amgr

Agent Manager CLI

Define your AI assistant configurations once, deploy them everywhere.

amgr lets you maintain a single repository of AI rules, commands, and settings, then selectively deploy them to any project — whether you're using Claude Code, Cursor, GitHub Copilot, or other AI tools.

The Problem

You have multiple projects:

~/Code/backend/     → needs coding rules, test commands, debugging skills
~/Code/frontend/    → needs coding rules, component patterns, different MCP servers
~/Code/blog/        → needs writing rules, no coding commands

Each AI tool (Claude Code, Cursor, Copilot) expects configurations in different formats and directories. Keeping them in sync manually is tedious and error-prone.

The Solution

┌─────────────────────────────────────────────────────────────────────────┐
│                        Your Rules Repository                            │
│  ~/Code/my-agents/                                                      │
│  ├── shared/           ← rules that apply everywhere                    │
│  │   └── rules/                                                         │
│  │       └── tone.md        (profiles: [development, writing])          │
│  │       └── testing.md     (profiles: [development])                   │
│  ├── development/      ← nested profile with sub-profiles               │
│  │   ├── _shared/      ← shared across frontend/backend                 │
│  │   ├── frontend/     ← sub-profile: development:frontend              │
│  │   └── backend/      ← sub-profile: development:backend               │
│  └── writing/          ← flat profile: style guides, templates          │
└─────────────────────────────────────────────────────────────────────────┘
                                    │
                                    │ amgr sync
                                    ▼
     ┌──────────────────────────────┬──────────────────────────────┐
     │                              │                              │
     ▼                              ▼                              ▼
┌─────────────┐             ┌─────────────┐             ┌─────────────┐
│  backend/   │             │  frontend/  │             │    blog/    │
│ .claude/    │             │ .claude/    │             │ .claude/    │
│ .cursor/    │             │ .cursor/    │             │ .cursor/    │
│             │             │             │             │             │
│ coding rules│             │ coding rules│             │ writing     │
│ test cmds   │             │ components  │             │ style guide │
│ debug skills│             │ debug skills│             │             │
└─────────────┘             └─────────────┘             └─────────────┘

One source of truth → Multiple projects → Multiple AI tools

Installation

npm install -g amgr

Quick Start

1. Create a rules repository (one-time setup)

mkdir ~/Code/my-agents && cd ~/Code/my-agents
amgr repo init --name "my-agents"
amgr repo add development --description "Coding and debugging"
amgr repo add writing --description "Documentation and content"

This creates a repository structure where you can add your rules, commands, and skills.

2. Set it as your global source

amgr source add ~/Code/my-agents --global

Now all your projects can use this repository.

3. Configure any project

cd ~/Code/my-project
amgr init                    # Interactive setup
amgr sync                    # Deploy configurations

That's it. Your AI tools now have the rules you defined, formatted correctly for each tool.

Commands

amgr or amgr sync

Synchronize agent configurations based on .amgr/config.json.

amgr sync
amgr sync --dry-run      # Preview changes without applying
amgr sync --verbose      # Show detailed output
amgr sync --replace      # Delete all tracked files before deploying

Options:

• -n, --dry-run - Show what would be done without making changes
• -v, --verbose - Enable verbose output
• -c, --config <path> - Use a custom config file path
• --replace - Delete all tracked files before deploying (use with caution in cloud-synced directories like iCloud, Dropbox, or OneDrive as it may create duplicate files)

amgr init

Initialize a new .amgr/config.json configuration file interactively.

amgr init

Prompts for:

• Target AI tools (claudecode, cursor, copilot, etc.)
• Features to include (rules, commands, skills, etc.)
• Use-cases (development, writing, product, etc.)
• Output directories (optional, per use-case)
• Advanced options (optional)

amgr config

Interactively edit an existing .amgr/config.json configuration file.

amgr config

Provides a menu to edit:

• Targets - AI tools to generate configs for
• Features - Content types to include
• Use-cases - Which use-cases to enable
• Output directories - Custom paths per use-case
• Options - Advanced settings

Options:

• -v, --verbose - Enable verbose output
• -c, --config <path> - Use a custom config file path

amgr list

List available profiles from the agents repository.

amgr list
amgr list --verbose    # Also show targets and features

Nested profiles are displayed with a tree structure showing their sub-profiles.

