augent 0.6.4

Lean package manager for various AI coding platforms

Augent

Augments AI coding platforms (such as Claude Code, OpenCode, Cursor) via bundles (of skills, commands, rules, MCP servers...) in a platform-independent, reproducible, and intuitive manner:

• It frees you from the burden of managing AI coding platform specific formats.
• It ensures reproducibility across the teams by locking bundle versions.
• You get and share bundles via public or private Git repositories. That's it.

Setup

You can install it from PyPI (recommended):

pip install --user augent

Or you can install it from npm:

npm install -g augent

If you have neither Python nor Node.js available, you can download a GitHub Release for your system and put the binary in your PATH.

Quick Start

To install a set of resources (bundles) for your AI coding platforms:

# Install bundle(s) from a public GitHub repository (select if many):
augent install @wshobson/agents

# List all installed bundles
augent list

# Show installation details
augent show @wshobson/agents

# Uninstall bundle (all under this prefix, select if many):
augent uninstall @wshobson/agents

Usage

Augent stores AI coding platform resources in universal format in bundles.

• Bundle: A path in a Git repository, optionally referencing other bundles
• Workspace: Your project's Git repository where you and your team work
• Resources: Universal resources transformed and installed for specific AI coding platforms

When you install a bundle from a remote Git repository, Augent:

1. Fetches the bundle(s) and adds it to .augent/augent.yaml in your workspace
2. Resolves and locks the Git ref on first install (by creating a lockfile)
3. By default, your AI coding platforms are auto-detected in the workspace.
4. Transforms the bundle's resources to match your AI coding platform's format
5. Installs resources to the platforms (and creates an index of what came where)

To ensure a coherent Augent setup across your team, store all three created files in .augent/ (yaml, index, and lock) in your Git repository.

Install Bundles

Install from local directory within workspace:

augent install ./local-bundle

Install only for specific platforms (otherwise installs to all detected):

augent install ./local-bundle --to cursor opencode

Install from GitHub repository, develop branch, subdirectory plugins/which:

augent install github:author/repo#develop:plugins/which

Install by using GitHub Web UI URL directly:

augent install https://github.com/author/bundle/tree/develop/plugins/which

Install from a Git repository over SSH:

augent install [e-mail]:author/bundled

The install command understands different repo formats, such as Claude Marketplace plugins.

If repository has many bundles (or Claude Marketplace plugins), you can select those from the menu (or pass --all-bundles).

Most commands will display an interactive menu if used without arguments.

Workspace Scope

All commands operate in your current workspace (you can pass -w, --workspace <PATH> to use different workspace).

Resolves remote bundles to the latest versions (and updates the lockfile):

augent install --update

List all installed bundles:

augent list

Show where bundle's resources are enabled:

augent show @author/repository/bundle

Uninstall the bundle and remove its resources:

augent uninstall @author/repository/bundle

Resources that came from the bundle are removed, unless you modified them first.

It also uninstalls the bundle's dependencies, unless used by other bundles.

Bundle Format

A bundle contains resources in platform-independent format, e.g.:

my-bundle/
├── augent.yaml          # Bundle metadata and dependencies (optional)
├── commands/            # Universal files for AI coding platforms
│   └── debug.md
├── skills/
│   └── web-browser.md
├── AGENTS.md
└── mcp.jsonc

Documentation

• Commands Reference - Detailed command documentation
• Bundle Format - Bundle structure and configuration
• Platform Support - Known platforms and adding new ones
• Workspace Configuration - Workspace setup and management

License

AGPL v3 - see LICENSE for details.

Acknowledgments

Platform conversion approach inspired by OpenPackage and rulesync.