The Agent Skills Directory

[DYNAMIC_EXECUTION]: The reference documentation in references/model_evaluation.md recommends and provides code for using pickle.load() and joblib.load() for model persistence. These methods are susceptible to unsafe deserialization attacks, where loading a maliciously crafted model file can lead to arbitrary code execution on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for reading and processing external data from CSV files (e.g., pd.read_csv('data.csv')) and user-provided arrays.
Ingestion points: Untrusted data enters the agent context via pandas read operations and script arguments in SKILL.md and scripts/classification_pipeline.py.
Boundary markers: None present in the code snippets to delimit data from instructions.
Capability inventory: The skill has capabilities to write files to disk (plots) and execute arbitrary Python code via the scikit-learn library.
Sanitization: No sanitization or validation of the ingested data structure is provided in the examples.
[EXTERNAL_DOWNLOADS]: The SKILL.md file contains instructions to download several Python packages from public registries.
Evidence: uv pip install scikit-learn matplotlib seaborn pandas numpy in SKILL.md.
Evidence: Mentions of imbalanced-learn, category-encoders, and umap-learn in reference files.
[COMMAND_EXECUTION]: The skill provides shell commands for environment setup and for running automated machine learning pipelines.
Evidence: python scripts/classification_pipeline.py and python scripts/clustering_analysis.py in SKILL.md.

scikit-learn