# Refactoring 12 Data Versioning

> Use when adding lightweight data versioning and dataset reproducibility practices.

- Skill: `majiayu000/refactoring-12-data-versioning` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/refactoring-12-data-versioning`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/refactoring-12-data-versioning/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/refactoring-12-data-versioning

---


# Refactoring 12: Data Versioning

## Goal

Make input data changes explicit and reproducible.

## Sequence

- Order: 12
- Previous: refactoring-11-ci-automation
- Next: none

## Workflow

- Define dataset sources, versions, and checksums.
  - Success: Each dataset has an identifiable source and version.
- Store metadata in a manifest (CSV/JSON/TOML).
  - Success: Manifest captures dataset metadata and checksums.
- Separate raw data from derived artifacts.
  - Success: Raw and derived data live in distinct locations.
- Record dataset version alongside experiment outputs.
  - Success: Outputs reference the dataset version used.
- Prefer lightweight tracking unless DVC or similar is already in use.
  - Success: Versioning stays minimal and non-disruptive.

## Guardrails

- Do not commit large datasets to git.
- Avoid tooling changes that block current workflows.
- Keep versioning easy to maintain.