worktrees

Workspaced Worktrees

A Claude Code plugin for parallel branch development in monorepos. Create isolated workspaces with auto-wired environments and generated scripts — so multiple agents (or humans) can work on different features simultaneously without conflicts.

Supports portless for stable named .localhost URLs, with automatic fallback to port-based assignment.

Install

/plugin marketplace add JoaquinCampo/workspaced-worktrees
/plugin install worktrees@workspaced-worktrees

When prompted, choose user scope to make it available across all projects.

Setup (per project)

Run from your monorepo root:

/worktrees:worktree feature/test

First run auto-detects your git repos and creates a template config at .claude/worktree.json:

{
  "repos": [
    {
      "name": "backend",
      "port": 8000,
      "install": "uv sync",
      "start": "uv run uvicorn app.main:app --port {port}",
      "env_files": [".env"],
      "env_patches": {},
      "cache_dirs": [".venv"]
    },
    {
      "name": "frontend",
      "port": 3000,
      "install": "pnpm install",
      "start": "pnpm dev --port {port}",
      "env_files": [".env"],
      "env_patches": {
        "NEXT_PUBLIC_API_URL": "http://localhost:{backend.port}"
      },
      "cache_dirs": []
    }
  ]
}

Fill in the install and start commands for your stack, then commit the file. Your whole team gets the config.

Config fields

Field	Description
name	Subdirectory name (must be a git repo)
port	Base port number. Auto-incremented to avoid collisions. Only used when portless is not installed
install	Command to install dependencies
start	Command to start the dev server. {port} is replaced with the assigned port
env_files	Environment files to copy into the worktree
env_patches	Key-value pairs to patch in env files. Values support {<repo_name>.url} (recommended) and {<repo_name>.port} placeholders
cache_dirs	Directories to clone before installing (e.g., .venv). Leave empty for package managers with global caches (pnpm, yarn berry)

Usage

/worktrees:worktree feature/my-feature

This creates:

../my-project-wt-feature-my-feature/
├── .claude/
│   └── rules/
│       └── worktree.md     # auto-loaded agent context
├── CLAUDE.md                # inherited from project root
├── start.sh                 # start all servers
├── stop.sh                  # stop all servers
├── cleanup.sh               # tear down everything
├── backend/                 # git worktree on feature/my-feature
└── frontend/                # git worktree on feature/my-feature

Then launch an agent:

cd ../my-project-wt-feature-my-feature && claude --dangerously-skip-permissions

The agent automatically knows its ports, how to start servers, and the workspace rules — all via the generated .claude/rules/worktree.md.

Portless integration

If portless is installed (npm i -g portless), workspaces automatically get stable named URLs instead of port numbers:

backend-feature-auth.localhost:1355      # instead of localhost:8001
frontend-feature-auth.localhost:1355     # instead of localhost:3001

No port assignment, offset counting, or conflict resolution needed. Cross-service env references (NEXT_PUBLIC_API_URL) resolve to portless URLs automatically.

If portless is not installed, the plugin falls back to port-based assignment — no changes needed.

Parallel work

Each workspace gets unique endpoints, so you can run multiple features simultaneously:

With portless:

backend-feature-auth.localhost:1355          # agent 1
frontend-feature-auth.localhost:1355
backend-feature-dashboard.localhost:1355     # agent 2
frontend-feature-dashboard.localhost:1355

Without portless:

your-project/                           # main checkout — :8000, :3000
../your-project-wt-feature-auth/        # agent 1 — :8001, :3001
../your-project-wt-feature-dashboard/   # agent 2 — :8002, :3002
../your-project-wt-fix-upload-bug/      # agent 3 — :8003, :3003

Generated scripts

Script	Purpose
./start.sh	Start all servers with assigned ports
./stop.sh	Stop all servers
./cleanup.sh	Stop servers, remove worktrees, delete workspace

What agents know automatically

The generated .claude/rules/worktree.md tells agents:

• Which branch they're on
• Service URLs (portless) or port assignments and start commands for each repo
• To use ./start.sh and ./stop.sh for server management
• No migrations — database is shared across worktrees
• Schema changes are a blocker — report back to the user
• Don't push unless explicitly asked

Cleanup

From the workspace:

./cleanup.sh

Or manually: