill uses dynamic module loading via importlib.util and __import__ to facilitate local script interactions and version checks. An AST-detected eval() call was manually reviewed and confirmed as a false positive, referring to the standard torch.nn.Module.eval() method used to set neural networks to evaluation mode.
  • [DATA_EXPOSURE]: codex_generate.py implements a temporary CODEX_HOME management system that symlinks the user's auth.json to a isolated directory during execution. This practice is designed to prevent session interference and is a documented part of the tool's local operation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 20, 2026, 03:25 PM