Multi-Agent Configuration System#
This directory contains documentation and configuration specifications for the ADK Runbooks multi-agent system’s intelligent delegation framework.
Overview#
The multi-agent system uses a configuration-based approach to intelligently route security tasks to specialized sub-agents without hardcoded logic. This enables dynamic agent management through YAML configuration files.
Files#
Core Documentation#
configuration_based_delegation.md
- Complete system architecture and implementation guideconfig_quick_reference.md
- Quick reference for YAML configuration syntaxagent_workflow_references.md
- 15+ agent workflow patterns with Mermaid diagrams
Key Features#
Data-Driven Delegation: Route tasks based on YAML configurations
Pattern Matching: Intelligent request analysis and agent selection
Tool Assignment: Automatic MCP tool provisioning per agent
Expertise Mapping: Agent capabilities and specialization areas
Escalation Paths: Defined agent-to-agent escalation routes
Quick Start#
Review Patterns: Start with
agent_workflow_references.md
for design patternsConfiguration Syntax: Check
config_quick_reference.md
for YAML structureImplementation: Follow
configuration_based_delegation.md
for full setup
Configuration Structure#
multi-agent/manager/config/
├── agents/*.yaml # Individual agent capabilities
├── mcp_tools_config.yaml # MCP server configurations
├── tool_agent_mapping.yaml # Tool-to-agent mappings
└── config_loader.py # Python configuration loader
Agent Types Supported#
SOC Analyst (Tier 1, 2, 3)
CTI Researcher
Incident Responder
Threat Hunter
Detection Engineer
SOAR Specialist
Benefits#
No Code Changes: Update agent behavior via configuration
Maintainable: Centralized capability definitions
Flexible: Easy addition of new agents and tools
Intelligent: Pattern-based task routing
Documented: Configuration serves as living documentation
Integration#
This system integrates with the main ADK Runbooks multi-agent system located in /multi-agent/manager/
and works with both traditional and A2A (Agent-to-Agent) deployment modes.