Running Multiple Claude Code Sessions Without Conflicts

Ran into a problem today. Wanted to work on two features at the same time with Claude Code, but opening multiple sessions on the same directory caused file conflicts.

Here’s how I fixed it.

The problem

When you open multiple Claude Code sessions in the same directory, they can step on each other’s toes. One session edits a file, another session doesn’t know about it, chaos ensues.

Git worktree to the rescue

Git worktree lets you check out multiple branches in separate directories, all sharing the same Git history. Perfect for this.

# Create a new worktree for feature work
git worktree add ../project-feature-auth -b feature-auth

# Now you have:
# /project          → main branch
# /project-feature-auth → feature-auth branch

# Run Claude in each directory
cd ../project-feature-auth && claude

Each Claude session gets its own isolated file system. No conflicts.

Making it easier with a custom skill

Typing those commands every time is annoying. So I made a /worktree skill.

Created skills/worktree/SKILL.md:

---
name: worktree
description: Git worktree management for parallel Claude sessions
aliases: [wt, parallel]
visibility: global
---

# Usage

/worktree new feature-auth    # Create new worktree
/worktree list                # Show all worktrees
/worktree remove feature-auth # Clean up

Now I just run /worktree new feature-auth and Claude handles the rest.

Auto-showing worktree status on session start

Added a SessionStart hook so I always know what worktrees exist:

#!/bin/bash
# hooks/worktree-status.sh

if ! git rev-parse --is-inside-work-tree &> /dev/null; then
  exit 0
fi

echo "📂 Git Worktree Status"
echo "Current branch: $(git rev-parse --abbrev-ref HEAD)"

worktree_count=$(git worktree list | wc -l | tr -d ' ')
if [ "$worktree_count" -gt 1 ]; then
  echo "Active worktrees:"
  git worktree list
fi

Registered it in settings.json:

{
  "hooks": {
    "SessionStart": [{
      "matcher": "startup",
      "hooks": [{
        "type": "command",
        "command": "~/.claude/hooks/worktree-status.sh"
      }]
    }]
  }
}

Now every new session shows me what worktrees are active.

Bonus: Saving tokens with script-based skills

While doing this, I realized my /git-sync skill was using a ton of tokens. Claude was reading the SKILL.md, thinking about each step, calling tools 5-7 times…

Switched to a script-based approach instead:

Before (token heavy):

SKILL.md explains steps → Claude interprets → Multiple tool calls

After (token efficient):

SKILL.md says "run this script" → One Bash call → Done

Created git-sync.sh that does everything:

• Checks status
• Creates branch if on main
• Commits with auto-generated message
• Pushes
• Opens PR (if gh CLI available)

# Now /git-sync just runs:
~/.claude/skills/git-sync/git-sync.sh "feat: Add feature"

Went from ~3000 tokens per run to ~500. Huge improvement.

Workflow now

1. Start main session: claude
2. Need parallel work: /worktree new feature-x
3. Open new terminal, cd to worktree, run claude
4. Work independently in each session
5. When done: /git-sync in each worktree
6. Clean up: /worktree remove feature-x

Gotchas

• Each worktree needs its own npm install or dependency setup
• Don’t forget to push from worktrees before removing them
• The main repo and worktrees share Git history, so commits are visible everywhere

Files created

~/.claude/
├── hooks/
│   └── worktree-status.sh    # SessionStart hook
├── skills/
│   ├── worktree/
│   │   └── SKILL.md          # /worktree skill
│   └── git-sync/
│       ├── SKILL.md          # Simplified
│       └── git-sync.sh       # All logic here
└── settings.json             # Hook registration

Pretty happy with this setup. Parallel Claude sessions without the headache.