Agent Creation

Mega Brain supports creating sophisticated AI agents with personality (SOUL), knowledge (MEMORY), and DNA-based expertise weighting. This guide covers the complete agent creation workflow.

Agent Architecture

Agents in Mega Brain follow a structured template system:

┌─────────────────────────────────────────────────────────────────────────────┐
│  AGENT STRUCTURE                                                             │
│                                                                              │
│  agents/cargo/AGENT-NAME/                                                    │
│  ├── AGENT.md          ← Main prompt (required)                             │
│  ├── SOUL.md           ← Personality definition (optional)                   │
│  ├── MEMORY.md         ← Knowledge base (optional)                           │
│  └── DNA-CONFIG.yaml   ← Expert DNA weights (optional)                       │
└─────────────────────────────────────────────────────────────────────────────┘

Agent Types

Mega Brain defines 4 agent categories:

CARGO (Functional)
MINDS (Expert Clones)
CONCLAVE (Multi-Agent)
SYSTEM (Core)

Purpose: Functional roles like Sales, Finance, OperationsLocation: agents/cargo/Count: 29 agentsExamples:

AGENT-CFO - Chief Financial Officer
AGENT-CLOSER - Sales closer
AGENT-SDR - Sales Development Rep
AGENT-CMO - Chief Marketing Officer

Template: TEMPLATE-AGENT-MD-ULTRA-ROBUSTO-V3.md

Purpose: Mind clones of specific expertsLocation: agents/minds/Count: 5 agentsExamples:

Alex Hormozi
Sam Ovens
Cole Gordon
Jeremy Haynes

DNA: High-fidelity replication of expert mental models

Purpose: Multi-perspective deliberationLocation: agents/conclave/Count: 3 agentsExamples:

Devil’s Advocate
Optimist
Pragmatist

Usage: Council sessions via /conclave

Purpose: System orchestrationLocation: agents/system/Count: 2 agentsExamples:

JARVIS - Orchestrator
Agent-Creator - Meta-agent

Total Active Agents: 37 (29 cargo + 5 minds + 3 conclave + 2 system)

AGENT.md Template

The AGENT.md file is the primary prompt for the agent. Mega Brain uses TEMPLATE-AGENT-MD-ULTRA-ROBUSTO-V3.md:

# ═══════════════════════════════════════════════════════════════════════════
# AGENT.md - [AGENT-NAME]
# ═══════════════════════════════════════════════════════════════════════════

╔════════════════════════════════════════════════════════════════════════════╗
║   [ASCII LOGO]                                                             ║
║   [AGENT TITLE]                                                            ║
║   "[Tagline]"                                                               ║
╠════════════════════════════════════════════════════════════════════════════╣
║   TIPO       │ [CARGO]                CATEGORIA  │ [CATEGORY]              ║
║   VERSÃO     │ 3.0.0                  ATUALIZADO │ [DATE]                  ║
║   MATURIDADE │ ████████████████░░░░░░ [%]                                  ║
╚════════════════════════════════════════════════════════════════════════════╝

