← ClaudeAtlas

messydatalisted

Use this skill when working with MessyData — a synthetic dirty data generator. Covers writing and validating YAML configs, using the CLI, and the Python API. Trigger on: "generate synthetic data", "messydata config", "create a dataset schema", "add anomalies", "fake data", "dirty data".
sodadata/messydata · ★ 37 · Data & Documents · score 69
Install: claude install-skill sodadata/messydata
# MessyData Skill MessyData generates realistic messy DataFrames from a YAML config. It produces structured data with configurable anomalies (missing values, duplicates, invalid categories, bad dates, outliers). ## Workflow Always follow this order: 1. Write or edit the YAML config 2. **Validate it** with the CLI before generating 3. Generate data ```bash # 1. Validate (fast — no generation, exits 0/1) uv run messydata validate my_config.yaml # 2. Generate to stdout uv run messydata generate my_config.yaml --rows 1000 --seed 42 # 3. Generate to file (format inferred from extension) uv run messydata generate my_config.yaml --rows 1000 --output data.csv uv run messydata generate my_config.yaml --rows 1000 --output data.parquet uv run messydata generate my_config.yaml --rows 1000 --output data.json uv run messydata generate my_config.yaml --rows 1000 --output data.jsonl ``` Always run `validate` after writing a config. Fix any errors before generating. --- ## YAML Config Structure ```yaml name: <string> # required — dataset identifier primary_key: <string> # optional, default: "id" records_per_primary_key: <distribution> # required — rows per PK group (continuous only) fields: [<field_spec>, ...] # required — column definitions anomalies: [<anomaly_spec>, ...] # optional — data quality injections ``` **Row count:** `--rows N` is approximate. Each PK group is sampled from `records_per_primary_key`; actual co