data-transform — Claude Code Skill

Convert data between any format with a single command.

CSV · JSON · SQL · YAML · Markdown · TOML · TSV · NDJSON

---

What it does

This Claude Code skill gives Claude two powerful Python scripts and expert knowledge to handle any data conversion task — reliably, every time, without re-inventing the logic from scratch:

You ask Claude…	What happens
"Convert this CSV to JSON"	Runs transform.py data.csv --to json
"Turn my JSON into SQL INSERT statements"	Runs transform.py data.json --to sql --table users
"Flatten this nested JSON and export to CSV"	Runs transform.py api.json --to csv --flatten
"What columns and types are in my data file?"	Runs infer_schema.py data.csv --output summary
"Generate a CREATE TABLE for this CSV"	Runs infer_schema.py data.csv --output sql
"Make this a markdown table"	Runs transform.py data.csv --to markdown

No external dependencies. Pure Python 3.8+ stdlib. Optional pip install pyyaml for YAML input/output.

---

Format Support

From \ To	CSV	JSON	NDJSON	SQL	YAML	Markdown	TOML
CSV / TSV	✓	✓	✓	✓	✓	✓	✓
JSON	✓	—	✓	✓	✓	✓	✓
NDJSON	✓	✓	—	✓	✓	✓	—
YAML	✓	✓	✓	✓	—	✓	—

---

Installation

Option A — Install as a Claude Code plugin (recommended)

# Clone the repo
git clone https://github.com/bernhardbrugger/claude-skill-data-transform.git

# Register with Claude Code
claude plugin add ./claude-skill-data-transform

Option B — Manual install

Copy the plugin folder into your Claude Code plugins directory:

cp -r claude-skill-data-transform ~/.claude/plugins/data-transform

Option C — Use scripts standalone

The scripts work on their own without Claude:

python skills/data-transform/scripts/transform.py data.csv --to json
python skills/data-transform/scripts/infer_schema.py data.csv --output summary

---

Quick Examples

Basic conversions

# CSV → JSON
python transform.py data.csv --to json --pretty

# JSON → SQL (PostgreSQL)
python transform.py data.json --to sql --table users --dialect postgres

# CSV → Markdown table
python transform.py data.csv --to markdown

# NDJSON logs → CSV
python transform.py app.log --to csv

# Pipe from stdin
cat records.ndjson | python transform.py --from ndjson --to csv

Nested JSON

# Flatten nested objects before converting to CSV
python transform.py api_response.json --to csv --flatten

# Explode array field: one row per item
python transform.py orders.json --to csv --explode items

Schema inference

# Human-readable column summary
python infer_schema.py data.csv --output summary

# JSON Schema draft-07
python infer_schema.py data.json --output json-schema

# SQL CREATE TABLE
python infer_schema.py data.csv --output sql --table my_table

Large files

# Stream a big CSV without loading into memory
python transform.py big.csv --to ndjson --stream

# Custom null values for messy data
python transform.py export.csv --to sql --null-values "N/A,#N/A,unknown,-,–"

---

What Claude handles automatically

When this skill is active, Claude:

• Auto-detects input format from file extension (.csv, .json, .ndjson, .yaml, .tsv)
• Infers types — integers, floats, booleans, dates, and nulls from string values
• Handles encoding — auto-detects UTF-8, UTF-8 BOM, and Latin-1; accepts --encoding override
• Normalises nulls — empty cells, NULL, N/A, n/a, none, nil, NaN, -
• Picks the right SQL dialect — SQLite, MySQL, or PostgreSQL type mappings
• Escapes SQL values safely — no SQL injection through ' escaping
• Streams large files — CSV and NDJSON output without loading the full dataset

---

Requirements

• Python 3.8+ (no install needed — included in macOS, most Linux, and Windows 10+)
• PyYAML (optional) — pip install pyyaml for YAML input or high-fidelity YAML output

---

transform.py — Full Options

usage: transform.py [-h] --to FORMAT [--from FORMAT] [--table NAME]
                    [--dialect sqlite|mysql|postgres] [--batch N]
                    [--delimiter CHAR] [--out-delimiter CHAR] [--no-header]
                    [--null-values LIST] [--flatten] [--sep SEP]
                    [--explode FIELD] [--no-infer] [--stream] [--pretty]
                    [--encoding ENC] [-o FILE]
                    [input]

Key flags:
  --to           Output format: json ndjson csv tsv sql yaml markdown toml
  --table        SQL table name (default: data)
  --dialect      SQL dialect: sqlite mysql postgres (default: sqlite)
  --flatten      Recursively flatten nested JSON objects
  --sep          Separator for flattened key paths (default: __)
  --explode      Expand array field: one output row per element
  --null-values  Extra null strings, comma-separated
  --no-infer     Keep all values as strings
  --stream       Stream output without loading full file into memory
  --pretty       Pretty-print JSON output
  -o FILE        Write to file instead of stdout

infer_schema.py — Full Options

usage: infer_schema.py [-h] [--output json-schema|sql|summary]
                       [--table NAME] [--sample N] [--enum-threshold N]
                       input

Key flags:
  --output          json-schema (default), sql, or summary
  --table           Table name for SQL output
  --sample          Max rows to analyze (default: 1000)
  --enum-threshold  Max unique values to list as enum (default: 20)

---

License

MIT — free to use, modify, and redistribute.

---

Built as a Claude Code skill by @bernhardbrugger