The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python scripts (slice_errors.py and largest_misses.py) to process user-provided datasets (CSV, Parquet, JSON). These scripts use standard libraries like pandas and numpy for data manipulation. The command execution is confined to local data processing based on user-supplied file paths.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for installing standard data science libraries (pandas, numpy, pyarrow) via pip from official package registries. These are well-known, trusted resources.
[DATA_EXPOSURE]: The skill processes prediction tables which could contain sensitive proprietary model outputs; however, the data remains local to the user's environment, and there are no network exfiltration patterns detected.

model-interpretation