TypePathREADSknowledge/dna/WRITESartifacts/DEPENDS_ONConstitution Article 7

```yaml DNA-CONFIG.yaml
version: "1.0"
agent_name: "AGENT-CFO"
category: "Finance"

dna:
  experts:
    - name: "Sam Ovens"
      domains:
        - "systems"
        - "scaling"
        - "operations"
      weight: 85
      source: "Consulting Accelerator"
    
    - name: "Alex Hormozi"
      domains:
        - "scaling"
        - "offers"
        - "unit-economics"
      weight: 80
      source: "$100M Offers, $100M Leads"
    
    - name: "Cole Gordon"
      domains:
        - "compensation"
      weight: 75
      source: "Closers 2.0"

dimensions:
  - name: "Cash Flow"
    score: 10
    description: "Cash flow is survival"
  
  - name: "Unit Economics"
    score: 9
    description: "LTV/CAC must be > 3"
  
  - name: "Risk Management"
    score: 9
    description: "Calculate downside first"

Referencing System

Agents use citation references to ensure traceability:

^[SOUL.md:17]           ← Direct citation from SOUL.md line 17
^[MEMORY.md:48-49]      ← Range citation from MEMORY.md
^[derivado:MEMORY.md:contagem-linhas-47-88]  ← Derived from analysis
^[DNA-CONFIG.yaml:43-47]  ← Configuration reference

All agent content must be 100% traceable to source files. Never make up citations.

Creating a New Agent

Choose Agent Category

Decide which category fits your agent:

CARGO - Functional role (Sales, Finance, Ops)
MINDS - Expert mind clone
CONCLAVE - Council participant
SYSTEM - Core orchestration

Create Agent Directory

# For cargo agents
mkdir -p agents/cargo/AGENT-[NAME]
cd agents/cargo/AGENT-[NAME]

# Copy template
cp agents/_templates/TEMPLATE-AGENT-MD-ULTRA-ROBUSTO-V3.md AGENT.md

Define Agent Identity

Edit AGENT.md and fill in:

Agent name - AGENT-[NAME] (uppercase)
Title - Role description
Category - Sales, Finance, Operations, etc.
Tagline - One-line description

# AGENT-CFO

> **Título:** "O Guardião da Saúde Financeira"
> **Categoria:** Finance
> **Versão:** 3.0.0

Define DNA (Optional)

Create DNA-CONFIG.yaml with expert weights:

version: "1.0"
agent_name: "AGENT-CFO"
category: "Finance"

dna:
  experts:
    - name: "Sam Ovens"
      domains: ["systems", "scaling"]
      weight: 85
    - name: "Alex Hormozi"
      domains: ["offers", "unit-economics"]
      weight: 80

Define Personality (Optional)

Create SOUL.md with personality traits:

# SOUL - AGENT-CFO

## IDENTITY

I am the guardian of financial health. The brake when needed, the accelerator when numbers allow.

## PHRASES I USE

- "Cash is king, everything else is noise"
- "Margin protects from errors"
- "Predictability > Spikes"

## DECISION RULES

1. No ROI = No approval
2. Payback > 12 months = Strong justification required
3. Never approve spend I can't explain with numbers

Add Knowledge (Optional)

Create MEMORY.md with knowledge base:

# MEMORY - AGENT-CFO

## INSIGHTS

### From Alex Hormozi

- LTV/CAC must be > 3 ^[source:100M-offers:p47]
- Break-even payback < 6 months ideal ^[source:100M-leads:p89]

### From Sam Ovens

- Systems before scale ^[source:consulting-accelerator:m3]
- Cash runway minimum 6 months ^[source:consulting-accelerator:m7]

Update AGENT-INDEX.yaml

Add agent to the master index:

agents:
  cargo:
    - name: AGENT-CFO
      category: Finance
      file: agents/cargo/AGENT-CFO/AGENT.md
      soul: agents/cargo/AGENT-CFO/SOUL.md
      memory: agents/cargo/AGENT-CFO/MEMORY.md
      dna: agents/cargo/AGENT-CFO/DNA-CONFIG.yaml
      status: active
      version: 3.0.0

Validate Agent

Run validation checks:

# Check AGENT.md structure
python3 core/intelligence/validate_agent.py agents/cargo/AGENT-CFO

# Verify all citations exist
python3 core/intelligence/verify_citations.py agents/cargo/AGENT-CFO/AGENT.md

DNA System

Mega Brain’s DNA system provides 5 layers of knowledge:

L1: PHILOSOPHIES

Core beliefs and worldviewExamples:

“Systems before scale” (Sam Ovens)
“Value arbitrage” (Alex Hormozi)
“Integrity closes deals” (Cole Gordon)

Format:

## PHILOSOPHIES

### [Expert Name]

- [Philosophy 1]
- [Philosophy 2]

L2: MENTAL-MODELS

Thinking and decision frameworksExamples:

“Unit economics model” (Alex Hormozi)
“Value ladder” (Russell Brunson)
“Flywheel” (Amazon)

Format:

## MENTAL-MODELS

### [Model Name]

**Description:** [What it is]
**Application:** [When to use]
**Formula:** [If applicable]

L3: HEURISTICS

Practical rules and decision shortcutsExamples:

“LTV/CAC > 3” (SaaS rule)
“20% max on comp” (Sales compensation)
“6-month cash runway minimum” (Startups)

