The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages Kaggle API credentials (username, legacy key, and scoped tokens). It provides tools to generate these keys and save them to ~/.kaggle/kaggle.json and .env files.
Evidence: modules/registration/scripts/setup_env.sh and modules/kllm/scripts/setup_env.sh create the kaggle.json file with user credentials.
[COMMAND_EXECUTION]: The skill makes extensive use of the kaggle command-line interface and the subprocess module to interact with the Kaggle platform, including downloading data, submitting predictions, and pushing notebooks.
Evidence: modules/badge-collector/scripts/utils.py contains the run_kaggle_cli function which wraps subprocess.run to execute Kaggle commands.
[PROMPT_INJECTION]: The competition reporting module scrapes user-generated content from Kaggle, such as competition problem statements and winner writeups, which could contain indirect prompt injection attacks.
Evidence: modules/comp-report/README.md describes a workflow that uses Playwright to extract content from competition overview and leaderboard pages, including solution descriptions from user-submitted writeups.
[COMMAND_EXECUTION]: The badge collector module includes a persistence mechanism for earning streak badges by generating a shell script and providing instructions for manual scheduling via cron or launchd.
Evidence: modules/badge-collector/scripts/phase_5_streaks.py generates daily_streak.sh and prints setup instructions for periodic execution.

kaggle