Back to Skills

git-worktrees

jamesrochabrun
Updated Yesterday
22 views
1
1
1
View on GitHub
Metaai

About

This Claude Skill helps developers manage Git worktrees to enable parallel Claude Code development sessions. It provides scripts and workflows for creating, managing, and cleaning up worktrees, allowing engineers to work on multiple features simultaneously without conflicts. Use this skill when you need to run parallel Claude sessions or maintain isolated development environments for different tasks.

Documentation

Git Worktrees for Claude Code

Overview

Run multiple Claude Code sessions in parallel on different branches using Git worktrees. This skill provides simple scripts and workflows to set up, manage, and clean up worktrees, enabling true parallel development without conflicts.

Why Worktrees?

  • Parallel Development: Run multiple Claude Code instances simultaneously
  • Zero Conflicts: Each worktree has independent file state
  • Fast Context Switching: No need to stash/commit when switching tasks
  • Isolated Experiments: Try different approaches without affecting main work
  • Long-Running Tasks: Let Claude work in background while you continue in main

Quick Start

1. Create a New Worktree (Super Easy!)

Just run the interactive script:

scripts/create_worktree.sh

This will:

  • ✅ Prompt for feature name
  • ✅ Create a new branch
  • ✅ Set up the worktree
  • ✅ Open in VS Code / editor
  • ✅ Give you the Claude Code setup command

That's it! The script handles all complexity.


2. View All Worktrees

scripts/list_worktrees.sh

Shows a clean, formatted list of all your worktrees with their branches and status.


3. Clean Up Old Worktrees

scripts/cleanup_worktrees.sh

Interactive removal of merged or abandoned worktrees.


Core Workflow

Pattern 1: Parallel Feature Development

Scenario: You want Claude to build feature A while you work on feature B.

Steps:

  1. Create worktree for feature A:

    scripts/create_worktree.sh
    # Enter: feature-a
    # Script creates: ../repo-feature-a/
    
  2. Open Claude Code in the new worktree:

    • The script outputs the path
    • Open Claude Code and navigate to that directory
    • Run /init to orient Claude
  3. Give Claude the task:

    "Build the user authentication feature with OAuth support"
    
  4. Continue your work in main:

    • Your original directory is unchanged
    • No conflicts, no waiting
    • Claude works in parallel
  5. When both are done, merge:

    # Review Claude's work
    cd ../repo-feature-a
    git log
    
    # If good, merge back
    git checkout main
    git merge feature-a
    
    # Clean up
    scripts/cleanup_worktrees.sh
    

Pattern 2: Long-Running Refactor

Scenario: You want Claude to refactor a large module while you continue development.

Steps:

  1. Create refactor worktree:

    scripts/create_worktree.sh
    # Enter: refactor-auth-module
    
  2. Start Claude in refactor worktree:

    "Refactor the authentication module to use dependency injection"
    
  3. Continue your daily work in main:

    • No interruption
    • No merge conflicts
    • Check progress periodically
  4. Review and integrate when ready:

    cd ../repo-refactor-auth-module
    git log --oneline
    # Review changes
    
    # Merge when satisfied
    git checkout main
    git merge refactor-auth-module
    

Pattern 3: Multiple AI Agents (Advanced)

Scenario: You want 3 Claude instances working on different features simultaneously.

Steps:

  1. Create 3 worktrees:

    scripts/create_worktree.sh  # feature-api-endpoints
    scripts/create_worktree.sh  # feature-ui-dashboard
    scripts/create_worktree.sh  # feature-email-notifications
    
  2. Open 3 Claude Code sessions:

    • Session 1: ../repo-feature-api-endpoints/
    • Session 2: ../repo-feature-ui-dashboard/
    • Session 3: ../repo-feature-email-notifications/
  3. Assign tasks to each Claude:

    • Claude 1: "Build REST API endpoints for user management"
    • Claude 2: "Create admin dashboard UI"
    • Claude 3: "Implement email notification system"
  4. Monitor progress:

    scripts/list_worktrees.sh
    
  5. Merge features as they complete:

    # Feature by feature
    git checkout main
    git merge feature-api-endpoints
    git merge feature-ui-dashboard
    git merge feature-email-notifications
    
    # Clean up
    scripts/cleanup_worktrees.sh
    

Pattern 4: Hotfix While Development Continues

Scenario: Production bug needs immediate fix while Claude is working on a feature.

