The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the official kaggle CLI using subprocess.run in scripts such as submission_quota.py and upload_dataset.py. These calls use the secure list-based argument format (preventing shell injection), and input parameters like competition slugs are validated against regular expressions defined in runtime.py.
[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of competition data, kernel source code, and dataset files exclusively from Kaggle's official domains (kaggle.com, api.kaggle.com) using standard tools.
[DATA_EXFILTRATION]: The skill manages the KAGGLE_API_TOKEN secret appropriately, with explicit instructions in SKILL.md and code checks ensuring the token is never logged, printed, or echoed.
[SAFE]: Robust sanitization logic is implemented to handle untrusted data from Kaggle. This includes html_to_markdown in runtime.py to strip potentially malicious HTML and sanitize_cli_output in upload_dataset.py to strip ANSI escape sequences and control characters, preventing terminal manipulation or prompt injection from reflected external content.

nvidia-kaggle-skill