easy-git

Let your AI Agent handle Git for you — an Agent Skills standard skill

Auto commit + push · Auto-managed .gitignore · Full worktree flow · Plain-language UX · Safety guardrails

---

What it solves

Two of the most painful things when working with an AI Agent:

• The Agent changed 20 files. You want to go back to 10 minutes ago — you can't. Nobody saved a checkpoint for you, and reconstructing it from memory isn't realistic. The faster AI writes, the bigger the price of having no version control.
• Want Git as a safety net? Staging / branch / merge / rebase / reset with three modes mixed together… You're three tutorials deep and still stuck on the first concept: "I just wanted to save my progress."

easy-git sits between you and Git — you describe what you need; Git is handled for you.

Features

Capability	Trigger	What the Agent does for you
Auto commit + push	A piece of work is done	Splits into atomic semantic commits, writes Conventional Commits-style messages, syncs to remote
.gitignore auto-management	Repo lacks .gitignore / dirty files about to be staged	Creates / updates .gitignore; blocks dependency dirs / build artifacts / secrets / IDE configs
Worktree management	You say "add a new feature ..."	Opens a worktree + new branch automatically; when done, asks (a) merge & clean (b) open a PR (c) keep
Plain-language UX	Any git interaction	"Saved a chunk of progress" / "Started a separate line for this" / "Going back to that checkpoint", instead of throwing Git jargon at you
Safety guardrails	Dangerous action triggered	force push to main / reset --hard / committing secrets / git add -A etc. — paused and confirmed

Tech stack

• Skill standard: Agent Skills open standard (agentskills.io) — unrelated to MCP, works across 32+ AI Agent tools
• Commit convention: Conventional Commits 1.0.0
• Underneath: Git + git-worktree
• Compatible Agents: Claude Code · Codex CLI · Cursor · GitHub Copilot · Gemini CLI · Junie · Goose · Amp · TRAE · …

Quick start

Option 1: Let the Agent install it for you (recommended)

Open the AI Agent you're using and paste this:

Install easy-git: please install https://github.com/xz1220/easy-git into the current Agent's skills directory. Use Marketplace for Claude Code; use ~/.codex/skills/easy-git for Codex CLI. After install, verify SKILL.md is readable and remind me to restart or open a new session.

No extra config needed after install. The AI Agent will automatically invoke easy-git at the right moments.

Option 2: Claude Code Marketplace

In Claude Code:

/plugin marketplace add xz1220/easy-git
/plugin install easy-git

Option 3: Manual install into each Agent's skills directory

If you want to install it manually, clone this repo into the corresponding tool's skills directory:

git clone --single-branch --depth 1 https://github.com/xz1220/easy-git.git <skills-dir>/easy-git

Common directories:

Agent	Install path
Codex CLI	~/.codex/skills/easy-git
Claude Code	~/.claude/skills/easy-git
Cursor	~/.cursor/skills/easy-git
OpenCode	~/.config/opencode/skills/easy-git
Gemini CLI / Goose / Amp / TRAE	Use the Agent Skills directory configured by that tool

Example, Codex CLI:

git clone --single-branch --depth 1 https://github.com/xz1220/easy-git.git ~/.codex/skills/easy-git

Example, Claude Code local install (when Marketplace isn't available):

git clone --single-branch --depth 1 https://github.com/xz1220/easy-git.git ~/.claude/skills/easy-git

If your environment uses a third-party skills installer:

npx skills install xz1220/easy-git

Claude.ai / Claude API users register SKILL.md via their respective Skills UI or API flow.

What it actually looks like

Scenario 1: The Agent broke something, and you want to go back

Without easy-git:

You: Optimize the search feature for me
Agent: OK, changed 12 files…
You: Wait, this is worse than before. Can you restore the old code?
Agent: Sorry, I didn't save a checkpoint. You'll have to recall it file by file…

With easy-git:

You: Optimize the search feature for me
Agent: Changed 12 files. Saved a chunk and synced to GitHub.
You: Not right, go back to the previous version
Agent: Done — rolled back to that checkpoint.

Scenario 2: Want to add a feature without touching main, but you don't know worktrees

Without easy-git:

You: Add a user login feature
Agent: (changes things directly on the current branch) Changed 30 files
You: Wait, you should've opened a branch first
Agent: Sorry, let me reset and redo…

With easy-git:

You: Add a user login feature
Agent: This is a sizeable change. I'm starting a separate line `feat-user-login`
       and I've moved to ~/repos/my-app-feat-user-login. When it's done I'll ask
       you how you want to bring it back.

Repo structure

easy-git/
├── SKILL.md            # Main skill file (Agent Skills spec)
├── references/
│   ├── translation.md  # Translation glossary (sourced from Git's official giteveryday)
│   ├── commit-style.md # Conventional Commits spec + public examples
│   ├── worktree-flow.md # Worktree lifecycle + naming rules
│   └── hook-recovery.md # pre-commit hook failure recovery flow
├── scripts/            # Safety scripts (secret scan, large-file scan)
├── docs/prd.md         # Full requirements and design
├── README.md
└── LICENSE

Roadmap

• PRD v4 stable
• User-story-style README
• Create docs/banner.png
• Write SKILL.md (Agent Skills spec)
• references/translation.md (full pass over giteveryday)
• references/commit-style.md (Conventional Commits 1.0.0 + public examples)
• references/worktree-flow.md
• references/hook-recovery.md
• Test fixture (empty sandbox + dry-run command set)
• Detailed comparison with peers like netresearch/git-workflow-skill to confirm differentiation
• Publish to Agent Skills marketplace

See docs/prd.md for details.

License

MIT