What problem does it solve? Exploring scientific datasets before modeling often means ad-hoc scripts that leak identifiers, silently truncate data, or make unsupported claims. This Skill performs bounded, local-only exploratory analysis of authorized files with strict safety rules, producing aggregate reports without exposing raw values or making causal claims. ## Core Features & Use Cases - Bounded format inspection: Profile CSV/TSV/JSON with the standard library, and optionally inspect NumPy NPY/NPZ, HDF5 metadata, FASTA/FASTQ sequences, and PNG/JPEG/TIFF image metadata with pinned dependencies. - Data quality audits: Run missingness, group/split leakage, duplicate-row, outlier, and transformation-sensitivity analyses without modifying or imputing the raw data. - Rigorous report scaffolds: Generate Markdown EDA reports with tokenized identifiers, documented limitations, and reproducible commands. - Use Case: A researcher receives a redacted CSV of experimental measurements and needs to check missingness patterns, train/test split leakage by subject, and outlier sensitivity before any modeling—without exposing subject identifiers or altering the raw file. ## Quick Start Use the exploratory-data-analysis skill to profile the file data.csv inside the approved project root and produce a bounded EDA report.
