The Agent Skills Directory

[DYNAMIC_EXECUTION]: The DefectDetectionModel class in SKILL.md uses torch.load() to load model weights. This function utilizes the pickle module for deserialization, which can execute arbitrary code if an untrusted or malicious model file is supplied via the model_path argument.
[EXTERNAL_DOWNLOADS]: The code references torchvision.models with the pretrained=True parameter (e.g., ResNet18 and Faster R-CNN). This triggers the download of model weights from official PyTorch infrastructure, which is a well-known and trusted service.
[SAFE]: The skill uses established libraries for construction quality control tasks, including OpenCV for edge detection and Pandas for generating inspection reports. The declared filesystem permission in claw.json is consistent with its intended purpose of reading images and writing Excel reports.

defect-detection-ai