The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes scripts that use subprocess.check_output to capture the current environment state (e.g., pip freeze, git rev-parse HEAD). This is a standard practice in machine learning for experiment reproducibility and does not pose a security risk in this context.
[EXTERNAL_DOWNLOADS]: The documentation and code snippets reference integration with well-known third-party experiment tracking services such as Weights & Biases (W&B), MLflow, and Neptune. These are well-known technology services used for logging training metrics and are considered safe.
[DATA_EXPOSURE]: The skill explicitly includes a security-enhancing 'PII removal' pipeline within its data preparation references to prevent sensitive information from entering training datasets.
[SAFE]: No prompt injection, obfuscation, unauthorized privilege escalation, or persistence mechanisms were found across the skill's instructions and reference files.

ai-model-training