Steps:

  1. Claude is already working in a feature worktree

    • Let it continue, don't interrupt
  2. Create hotfix worktree from main:

    scripts/create_worktree.sh
    # Enter: hotfix-login-bug
    # Choose base branch: main
    
  3. Fix the bug yourself or with another Claude session:

    cd ../repo-hotfix-login-bug
    # Make fixes
    git add .
    git commit -m "Fix login redirect bug"
    
  4. Merge hotfix to main:

    git checkout main
    git merge hotfix-login-bug
    git push origin main
    
  5. Sync feature worktree with latest main:

    cd ../repo-feature-xyz
    scripts/sync_worktree.sh
    # Merges latest main into feature branch
    

Essential Scripts

scripts/create_worktree.sh

Interactive worktree creation with all the right defaults.

What it does:

  1. Prompts for feature name
  2. Validates git repo
  3. Creates branch from main (or specified base)
  4. Sets up worktree in parallel directory
  5. Outputs setup instructions

Usage:

scripts/create_worktree.sh

# Prompts:
# Feature name: my-feature
# Base branch (default: main):
#
# Creates: ../repo-my-feature/
# Branch: my-feature

Advanced usage:

# Create from specific branch
scripts/create_worktree.sh --base develop

# Specify location
scripts/create_worktree.sh --dir ~/worktrees/my-feature

scripts/list_worktrees.sh

Shows all active worktrees with status.

Output:

╔════════════════════════════════════════════════╗
║            Active Git Worktrees                ║
╚════════════════════════════════════════════════╝

📁 Main Worktree
   Path: /home/user/project
   Branch: main
   Status: clean

📁 Feature Worktrees
   Path: /home/user/project-feature-api
   Branch: feature-api
   Status: 2 uncommitted changes

   Path: /home/user/project-refactor
   Branch: refactor-auth
   Status: clean

scripts/cleanup_worktrees.sh

Interactive cleanup of old worktrees.

Features:

  • Lists all worktrees with merge status
  • Identifies merged branches (safe to remove)
  • Prompts for confirmation
  • Safely removes worktree and branch

Usage:

scripts/cleanup_worktrees.sh

# Output:
# Found 3 worktrees
#
# 1. feature-api (merged to main) - Safe to remove
# 2. feature-dashboard (not merged) - Keep
# 3. old-experiment (not merged, 30 days old) - Consider removing
#
# Which worktrees to remove? (1,3): 1

scripts/sync_worktree.sh

Keep worktree up-to-date with main branch.

What it does:

  1. Fetches latest from remote
  2. Merges main into current worktree branch
  3. Handles conflicts with guidance

Usage:

# From within a worktree
cd ../repo-feature-api
scripts/sync_worktree.sh

# Or specify worktree
scripts/sync_worktree.sh ../repo-feature-api

Claude Code Integration

Important: Run /init in Each Worktree

When you open a new Claude Code session in a worktree, always run /init first:

/init

This ensures Claude:

  • Understands the codebase structure
  • Has proper context
  • Can navigate files correctly

Recommended Claude Code Workflow

  1. Create worktree (using script)
  2. Open Claude Code in worktree directory
  3. Run /init to orient Claude
  4. Give task to Claude
  5. Monitor via git log or file watching
  6. Review when complete
  7. Merge if satisfied
  8. Cleanup worktree

Best Practices

Naming Conventions

Good names:

  • feature-user-auth
  • refactor-api-layer
  • hotfix-login-bug
  • experiment-new-db

Bad names:

  • test ❌ (too vague)
  • wt1 ❌ (meaningless)
  • fixes ❌ (unclear)

Recommended format:

<type>-<short-description>

Types:
- feature-*
- refactor-*
- hotfix-*
- experiment-*
- review-*

Directory Structure

Recommended layout:

~/projects/
  └── myapp/              # Main worktree (main branch)
      ├── myapp-feature-api/      # Feature worktree
      ├── myapp-refactor-auth/    # Refactor worktree
      └── myapp-hotfix-bug/       # Hotfix worktree

The scripts default to creating worktrees as siblings to your main directory with a naming pattern: <repo>-<branch-name>.