Format:

## HEURISTICS

- If [condition] → [action]
- [Rule]: [threshold] = [consequence]

L4: FRAMEWORKS

Structured methodologies and processesExamples:

“CLOSER Framework” (Cole Gordon)
“Value Equation” (Alex Hormozi)
“Consulting Process” (Sam Ovens)

Format:

## FRAMEWORKS

### [Framework Name]

1. [Step 1]
2. [Step 2]
3. [Step 3]

L5: METHODOLOGIES

Step-by-step implementationsExamples:

“7-step sales call” (Cole Gordon)
“Offer creation process” (Alex Hormozi)
“Client onboarding” (Sam Ovens)

Format:

## METHODOLOGIES

### [Process Name]

**Step 1:** [Action]
**Step 2:** [Action]
**Step 3:** [Action]

Agent Lifecycle

Automatic Creation

Agents can be auto-created via hooks:

# agent_creation_trigger.py (PostToolUse hook)

def detect_agent_request(tool_output: str) -> bool:
    """Detects if user wants to create an agent."""
    patterns = [
        r'create agent',
        r'novo agente',
        r'new agent'
    ]
    return any(re.search(p, tool_output, re.I) for p in patterns)

if detect_agent_request(tool_output):
    trigger_agent_creation()

Agent Activation

Agents are activated via:

Direct invocation - Mention agent by name
Skill routing - Auto-activated by keywords
Hook trigger - Activated by lifecycle events
Slash commands - /agent-name

Agent Index

AGENT-INDEX.yaml maintains the registry:

version: "2.0.0"
last_updated: "2024-12-25"

agents:
  cargo:
    - name: AGENT-CFO
      category: Finance
      file: agents/cargo/AGENT-CFO/AGENT.md
      soul: agents/cargo/AGENT-CFO/SOUL.md
      memory: agents/cargo/AGENT-CFO/MEMORY.md
      dna: agents/cargo/AGENT-CFO/DNA-CONFIG.yaml
      status: active
      version: 3.0.0
      keywords: ["finance", "cfo", "financial"]
  
  minds:
    - name: ALEX-HORMOZI
      category: Expert
      file: agents/minds/ALEX-HORMOZI/AGENT.md
      status: active
      version: 2.0.0

agent_index_updater.py hook automatically updates this file when agents are created/modified.

Agent Templates

Mega Brain provides templates in agents/_templates/:

TEMPLATE-AGENT-MD-ULTRA-ROBUSTO-V3.md
INDEX.md

Version: 3.1Size: 78KBFeatures:

Full DNA integration
Citation system
Incremental sections
Dependencies tracking

Use for: Production cargo agents

Best Practices

Agent Design Principles

Single Responsibility - One agent, one role
Traceable - All content cited from sources
DNA-Driven - Base on expert mental models
Iterative - Start minimal, expand over time
Validated - Use validation tools before deployment

Quality Checklist

Troubleshooting

Agent Not Found

Problem: Agent exists but isn’t activatedSolutions:

Check AGENT-INDEX.yaml includes the agent
Verify file path is correct
Ensure status is “active” not “draft”
Run agent_index_updater.py to rebuild index

Citations Broken

Problem: References like ^[SOUL.md:17] don’t resolveSolutions:

Verify SOUL.md exists at correct path
Check line number exists in file
Run verify_citations.py to find all broken refs
Update references to correct line numbers

DNA Conflicts

Problem: Multiple experts conflict in decision-makingSolutions:

Adjust expert weights in DNA-CONFIG.yaml
Create resolution rules in AGENT.md
Define priority order for conflicts
Consider splitting into multiple agents

DNA System

Deep dive into the 5-layer DNA knowledge system

Skills Routing

Auto-activate agents with keyword routing

Hooks System

Agent lifecycle hooks and automation

Validation

Validate agent structure and content

Get Started

Core Concepts

CLI Commands

Guides

Advanced

Agent Creation

Agent Creation

Agent Architecture

Agent Types

AGENT.md Template

Referencing System

Creating a New Agent

DNA System

Agent Lifecycle

Automatic Creation

Agent Activation

Agent Index

Agent Templates

Best Practices

Agent Design Principles

Quality Checklist

Troubleshooting

DNA System

Skills Routing

Hooks System

Validation

​Agent Creation

​Agent Architecture

​Agent Types

​AGENT.md Template

​Referencing System

​Creating a New Agent

​DNA System

​Agent Lifecycle

​Automatic Creation

​Agent Activation

​Agent Index

​Agent Templates

​Best Practices

Agent Design Principles

​Quality Checklist

​Troubleshooting

​Related Documentation

DNA System

Skills Routing

Hooks System

Validation

Agent Creation

Agent Architecture

Agent Types

AGENT.md Template

Referencing System

Creating a New Agent

DNA System

Agent Lifecycle

Automatic Creation

Agent Activation

Agent Index

Agent Templates

Best Practices

Quality Checklist

Troubleshooting

Related Documentation