The Agent Skills Directory

[SAFE]: The skill implements comprehensive input validation through a shared utility module (_common.py), enforcing strict limits on file size (32MB), row counts (20,000), and feature dimensions.- [DATA_EXFILTRATION]: The bundled scripts explicitly reject network URLs and symbolic links for both input and output operations, ensuring that processing remains strictly local.- [DYNAMIC_EXECUTION]: A lazy import of the numpy package is used for a single calculation in the validation report; it uses a static string and poses no security risk.- [DYNAMIC_EXECUTION]: Data serialization via NumPy is configured with allow_pickle=False by default, preventing potential unsafe deserialization attacks.- [INDIRECT_PROMPT_INJECTION]: The skill processes external CSV data but utilizes rigorous type checking and structural validation, effectively minimizing the risk from untrusted data inputs.

scikit-survival