The Agent Skills Directory

[SAFE]: The skill serves as a high-quality educational resource and toolset for the scikit-learn library. All provided Python scripts and markdown instructions demonstrate idiomatic and safe usage of the framework.
[INDIRECT_PROMPT_INJECTION]: The skill includes common patterns for loading datasets from external files (e.g., CSV). While standard for machine learning, this represents a potential surface for the ingestion of untrusted data if an attacker controls the source file.
Ingestion points: pd.read_csv('data.csv') used in SKILL.md and references/quick_reference.md.
Boundary markers: Not present for the content of data files.
Capability inventory: The skill allows the Bash tool and performs local file operations (writing .png plots and .pkl models).
Sanitization: Standard scikit-learn/pandas processing without additional filtering for malicious text content in data rows.
[DYNAMIC_EXECUTION]: Reference documentation (e.g., references/model_evaluation.md) describes model persistence using joblib and pickle. These are standard utilities for saving ML models but can execute arbitrary code if loading a malicious file from an untrusted source.

scikit-learn