Agent Creator

Agent Creator

Overview

This skill enables the automatic creation of specialized agents for the Miyabi framework. It generates agent definition files (.md) with complete workflows, tool access configurations, and quality standards based on established templates. Perfect for quickly creating domain-specific agents without manual configuration.

Quick Start

Create Your First Agent

# Initialize the agents directory
python .claude/skills/agent-creator/scripts/init_agent.py --init

# Generate a code review agent
python .claude/skills/agent-creator/scripts/generate_agent.py code-review-expert code-review

# Generate a deployment agent
python .claude/skills/agent-creator/scripts/generate_agent.py deployment-automator deployment

Available Agent Types

• code-review: Comprehensive code review specialist (architecture, security, performance, testing)
• deployment: CI/CD automation and deployment pipeline management
• testing: Test automation and quality assurance specialist
• documentation: Documentation generation and maintenance expert
• security: Security analysis and vulnerability assessment specialist
• general-purpose: Research and complex multi-step task coordinator

Agent Creation Workflow

Step 1: Directory Setup

First, ensure the agents directory exists:

# Check if agents directory exists
python .claude/skills/agent-creator/scripts/init_agent.py --status

# Initialize if needed
python .claude/skills/agent-creator/scripts/init_agent.py --init

Step 2: Choose Agent Type

Select the appropriate agent type based on your needs:

Step 3: Generate Agent

Use the generator script with appropriate parameters:

# Basic generation
python .claude/skills/agent-creator/scripts/generate_agent.py <name> <type>

# With custom description
python .claude/skills/agent-creator/scripts/generate_agent.py \
  security-auditor security \
  --description "Specialized security auditor for web applications"

# With custom capabilities
python .claude/skills/agent-creator/scripts/generate_agent.py \
  api-tester testing \
  --capabilities "API testing" "Load testing" "Integration testing"

# Dry run to preview
python .claude/skills/agent-creator/scripts/generate_agent.py \
  custom-agent general-purpose --dry-run

Step 4: Validate Agent

Always validate generated agents:

# Validate specific agent
python .claude/skills/agent-creator/scripts/validate_agent.py <agent-name>.md

# Validate all agents
python .claude/skills/agent-creator/scripts/validate_agent.py --all

Agent Customization

Custom Descriptions

Add domain-specific context to your agent:

python .claude/skills/agent-creator/scripts/generate_agent.py \
  react-reviewer code-review \
  --description "Specialized code reviewer for React applications with focus on hooks, performance, and accessibility"

Custom Capabilities

Override default capabilities for your specific use case:

python .claude/skills/agent-creator/scripts/generate_agent.py \
  microservices-deployer deployment \
  --capabilities \
    "Kubernetes deployment automation" \
    "Service mesh configuration" \
    "Container orchestration" \
    "Health monitoring setup"

Custom Tool Access

Restrict or expand tool access as needed:

python .claude/skills/agent-creator/scripts/generate_agent.py \
  documentation-writer documentation \
  --tools Read Write Edit Grep Glob

Custom Workflow Instructions

Provide domain-specific workflow guidance:

python .claude/skills/agent-creator/scripts/generate_agent.py \
  blockchain-auditor security \
  --workflow "Focus on smart contract security, DeFi vulnerability patterns, and gas optimization issues"

Agent Types Reference

Code Review Agent

Best for: Development teams needing systematic code quality analysis

Key Features:

• Architecture & design analysis (25% weight)
• Code quality assessment (20% weight)
• Security vulnerability detection (20% weight)
• Performance & scalability evaluation (15% weight)
• Testing coverage analysis (10% weight)
• Documentation & API design review (10% weight)

Quality Threshold: 80+ points required for progression

Deployment Agent

Best for: DevOps teams managing deployment pipelines

Key Features:

• Pre-deployment validation
• Automated deployment execution
• Health check monitoring
• Rollback management
• Post-deployment documentation

Integration: Works with CoordinatorAgent for task delegation

Testing Agent

Best for: QA teams ensuring comprehensive testing coverage

Key Features:

• Test automation execution
• Coverage report generation (80%+ requirement)
• Quality assurance validation
• Test result analysis
• Performance testing

Test Types: Unit, integration, E2E, security, performance

Documentation Agent

Best for: Technical writers and development teams

Key Features:

• API documentation generation
• README and guide creation
• Technical writing assistance
• Documentation maintenance
• Example code generation

Documentation Types: API docs, user guides, technical specifications

Security Agent

Best for: Security teams and compliance officers

Key Features:

• OWASP Top 10 vulnerability scanning
• Security best practices validation
• Compliance assessment
• Risk analysis and categorization
• Security reporting

Risk Categories: Critical, High, Medium, Low with remediation timelines

General Purpose Agent

Best for: Complex research and multi-step coordination tasks

Key Features:

• Broad research capabilities
• Multi-step task coordination
• Information synthesis
• Problem decomposition
• Full tool access (*)

Use Cases: Research projects, problem analysis, cross-functional coordination

Best Practices

Agent Naming

Follow these naming conventions:

# Good examples
code-review-expert
deployment-automator
security-analyzer
documentation-generator
testing-orchestrator

# Use kebab-case
# Include descriptive suffixes
# Make purpose clear from name

Quality Standards

All generated agents should:

• Maintain 80+ quality score thresholds
• Provide clear, actionable feedback
• Follow established coding standards
• Document significant decisions
• Include proper error handling
• Validate inputs and outputs

Integration Patterns

# Sequential agent usage
code-review-expert → testing → deployment

# Parallel agent usage
(code-review AND security) → testing

# Research to implementation
general-purpose → specialized-agent

Resources

scripts/

Core Scripts:

• generate_agent.py - Main agent generation script with template system
• init_agent.py - Directory setup and initialization utility
• validate_agent.py - Agent definition validation and quality checking

Usage Examples:

# Generate with all options
python scripts/generate_agent.py \
  my-specialized-agent code-review \
  --description "Custom description here" \
  --capabilities "Custom cap 1" "Custom cap 2" \
  --tools Read Grep Edit Bash \
  --workflow "Custom workflow instructions"

# List available templates
python scripts/generate_agent.py --list-templates

# Initialize directory
python scripts/init_agent.py --init

# Validate all agents
python scripts/validate_agent.py --all

references/

Documentation Files:

• agent_types.md - Comprehensive reference for all agent types, characteristics, and use cases
• workflow_patterns.md - Detailed workflow patterns and quality standards for each agent type

Key Information:

• Agent type selection guidelines
• Workflow quality standards
• Integration patterns and best practices
• Error handling patterns
• Performance optimization guidelines

assets/

Template Files:

• template_example.md - Example of a generated agent definition file showing proper structure and format

Purpose:

• Demonstrates expected output format
• Provides reference for manual customization
• Shows best practices for agent documentation

Troubleshooting

Common Issues

Agent not found after creation:

# Restart Claude Code to reload agents
# Verify agent exists in .claude/agents/
ls .claude/agents/

Validation errors:

# Check detailed validation output
python scripts/validate_agent.py agent-name.md

# Common fixes:
# - Add missing sections
# - Fix tool references
# - Improve workflow instructions

Directory not found:

# Initialize agents directory
python scripts/init_agent.py --init

# Check directory status
python scripts/init_agent.py --status

Getting Help

• Use --help flag with any script for usage information
• Check validation output for specific issues
• Review agent type references for proper configuration
• Consult workflow patterns for best practices

This agent-creator skill follows the Miyabi framework principles and integrates with the common library system for consistent environment management.