The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes components that ingest external data (e.g., pd.read_parquet and pd.read_csv) to drive training and feature engineering workflows.
Ingestion points: Data is loaded from paths specified in pipeline parameters or input datasets in SKILL.md and references/pipeline-orchestration.md.
Boundary markers: The skill explicitly recommends and provides templates for data validation checkpoints using great_expectations to verify schemas and distributions before processing.
Capability inventory: Scripts include model serialization and loading via pickle, joblib, and torch, as well as file operations for deployment.
Sanitization: Integrated validation logic reduces the risk of malicious data influencing agent behavior.
[DYNAMIC_EXECUTION]: The provided templates utilize pickle.load(), joblib.load(), and torch.load() for deserializing models and transformation pipelines.
Evidence: Found in references/feature-engineering.md (FeaturePipeline.load), references/pipeline-orchestration.md (evaluate_model), and references/training-pipelines.md (load_checkpoint).
Context: These are standard practices within MLOps for managing model artifacts; the skill's primary purpose is providing these infrastructure patterns for developer use.

ml-pipeline