メニュー
Comprehensive data cleaning — handles missing/null values (drop, fill-mean, fill-median, fill-constant), deduplication (exact and fuzzy), type coercion, whitespace trimming, and outlier detection in a single pass.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Validate tabular data against configurable business rules — range checks, cross-column constraints, referential integrity against lookup tables, and temporal consistency.
Load and parse CSV files into structured data frames with configurable encoding, delimiter, and date parsing options.
Export pandas DataFrames or processed datasets to well-formatted CSV files with configurable encoding, delimiter, quoting, and column ordering.
Aggregate tabular data by one or more dimensions with configurable metrics — sum, mean, count, min, max, median. Supports multi-level grouping and pivot table generation.
Comprehensive data cleaning toolkit — handles missing values, duplicate removal, type coercion, whitespace trimming, and outlier detection in a single pass over the dataset.
Identify and remove duplicate records from tabular datasets using exact-match, fuzzy-match, or composite-key-based deduplication strategies.
| name | data-cleaning |
| description | Comprehensive data cleaning — handles missing/null values (drop, fill-mean, fill-median, fill-constant), deduplication (exact and fuzzy), type coercion, whitespace trimming, and outlier detection in a single pass. |
| version | 1.0.0 |
All-in-one data quality tool performing multiple cleaning operations in a single pipeline pass. Handles missing value imputation (configurable per-column strategy), duplicate row detection and removal (exact-match and fuzzy-match), automatic type coercion (string→numeric, string→date), whitespace trimming, and statistical outlier flagging. Generates comprehensive data quality reports.
Merges functionality from: null_value_processor, data_deduplicator, data_cleaning_toolkit.
drop, fill-mean, fill-median, fill-constant, fill-forward, flag-only--input / -i: Path to input file (CSV, Parquet, Excel) (required)--output / -o: Output path for cleaned data--null-strategy: Default strategy for missing values (default: drop)--critical-columns: Columns where nulls always trigger row drop--column-config: JSON mapping columns to specific null strategies--dedup-mode: exact-all, exact-key, or fuzzy (default: exact-all)--dedup-keys: Columns for composite key matching--keep: Which duplicate to keep: first, last, none (default: first)--fuzzy-threshold: Similarity threshold 0-100 for fuzzy mode (default: 90)--outlier-method: zscore or iqr (default: zscore)--no-outlier: Skip outlier detection--audit-only: Only report without modifying dataquality_report.json: Null counts before/after, duplicates found/removed, type coercion results, outlier counts, row counts at each stagenull_report.json: Per-column null analysisduplicates_report.json: Duplicate group detailspython3 ./skills/data_cleaning/clean.py -i <file> -o <output>pandas, numpy, scipy, fuzzywuzzy