amgr validate

Validate the configuration file without syncing.

amgr validate
amgr validate --verbose    # Show configuration summary

amgr clean

Remove all generated agent configuration files tracked by amgr.

amgr clean
amgr clean --dry-run    # Preview what would be removed

amgr detach

Remove all amgr-created files and optionally the configuration.

amgr detach

This command:

1. Removes all files tracked in the lock file
2. Deletes the lock file
3. Optionally removes .amgr/config.json and the .amgr/ directory

amgr source

Manage rules sources for your project. Sources can be local paths or Git URLs, allowing you to use rules from multiple repositories.

amgr source add <url-or-path>

Add a rules source to the project configuration.

amgr source add https://github.com/company/agent-rules
amgr source add ~/my-personal-agents
amgr source add ./local-rules

Options:

• --position <index> - Insert at specific position (default: append at end)
• --name <name> - Optional display name for the source

amgr source remove <index-or-name>

Remove a source from the project configuration.

amgr source remove 0              # Remove by index
amgr source remove company        # Remove by name

Options:

• -f, --force - Skip confirmation prompt

amgr source list

Show configured sources and their status.

amgr source list
amgr source list --verbose    # Show full URLs/paths

amgr source update

Manually refresh all Git sources (fetch latest changes).

amgr source update
amgr source update --global    # Update only global sources

Global Sources

Global sources are configured once in ~/.amgr/config.json and are automatically available to all projects.

# Add a global source (available to all projects)
amgr source add ~/Code/my-agents --global
amgr source add https://github.com/company/rules --global

# List global sources
amgr source list --global

# Remove a global source
amgr source remove agents --global

Global config (~/.amgr/config.json):

{
  "globalSources": [
    { "type": "local", "path": "~/Code/agents", "name": "agents" }
  ]
}

By default, global sources are prepended to project sources (project sources override global). Projects can control this behavior via options.ignoreGlobalSources and options.globalSourcesPosition.

Source Layering

When multiple sources are configured, they are applied in order:

1. First source provides the base content
2. Later sources override earlier ones
3. This allows "company rules + personal overrides" patterns

{
  "sources": [
    { "type": "git", "url": "https://github.com/company/agent-rules" },
    { "type": "local", "path": "~/my-personal-agents" }
  ]
}

amgr repo

Manage standalone amgr repositories. An amgr repo is a directory containing agent configurations that can be used with amgr sync.

amgr repo init

Initialize a new amgr repository in the current directory.

amgr repo init
amgr repo init --name "my-agents" --description "My agent configs" --author "Your Name"

Creates the following structure:

my-agents-repo/
├── repo.json          # Repository manifest
├── shared/            # Shared content across all use-cases
│   ├── rules/
│   ├── commands/
│   ├── skills/
│   └── subagents/
└── use-cases/         # Use-case specific content

Options:

• --name <name> - Repository name (defaults to directory name)
• --description <desc> - Repository description
• --author <author> - Repository author

amgr repo add <name>

Add a new profile to the repository. Supports flat profiles, nested profiles, and sub-profiles.

Flat profile:

amgr repo add writing
amgr repo add writing --description "Documentation and content"

Nested profile (with shared content across sub-profiles):

amgr repo add development --nested
amgr repo add development --nested --description "Coding and debugging"

Sub-profile:

amgr repo add development:frontend
amgr repo add development:frontend --description "React, Vue, browser APIs"

Options:

• --description <desc> - Profile description (prompted if not provided)
• --nested - Create a nested profile with _shared/ directory

amgr repo remove <name>

Remove a profile from the repository.

amgr repo remove writing              # Remove flat profile
amgr repo remove development:frontend # Remove sub-profile
amgr repo remove development --force  # Skip confirmation

Options:

• -f, --force - Skip confirmation prompt

amgr repo list

List profiles in the current repository.

amgr repo list
amgr repo list --verbose    # Show orphaned directories

Nested profiles are displayed with a tree structure showing their sub-profiles.

Repository Auto-Detection

When running amgr repo list inside a directory containing repo.json, amgr automatically uses that directory as the agents source. This means you can:

cd /path/to/my-agents-repo
amgr repo list               # Lists use-cases from this repo

To use a repository as a source in your project, add it via amgr source add:

amgr source add /path/to/my-agents-repo
amgr source add https://github.com/company/agent-rules

Configuration

Configuration lives in .amgr/config.json in your project directory.