When to Use Worktrees

✅ Use worktrees when:

  • Running multiple Claude Code sessions in parallel
  • Working on independent features simultaneously
  • Doing long-running refactors
  • Experimenting with major changes
  • Quick hotfixes during feature development
  • Code reviews without interrupting work

❌ Don't use worktrees for:

  • Simple branch switching (just use git checkout)
  • Minor tweaks (stash instead)
  • Single-task workflows
  • Very short-lived branches (< 1 hour)

Keeping Worktrees in Sync

Problem: Feature branches can get out of date with main.

Solution: Regularly sync worktrees:

# Option 1: Use the sync script
cd ../repo-feature-api
scripts/sync_worktree.sh

# Option 2: Manual sync
cd ../repo-feature-api
git fetch origin
git merge origin/main

Recommended frequency:

  • Daily for long-running features
  • After major merges to main
  • Before creating PRs

Managing Many Worktrees

Quick status of all worktrees:

scripts/list_worktrees.sh

Regular cleanup (weekly):

scripts/cleanup_worktrees.sh

Find stale worktrees:

# Worktrees with old commits
git worktree list | while read path branch commit; do
  cd "$path"
  last_commit=$(git log -1 --format=%cr)
  echo "$branch: Last commit $last_commit"
done

Troubleshooting

Issue: "fatal: invalid reference"

Cause: Branch name has invalid characters

Solution: Use alphanumeric + hyphens only:

# Bad
feature/my_feature  ❌

# Good
feature-my-feature  ✅

Issue: "cannot remove worktree, path is not a working tree"

Cause: Worktree directory was manually deleted

Solution: Prune worktree references:

git worktree prune

Issue: Claude seems "lost" in worktree

Cause: Didn't run /init after opening worktree

Solution: Always run /init when starting Claude in a new worktree:

/init

Issue: Merge conflicts when syncing worktree

Cause: Changes in worktree conflict with main

Solution 1: Resolve conflicts manually:

cd ../repo-feature-api
git fetch origin
git merge origin/main
# Fix conflicts in editor
git add .
git commit -m "Merge main and resolve conflicts"

Solution 2: If you want to discard worktree changes and restart:

cd ../repo-feature-api
git fetch origin
git reset --hard origin/main
# Start feature over

Issue: "Cannot lock ref" error

Cause: Branch exists in multiple worktrees

Solution: You can't have the same branch checked out in multiple worktrees. Create a new branch:

scripts/create_worktree.sh
# Use a different branch name

Issue: Worktree takes up too much disk space

Cause: Git worktrees share the .git directory, but files are duplicated

Solution 1: Clean up old worktrees:

scripts/cleanup_worktrees.sh

Solution 2: For large repos, use sparse-checkout:

git -C ../repo-feature-api sparse-checkout set src/ tests/

Advanced Usage

Custom Slash Command

Create a /worktree command for Claude Code. See references/slash_command_template.md for the complete setup.

Usage after setup:

/worktree feature-new-api

Claude will:

  1. Create the worktree
  2. Provide setup instructions
  3. Offer to continue the conversation in the new worktree

Sharing Worktrees in Team

Worktrees are local only, but you can share the workflow:

  1. Share branch, not worktree:

    cd ../repo-feature-api
    git push origin feature-api
    
    # Teammate creates their own worktree
    git worktree add ../repo-feature-api feature-api
    
  2. Document your worktree structure in team docs

  3. Use consistent naming across team


Worktree Performance Optimization

For large repos:

  1. Use sparse-checkout (only checkout files you need):

    git -C ../repo-feature-api sparse-checkout init
    git -C ../repo-feature-api sparse-checkout set src/ tests/
    
