The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides templates using pickle and joblib for model serialization. These utilities are standard in ML workflows but pose a security risk if used to deserialize untrusted artifacts, as they can execute arbitrary code.\n- [INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface for indirect prompt injection via the ingestion of external datasets (Parquet/CSV) in references/pipeline-orchestration.md and references/feature-engineering.md. The skill includes capability for file writing and network operations (MLflow, WandB, GCS) but mitigates risk through provided data validation and schema checking templates (Great Expectations).\n- [EXTERNAL_DOWNLOADS]: The skill references the installation of standard, well-known libraries (e.g., scikit-learn, pandas, mlflow, google-cloud-storage) and interacts with established services (MLflow, Weights & Biases, GCP, AWS) required for ML pipeline orchestration.

ml-pipeline