The Agent Skills Directory

[SAFE]: The skill implements a structured 4-phase exploration process (isolation, incremental, ablation, and refinement) using local Python scripts and well-known data science libraries.
[SAFE]: Data handling is performed locally via pandas and xgboost. The skill reads from user-provided file paths (CSV/Parquet) and writes results to a local output directory.
[SAFE]: No network operations, remote code execution patterns, or attempts to exfiltrate data were identified.
[SAFE]: Dependencies are standard and verifiable, including numpy, pandas, xgboost, scikit-learn, and scipy.
[SAFE]: No obfuscation techniques, privilege escalation attempts, or persistence mechanisms were found in the codebase.
[SAFE]: Prompt instructions in SKILL.md are focused on experiment logic and reporting format, with no attempts to bypass safety filters or override agent constraints.

auto-experiment