Bundles Guide

Bundles are the primary way to customize Amplifier's behavior. A bundle packages together tools, agents, context, and configuration into a cohesive, reusable unit. This section teaches you how to use existing bundles, create your own, and compose them for powerful applications.

Section Contents

Page Description
Foundation Core Amplifier bundle with essential capabilities
Recipes Multi-step workflow orchestration
LSP Python Python code intelligence via Language Server Protocol
Design Intelligence Design system and UI expertise

Quick Tips

  • Thin bundles - Keep bundles focused; compose for complexity
  • Reuse context - Reference shared context files instead of duplicating
  • Version carefully - Bundle changes affect all users
  • Test in isolation - Verify bundles work independently before composing
  • Document behavior - Clear descriptions help users and AI understand intent

Bundle Anatomy

my-bundle/
├── bundle.yaml        # Bundle manifest
├── context/           # Knowledge and instructions
│   ├── README.md      # Primary context
│   └── examples/      # Example files
├── agents/            # Agent definitions
│   └── specialist.yaml
└── skills/            # Optional skills
    └── domain-skill.md

The Thin Bundle Philosophy

Bundles should be minimal compositions, not monolithic packages:

Do Don't
Reference shared context Duplicate instructions
Compose multiple thin bundles Build one giant bundle
Single responsibility Kitchen sink approach
Clear extension points Tightly coupled internals

Where to Start

Core capabilities? Begin with Foundation for the essential Amplifier bundle.

Workflow automation? Jump to Recipes for multi-step orchestration.

Python development? See LSP Python for code intelligence.

Example: Minimal Bundle

# bundle.yaml
name: my-assistant
description: Custom assistant behavior
version: 1.0.0

context:
  - context/README.md

extends:
  - foundation  # Inherit base capabilities

Next Steps

After mastering bundles, explore Skills for adding domain knowledge or Advanced for complex patterns.