Example Configuration

{
  "$schema": "https://raw.githubusercontent.com/oztamir/amgr/main/schemas/amgr.schema.json",
  "targets": ["claudecode", "cursor"],
  "features": ["rules", "commands", "skills"],
  "profiles": ["development:frontend"]
}

Example with Custom Sources

{
  "$schema": "https://raw.githubusercontent.com/oztamir/amgr/main/schemas/amgr.schema.json",
  "sources": [
    { "type": "git", "url": "https://github.com/company/agent-rules" },
    { "type": "local", "path": "~/my-personal-agents" }
  ],
  "targets": ["claudecode", "cursor"],
  "features": ["rules", "commands", "skills"],
  "profiles": ["development:frontend", "writing"]
}

Example with Custom Output Directories

{
  "$schema": "https://raw.githubusercontent.com/oztamir/amgr/main/schemas/amgr.schema.json",
  "targets": ["claudecode", "cursor"],
  "features": ["rules", "commands"],
  "profiles": ["development:frontend", "product"],
  "outputDirs": {
    "product": "docs/"
  }
}

Migration Note: The use-cases field is deprecated. Use profiles instead. Both are supported for backwards compatibility.

This places product use-case files in docs/.claude/ and docs/.cursor/ instead of the project root.

Properties

sources (optional)

Array of rules sources. Each source can be:

• A git URL (e.g., https://github.com/company/agent-rules)
• A local path (e.g., ~/my-agents or ./local-rules)
• An object with type, url/path, and optional name

Sources are processed in order - later sources override earlier ones.

| Format | Example | |--------|---------| | Git URL string | "https://github.com/company/rules" | | Local path string | "~/my-agents" | | Git object | { "type": "git", "url": "...", "name": "company" } | | Local object | { "type": "local", "path": "...", "name": "personal" } |

targets (required)

AI tools to generate configurations for:

| Target | Description | |--------|-------------| | agentsmd | AGENTS.md (generic) | | antigravity | Antigravity AI | | augmentcode | Augment Code | | augmentcode-legacy | Augment Code (legacy) | | claudecode | Claude Code (Anthropic's CLI) | | claudecode-legacy | Claude Code (legacy) | | cline | Cline VS Code extension | | codexcli | OpenAI Codex CLI | | copilot | GitHub Copilot | | cursor | Cursor IDE | | geminicli | Gemini CLI | | junie | JetBrains Junie | | kilo | Kilo Code | | kiro | Kiro AI | | opencode | OpenCode | | qwencode | Qwen Code | | roo | Roo Code | | warp | Warp Terminal | | windsurf | Windsurf IDE | | zed | Zed Editor |

Use "*" to generate for all supported tools.

features (required)

Content types to include:

| Feature | Description | |---------|-------------| | rules | General guidelines and instructions | | ignore | File patterns to exclude from AI context | | mcp | MCP server configurations | | commands | Slash commands (e.g., /commit) | | subagents | Specialized AI assistant definitions | | skills | Directory-based capability definitions |

profiles (required)

Profile identifiers that map to folders in your configured sources. Supports both flat profiles and nested profiles with sub-profiles.

Profile selection syntax:

| Syntax | Meaning | |--------|---------| | "writing" | Flat profile | | "development:frontend" | Single sub-profile | | "development:*" | All sub-profiles (wildcard) | | "development" | Shorthand for development:* if nested |

Note: Sources must be configured for amgr sync to work. Profiles are defined by source repositories. Run amgr list to see available profiles from your configured sources.

use-cases (deprecated)

Deprecated: Use profiles instead. Still supported for backwards compatibility.

options (optional)

Advanced configuration options:

{
  "options": {
    "simulateCommands": false,
    "simulateSubagents": false,
    "simulateSkills": false,
    "modularMcp": false,
    "ignoreGlobalSources": false,
    "globalSourcesPosition": "prepend"
  }
}

| Option | Type | Default | Description | |--------|------|---------|-------------| | simulateCommands | boolean | false | Generate simulated commands for tools without native support | | simulateSubagents | boolean | false | Generate simulated subagents for tools without native support | | simulateSkills | boolean | false | Generate simulated skills for tools without native support | | modularMcp | boolean | false | Enable modular-mcp for Claude Code (reduces token usage) | | ignoreGlobalSources | boolean | false | Ignore global sources for this project | | globalSourcesPosition | string | "prepend" | Where to merge global sources: "prepend" (project overrides global) or "append" (global overrides project) |

outputDirs (optional)

Custom output directory prefixes per use-case. By default, all use-cases deploy files to the project root (e.g., .claude/, .cursor/). Use outputDirs to place specific use-cases in subdirectories.

