The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/scaffold.py script creates Python source files by interpolating user-controlled strings—such as manifest CSV headers, manifest filenames, and the --from-pretrained CLI argument—directly into code templates. A lack of sanitization or escaping allows for arbitrary Python code to be injected into the generated project files, such as dataset.py or model.py.
[REMOTE_CODE_EXECUTION]: The skill's verification and testing suites (scripts/scaffold_challenge/verify.sh and tests/test_training_hygiene.sh) utilize importlib and exec_module to dynamically load and run the generated code for validation. This behavior results in the immediate execution of any code injected during the generation phase.
[COMMAND_EXECUTION]: The skill uses the Bash tool to orchestrate the execution of its internal scaffolding scripts and testing utilities, which involves running local Python and shell scripts.
[EXTERNAL_DOWNLOADS]: The generated requirements.txt and various documentation files (e.g., training_guide.md, mlops_guide.md) encourage the installation and use of external libraries including torch, numpy, monai, timm, torchvision, torchio, nibabel, pydicom, tifffile, tensorboard, wandb, mlflow, dvc, hydra-core, and omegaconf. While these are established tools in the machine learning domain, they represent external dependencies managed by the skill's output.

model-scaffold