AgentSkillsCN

refactoring-12-data-versioning

在添加轻量级数据版本控制和数据集可重复性实践时使用。

SKILL.md
--- frontmatter
name: refactoring-12-data-versioning
description: Use when adding lightweight data versioning and dataset reproducibility practices.

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.