{
  "use-cases": ["development", "product"],
  "outputDirs": {
    "product": "docs/"
  }
}

With this configuration:

• development rules → .claude/, .cursor/ (default)
• product rules → docs/.claude/, docs/.cursor/

Rules from shared/ are deployed to all configured output directories.

| Property | Type | Description | |----------|------|-------------| | <use-case> | string | Directory prefix for this use-case. Will be normalized to end with /. Example: "docs" or "docs/" |

Validation rules:

• Keys must match use-cases in the use-cases array
• Values must be relative paths (no leading /)
• Values must not contain .. path traversal

File Tracking

amgr maintains a lock file (.amgr/amgr-lock.json) to track files it creates. This ensures:

• Safe updates: Only amgr-managed files are modified during sync
• Native file preservation: Files you create manually are never touched
• Clean removal: amgr clean and amgr detach only remove tracked files

Recommended .gitignore

# amgr lock file (project-local state)
.amgr/amgr-lock.json

# Or ignore the entire directory
.amgr/

You may choose to commit .amgr/config.json to share the configuration with your team.

Git Source Caching

Git sources are cached locally in ~/.amgr/cache/:

~/.amgr/
├── config.json                        # Global sources configuration
└── cache/
    ├── github.com-company-agent-rules/
    │   ├── .git/
    │   ├── repo.json
    │   ├── shared/
    │   └── use-cases/
    └── github.com-other-repo/
        └── ...

• On amgr sync: Git sources are automatically pulled (updated)
• Cache is created on first use and reused for subsequent syncs
• Use amgr source update to manually refresh all git sources

Environment Variables

| Variable | Description | |----------|-------------| | AMGR_CONFIG | Override the config file path | | AMGR_VERBOSE | Enable verbose logging (true/false) |

Repository Configuration (repo.json)

When managing a standalone amgr repository, repo.json defines the repository metadata and available use-cases.

Example repo.json

{
  "$schema": "https://raw.githubusercontent.com/oztamir/amgr/main/schemas/amgr-repo.schema.json",
  "name": "my-agents",
  "description": "My personal agent configurations",
  "version": "1.0.0",
  "author": "Your Name",
  "profiles": {
    "development": {
      "description": "Coding and debugging",
      "sub-profiles": {
        "frontend": { "description": "React, Vue, browser APIs" },
        "backend": { "description": "Node.js, APIs, databases" }
      }
    },
    "writing": {
      "description": "Documentation and content creation"
    }
  }
}

Properties

| Property | Required | Description | |----------|----------|-------------| | name | Yes | Repository name | | description | No | Repository description | | version | No | Semantic version (e.g., "1.0.0") | | author | No | Repository author or maintainer | | profiles | Yes* | Object mapping profile names to their metadata | | use-cases | Yes* | (Deprecated) Object mapping use-case names to their metadata |

*Either profiles or use-cases is required. Prefer profiles for new repositories.

Each profile entry must include a description field. Nested profiles can include a sub-profiles object.

How It Works

1. Parse Configuration - Reads .amgr/config.json and validates it
2. Clean Previous Files - Removes files listed in the lock file
3. Compose Content - Merges shared content with use-case specific content
4. Generate Configs - Runs rulesync to create tool-specific configurations
5. Deploy Files - Copies generated files to target directories (.claude/, .cursor/, etc.)
6. Update Lock File - Records all created files for future reference

Conflict Handling

When amgr encounters a file that exists but isn't tracked:

1. The file is skipped (not overwritten)
2. A warning is displayed
3. Other files continue to be processed

To resolve conflicts:

• Remove or rename the conflicting file
• Re-run amgr sync