  2. Use worktree prune regularly:

    git worktree prune
    
  3. Limit concurrent worktrees to 3-5 for best performance


Common Workflows Summary

Quick Reference

TaskCommand
Create new worktreescripts/create_worktree.sh
List all worktreesscripts/list_worktrees.sh
Clean up worktreesscripts/cleanup_worktrees.sh
Sync with mainscripts/sync_worktree.sh
Manual creategit worktree add ../path branch-name
Manual listgit worktree list
Manual removegit worktree remove ../path
Prune referencesgit worktree prune

Real-World Examples

Example 1: Parallel API & UI Development

Setup:

# Create API worktree
scripts/create_worktree.sh
# Name: feature-api

# Create UI worktree
scripts/create_worktree.sh
# Name: feature-ui

Execution:

  • Claude 1 in ../repo-feature-api: "Build REST API for user management"
  • Claude 2 in ../repo-feature-ui: "Create React UI for user management"
  • You in main: Continue other work

Merge:

git checkout main
git merge feature-api
git merge feature-ui
scripts/cleanup_worktrees.sh

Example 2: Code Review Without Context Switching

Scenario: Need to review PR while Claude works on feature

Setup:

# Claude is working in feature worktree
# You need to review PR #123

scripts/create_worktree.sh
# Name: review-pr-123
# Base: pr-branch-name

Execution:

  • Review code in ../repo-review-pr-123
  • Claude continues work in ../repo-feature-xyz
  • No context switching needed

Cleanup:

# After review
scripts/cleanup_worktrees.sh
# Select review-pr-123

Example 3: Experimentation Without Risk

Scenario: Want to try a radical refactor without breaking current work

Setup:

scripts/create_worktree.sh
# Name: experiment-new-architecture

Execution:

  • Experiment freely in ../repo-experiment-new-architecture
  • Main worktree remains stable
  • If experiment fails, just delete worktree

Decision:

# If experiment succeeded
git checkout main
git merge experiment-new-architecture

# If experiment failed
scripts/cleanup_worktrees.sh
# Delete experiment worktree (no harm done)

Tips for Maximum Productivity

1. Name Worktrees Descriptively

  • You'll thank yourself later when you have 5 worktrees open

2. Always Run /init

  • First thing when opening Claude in a new worktree

3. Merge Frequently

  • Don't let worktrees diverge for weeks
  • Sync with main regularly

4. Clean Up Weekly

  • Run scripts/cleanup_worktrees.sh every Friday
  • Keep your workspace tidy

5. Use Scripts

  • Don't memorize git worktree commands
  • Scripts handle edge cases and validation

6. Monitor Progress

  • Use scripts/list_worktrees.sh to see what's active
  • Check git logs in worktrees periodically

7. Limit Concurrent Sessions

  • 3-5 worktrees max for best performance
  • More than that gets hard to manage

Resources

Scripts

  • create_worktree.sh - Interactive worktree creation
  • list_worktrees.sh - Show all active worktrees
  • cleanup_worktrees.sh - Remove old worktrees
  • sync_worktree.sh - Keep worktrees synchronized

References

  • worktree_commands.md - Complete Git worktree command reference
  • best_practices.md - Detailed best practices and patterns
  • slash_command_template.md - Create custom /worktree command

Summary

Git worktrees enable true parallel development with Claude Code:

Run multiple Claude sessions without conflicts ✅ Switch contexts instantly without stashing ✅ Experiment safely without breaking main ✅ Simple scripts handle all complexity ✅ Clean workflows for common scenarios

Get started:

scripts/create_worktree.sh

That's it! The scripts make worktrees simple and practical.

Quick Install

/plugin add https://github.com/jamesrochabrun/skills/tree/main/git-worktrees

Copy and paste this command in Claude Code to install this skill

GitHub 仓库

jamesrochabrun/skills
Path: git-worktrees

Related Skills

sglang

Meta

SGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.

View skill

evaluating-llms-harness

Testing

This Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.

View skill

llamaguard

Other

LlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.

View skill

langchain

Meta

LangChain is a framework for building LLM applications using agents, chains, and RAG pipelines. It supports multiple LLM providers, offers 500+ integrations, and includes features like tool calling and memory management. Use it for rapid prototyping and deploying production systems like chatbots, autonomous agents, and question-answering services.

View skill