StarterApp Docs
AI WorkflowMethodologies

BMAD

Breakthrough Method of Agile AI-Driven Development

BMAD (Breakthrough Method of Agile AI-Driven Development) orchestrates multiple AI agents working on complex features simultaneously. The /bmad-init command provides a lite implementation focusing on parallel development with bounded contexts.

Parallel AI Development

BMAD decomposes complex features into isolated work streams, allowing multiple AI agents to implement different components simultaneously while maintaining integration through shared contracts and bounded contexts.

Core Implementation

The codebase includes BMAD initialization through:

Initialize BMAD Workflow
/bmad-init "real-time collaboration system" --agents=3 --context=full

This command decomposes features into parallel work streams with specialized agents.

Agent Structure

The current implementation creates three default agent types:

  • Core Functionality Specialist - Implements primary feature logic
  • Testing and Validation Specialist - Creates comprehensive test suites
  • Integration and Deployment Specialist - Handles system connections

Additional agents can be specified for complex features:

Configure Multiple Agents
/bmad-init "payment processing" --agents=5 --prp

Generated Artifacts

The command creates coordination structure:

context.md
agent-1-prp.md
agent-2-prp.md
agent-3-prp.md
contracts.ts
integration-plan.md
  • context.md - Shared context for all agents
  • agent-1-prp.md - Core functionality PRP
  • agent-2-prp.md - Testing PRP
  • agent-3-prp.md - Integration PRP
  • contracts.ts - Interface definitions
  • integration-plan.md - Rollout strategy

Context Files

Each agent receives bounded context through individual PRPs:

agent-1-prp.md
# agent-1-prp.md
## Scope: Core Notification Engine
- WebSocket connection management
- Message broadcasting logic
- State synchronization

## Interfaces
See contracts.ts for MessageBroker interface

## Constraints
- No database access (use agent-3)
- Must implement all contracts.ts interfaces

Interface Contracts

TypeScript interfaces define boundaries:

bmad/notifications/contracts.ts
export interface MessageBroker {
  publish(channel: string, message: Message): Promise<void>;
  subscribe(channel: string, handler: Handler): Subscription;
}

Agents implement against these contracts ensuring compatibility.

Workflow Process

  1. Feature Analysis - Command analyzes complexity and identifies work streams
  2. Agent Assignment - Each stream gets dedicated agent context
  3. Contract Generation - Shared interfaces prevent integration conflicts
  4. PRP Creation - Detailed implementation plans for each agent
  5. Progress Tracking - Monitor multi-agent coordination

Parallel Execution

Agents work simultaneously within boundaries:

Agent Work Streams
Agent 1: Implements core business logic
Agent 2: Creates tests for agent 1 outputs
Agent 3: Integrates with existing systems

Clear separation enables true parallel development. Dependencies tracked in integration plan.

Command Options

Context Levels

Context Configuration
--context=minimal   # Basic feature requirements only
--context=standard  # Include architecture constraints
--context=full      # Complete codebase patterns

PRP Generation

Enable PRP Generation
--prp  # Generate detailed PRPs for each agent

Without this flag, agents receive lighter context files.

Limitations

This implementation provides BMAD-lite functionality:

  • No automated sprint ceremonies
  • No role-based agent specialization (Analyst, PM, etc.)
  • No story file generation
  • Manual coordination between agents
  • No built-in hand-off protocols

Full BMAD methodology includes agile ceremonies and specialized agent roles not currently implemented.

When to Use

The /bmad-init command suits:

  • Features with clear component boundaries
  • Parallel development opportunities
  • Need for isolated agent contexts
  • Complex multi-layer implementations

Simple features gain little from agent orchestration overhead.

Integration Benefits

Despite being a lite implementation, the command provides:

  • Bounded contexts preventing agent confusion
  • Interface contracts ensuring compatibility
  • Task decomposition for parallel work
  • Progress tracking through structured outputs

These benefits make complex features manageable through AI agent coordination.

Example Usage

Implementing a real-time collaboration system:

Real-Time Collaboration Example
/bmad-init "real-time collaboration" --agents=3 --context=full --prp

Generates:

  • Core agent builds WebSocket infrastructure
  • Testing agent creates integration tests
  • Integration agent connects to authentication and database

Agents work in parallel following their PRPs and respecting interface contracts.