The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents MLOps workflows that ingest and process external data sources (CSV, Parquet, and cloud storage) which are then used to drive automated decisions such as model performance evaluation and promotion to production.
Ingestion points: Multiple examples in EXAMPLES.md and SKILL.md show data being loaded from local files and cloud buckets (e.g., pd.read_csv, pd.read_parquet).
Boundary markers: The provided code templates do not include explicit boundary markers or instructions to the model to ignore potential control instructions embedded within the ingested data.
Capability inventory: The documented workflows involve file system write operations, network server execution (mlflow server), and deployment triggers via cloud APIs (AWS SageMaker, Azure ML, GCP Vertex AI).
Sanitization: No explicit data sanitization or schema validation logic is demonstrated in the examples to protect against adversarial payloads in the input data.
[EXTERNAL_DOWNLOADS]: The skill references the installation of standard machine learning and data processing packages from public registries.
Evidence: README.md and SKILL.md include instructions to install well-known libraries such as mlflow, scikit-learn, pandas, numpy, torch, and xgboost using standard package managers.
[COMMAND_EXECUTION]: The documentation provides shell command examples for infrastructure management and model serving.
Evidence: README.md contains bash commands for starting MLflow tracking servers and deploying models using Docker and Kubernetes, which are standard operational tasks for MLOps engineers.

mlops-workflows