Interactive agent builder creating BMad Core compliant agents as YAML source files that compile to .md during installation.
# Direct workflow
workflow create-agent
# Via BMad Builder
*create-agent- Self-contained functionality
- Basic command structure
- No external resources
- Sidecar resources for domain knowledge
- Extended capabilities
- Knowledge base integration
- Full-featured with workflows
- Module-specific commands
- Integrated with module structure
- Creative ideation session
- Explore concepts and personalities
- Generate command ideas
- Output feeds into persona development
- Choose agent type (Simple/Expert/Module)
- Define identity (name, title, icon, filename)
- Assign to module (if Module agent)
- Define role and responsibilities
- Craft unique identity/backstory
- Select communication style
- Establish guiding principles
- Add critical actions (optional)
- Add required commands (*help, *exit)
- Define workflow commands
- Add task commands
- Create action commands
- Configure attributes
- Generate .agent.yaml file
- Create customize file (optional)
- Setup sidecar resources (Expert agents)
- Validate and compile
- Provide usage instructions
Standalone Agents:
- Source:
{bmad_folder}/agents/{filename}.agent.yaml - Compiled:
{bmad_folder}/agents/{filename}.md
Module Agents:
- Source:
src/modules/{module}/agents/{filename}.agent.yaml - Compiled:
{bmad_folder}/{module}/agents/{filename}.md
agent:
metadata:
id: {bmad_folder}/{module}/agents/{filename}.md
name: Agent Name
title: Agent Title
icon: 🤖
module: module-name
persona:
role: '...'
identity: '...'
communication_style: '...'
principles: ['...', '...']
menu:
- trigger: command-name
workflow: path/to/workflow.yaml
description: Command descriptionLocation: {bmad_folder}/_cfg/agents/{module}-{filename}.customize.yaml
Allows persona and menu overrides that persist through updates.
Quick Rebuild:
bmad compile-agentsDuring Module Install: Automatic compilation when installing modules
Manual Compilation:
node tools/cli/bmad-cli.js compile-agentsUser: I need a code review agent
Builder: Let's brainstorm first...
[Brainstorming generates ideas for strict vs friendly reviewer]
Builder: Now let's build your agent:
- Type: Simple
- Name: Code Reviewer
- Role: Senior developer conducting thorough reviews
- Style: Professional but approachable
- Commands:
- *review-pr: Review pull request
- *review-file: Review single file
- *review-standards: Check coding standards
Type: Expert
Name: Legal Advisor
Sidecar: legal-knowledge/
Commands:
- *contract-review
- *compliance-check
- *risk-assessment
create-agent/
├── workflow.yaml # Configuration
├── instructions.md # Step guide
├── checklist.md # Validation
├── README.md # This file
├── agent-types.md # Type details
├── agent-architecture.md # Patterns
├── agent-command-patterns.md # Commands
└── communication-styles.md # Styles
- Use brainstorming for complex agents
- Start simple - Add commands incrementally
- Test commands before finalizing
- Document thoroughly in descriptions
- Follow naming conventions consistently