# Git Split

> Split the current `git` commit into multiple focused, self-contained, easy to review commits.

- Skill: `tomeraberbach/git-split` (Agent Skill)
- Install (CLI): `npx skillmds add tomeraberbach/git-split`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tomeraberbach/git-split/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: TomerAberbach (https://skillmd.com/u/tomeraberbach)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/tomeraberbach/git-split

---


Split the current `git` commit into multiple focused, self-contained, easy to
review commits.

# Current commit

```!
git show
```

$ARGUMENTS

# Principles

- One concern per commit: e.g. bug fix, feature, refactor, or config change
- Each commit passes all checks: format, lint, typecheck, build, test
- Commit tests with the code they test
- Never mix refactors with behavioral changes
- Prefer thin vertical slices, one complete feature end-to-end, over horizontal
  layers
- Use every added API, abstraction, or stub within the same commit. Don't split
  so small that a commit introduces dead code

# Workflow

1. Read the diff above. If the commit is already small and self-contained, tell
   the user and stop
2. Identify logical groupings and order them from most foundational to most
   dependent. When a boundary is ambiguous, keep changes together
3. Infer verification commands from project files (e.g. `package.json`,
   `Makefile`, CI config). If there are none, ask the user
4. Present the plan as a numbered list:
   - Commit N: `<imperative-mood description>: <file or hunk list>`
   - One sentence justifying each split boundary
   - The verification commands you will run after each split
   - Ask for confirmation before proceeding
5. Run `git reset HEAD~1` so all changes return to the working tree
6. Stage and commit each group in order, running verification commands after
   each:
   ```sh
   git stash
   <verification commands>
   git stash pop
   ```
7. After all splits, run `git log --oneline -<commit count>` and show the
   resulting stack

# How to split

## By file

Stage specific files for the first commit, then commit:

```sh
git add <files for first commit>
git commit -m "<first commit description>"
```

## By hunk

When a single file contains changes that belong in different commits:

1. Edit the file to contain only the changes for the first commit, removing the
   hunks that belong later
2. Stage and commit it
3. Re-edit the file to restore the removed hunks before the next commit