Claude Code Branch Memory Manager

Automatic branch-specific CLAUDE.md memory management for Claude Code. When you switch git branches, your Claude Code context switches with you.

The Problem

When working across multiple git branches, your CLAUDE.md file contains context specific to one branch. Switching branches means Claude Code loses that context, or worse, has the wrong context for your current work.

The Solution

Branch Memory Manager saves and loads branch-specific CLAUDE.md files automatically. Each branch gets its own memory, stored in .claude/memories/.

As a Claude Code plugin — memories switch automatically when you run git checkout or git switch inside Claude Code. No manual intervention needed.

As a CLI — manually save, load, and manage branch memories from your terminal.

Quick Start

Install as a Claude Code Plugin (Recommended)

Install from the plugin marketplace inside Claude Code:

/plugin marketplace add Davidcreador/claude-code-branch-memory-manager
/plugin install branch-memory@davidcreador

Or install the npm package and load it with --plugin-dir:

# Install the npm package in your project
npm install --save-dev claude-code-branch-memory-manager

# Or install globally
npm install -g claude-code-branch-memory-manager

Then load as a plugin:

claude --plugin-dir ./node_modules/claude-code-branch-memory-manager

Once installed, branch memory switching happens automatically:

1. You run git checkout feature/auth inside Claude Code
2. The plugin saves your current CLAUDE.md to .claude/memories/main.md
3. It loads .claude/memories/feature_auth.md into CLAUDE.md
4. Claude Code now has the right context for your feature/auth work

Use as a Standalone CLI

# Initialize in your repo
branch-memory init

# Save current context
branch-memory save "Implemented user auth"

# Switch branches (saves + loads automatically)
branch-memory switch feature/api

# List all saved memories
branch-memory list

# Check status
branch-memory status

Commands

How It Works

Storage

Branch memories are stored in .claude/memories/ within your repository:

your-repo/
├── CLAUDE.md                          # Active memory (read by Claude Code)
├── .claude/
│   └── memories/
│       ├── main.md                    # Memory for main branch
│       ├── feature_auth.md            # Memory for feature/auth
│       ├── feature_api.md             # Memory for feature/api
│       └── .backups/                  # Automatic backups
│           └── main.2025-03-18T...md

Plugin Integration

When installed as a Claude Code plugin, four hooks handle automatic switching. Each one injects the active branch memory into the live session via Claude Code's additionalContext mechanism (writing the file on disk alone does not update an already-loaded context):

• SessionStart — Loads the correct branch memory when Claude Code starts
• PostToolUse (Bash) — Detects git checkout / git switch commands and swaps memories
• UserPromptSubmit — Detects branch switches made outside Claude Code (e.g. another terminal) and swaps memories. Cheap no-op when the branch hasn't changed, so it does not bloat every prompt
• SessionEnd — Saves the active CLAUDE.md back to the current branch when the session ends

Fallback Behavior

When switching to a branch with no saved memory, the tool looks for a fallback in this order: main → master → develop. This is configurable.

Configuration

Create a .branch-memory.json in your repo root:

{
  "memoryDir": ".claude/memories",
  "memoryFileName": "CLAUDE.md",
  "autoSaveOnSwitch": true,
  "autoSaveOnSessionEnd": true,
  "watchExternalSwitch": true,
  "fallbackToDefault": true,
  "fallbackBranches": ["main", "master", "develop"]
}

Environment Variables

Plugin Skills

When installed as a Claude Code plugin, these skills are available as namespaced slash commands:

• /branch-memory:save — Save current context
• /branch-memory:load — Load a branch's context
• /branch-memory:status — Show memory status
• /branch-memory:list — List all memories

These are Agent Skills: you can invoke them directly with the slash command, and Claude may also use them automatically when their description matches what you're doing.

Tips

• Add .claude/memories/ to .gitignore if you don't want to share memories across the team
• Commit .claude/memories/ if you want shared branch context
• Run branch-memory clean periodically to remove memories for deleted branches
• Branch names are sanitized for filenames: feature/auth becomes feature_auth.md

Requirements

• Node.js 18+
• Git 2.0+

License

MIT

Author

David Pastor