-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Agent System Overview
rUv edited this page Jul 25, 2025
·
1 revision
Claude-Flow v2 Alpha features a sophisticated 64-agent system designed for enterprise-grade AI orchestration. These specialized agents work together to create intelligent swarms capable of handling complex development tasks through coordinated collaboration.
- Total Agents: 64 specialized agents
- Categories: 12 distinct categories
- Directory Structure: 25 organized subdirectories
- Configuration Format: YAML frontmatter with markdown documentation
- Integration: Full MCP tool integration with 87+ available tools
All agents follow a standardized configuration format:
---
name: agent-name
type: agent-type
color: "#HEX_COLOR"
description: Brief description of agent purpose
capabilities:
- capability_1
- capability_2
- capability_3
priority: high|medium|low|critical
hooks:
pre: |
echo "Pre-execution commands"
post: |
echo "Post-execution commands"
---
# Agent Documentation
Detailed agent description and usage instructions| Type | Purpose | Examples |
|---|---|---|
coordinator |
Orchestrates other agents |
hierarchical-coordinator, mesh-coordinator
|
developer |
Code implementation |
coder, backend-dev
|
tester |
Testing and validation |
tester, production-validator
|
analyzer |
Analysis and optimization |
perf-analyzer, code-analyzer
|
security |
Security and compliance | security-manager |
synchronizer |
Data synchronization | crdt-synchronizer |
Location: .claude/agents/core/
Essential agents for basic development tasks:
| Agent | Type | Description | Priority |
|---|---|---|---|
coder |
developer | Implementation specialist for clean, efficient code | high |
reviewer |
reviewer | Code quality assurance and review specialist | high |
tester |
tester | Test creation and validation expert | high |
planner |
planner | Strategic planning and task orchestration | high |
researcher |
researcher | Information gathering and analysis specialist | high |
Usage Example:
# Deploy full development swarm
Task("Research requirements", "...", "researcher")
Task("Plan architecture", "...", "planner")
Task("Implement features", "...", "coder")
Task("Create tests", "...", "tester")
Task("Review code", "...", "reviewer")Location: .claude/agents/swarm/
Advanced coordination patterns for distributed agent networks:
| Agent | Type | Description | Topology |
|---|---|---|---|
hierarchical-coordinator |
coordinator | Queen-led coordination with specialized workers | Tree structure |
mesh-coordinator |
coordinator | Peer-to-peer networks with fault tolerance | Mesh network |
adaptive-coordinator |
coordinator | Dynamic topology switching based on workload | Adaptive hybrid |
Concurrent Swarm Deployment:
Task("Hierarchical coordination", "...", "hierarchical-coordinator")
Task("Mesh network backup", "...", "mesh-coordinator")
Task("Adaptive optimization", "...", "adaptive-coordinator")Location: .claude/agents/hive-mind/
Collective intelligence and shared decision-making:
| Agent | Type | Description | Capabilities |
|---|---|---|---|
collective-intelligence-coordinator |
coordinator | Shared memory and knowledge aggregation | decision-making, knowledge_aggregation |
consensus-builder |
coordinator | Byzantine fault-tolerant consensus mechanisms | consensus_algorithms, voting_systems |
swarm-memory-manager |
coordinator | Distributed memory coordination | memory_sync, context_sharing |
Location: .claude/agents/consensus/
Enterprise-grade distributed system coordination:
| Agent | Type | Description | Algorithm |
|---|---|---|---|
byzantine-coordinator |
coordinator | Byzantine fault tolerance with malicious actor detection | PBFT, HoneyBadger BFT |
raft-manager |
coordinator | Leader election and log replication | Raft consensus |
gossip-coordinator |
coordinator | Epidemic dissemination for eventual consistency | Gossip protocols |
security-manager |
security | Cryptographic security and attack detection | Threshold cryptography |
crdt-synchronizer |
synchronizer | Conflict-free replicated data types | State-based CRDTs |
performance-benchmarker |
analyst | Consensus protocol performance analysis | Benchmarking suites |
quorum-manager |
coordinator | Dynamic quorum adjustment and membership | Quorum algorithms |
Location: .claude/agents/optimization/
High-performance coordination and optimization:
| Agent | Type | Description | Optimization Type |
|---|---|---|---|
load-balancer |
coordinator | Work-stealing algorithms for task distribution | Load balancing |
performance-monitor |
monitor | Real-time metrics collection and bottleneck analysis | Performance monitoring |
topology-optimizer |
optimizer | Dynamic swarm topology reconfiguration | Topology optimization |
resource-allocator |
allocator | Adaptive resource allocation with ML prediction | Resource management |
benchmark-suite |
tester | Automated performance testing and regression detection | Performance testing |
Location: .claude/agents/github/
Complete GitHub workflow automation:
| Agent | Type | Description | GitHub Feature |
|---|---|---|---|
github-modes |
coordinator | Comprehensive GitHub integration modes | Multi-mode coordination |
pr-manager |
manager | Pull request lifecycle management | PR automation |
code-review-swarm |
reviewer | Multi-agent intelligent code reviews | Code review |
issue-tracker |
tracker | Issue management and project coordination | Issue tracking |
release-manager |
manager | Release coordination and deployment | Release management |
workflow-automation |
automation | CI/CD pipeline creation and optimization | GitHub Actions |
project-board-sync |
synchronizer | Project board synchronization | Project management |
repo-architect |
architect | Repository structure optimization | Repository design |
multi-repo-swarm |
coordinator | Cross-repository coordination | Multi-repo management |
release-swarm |
coordinator | Coordinated multi-package releases | Release orchestration |
swarm-issue |
coordinator | Issue-based swarm task coordination | Issue automation |
swarm-pr |
coordinator | PR-based swarm workflows | PR coordination |
sync-coordinator |
synchronizer | Multi-repository synchronization | Sync management |
Location: .claude/agents/sparc/
Test-driven development with SPARC methodology:
| Agent | Type | Description | SPARC Phase |
|---|---|---|---|
specification |
analyst | Requirements analysis and specification creation | Specification |
pseudocode |
designer | Algorithm design and pseudocode development | Pseudocode |
architecture |
architect | System architecture and design patterns | Architecture |
refinement |
refiner | Iterative improvement and optimization | Refinement |
Locations: Various specialized directories
Domain-specific development expertise:
| Agent | Type | Description | Specialization |
|---|---|---|---|
backend-dev |
developer | API development specialist | Backend/API |
mobile-dev |
developer | React Native mobile development | Mobile apps |
ml-developer |
developer | Machine learning model development | AI/ML |
cicd-engineer |
engineer | CI/CD pipeline creation | DevOps |
api-docs |
documenter | OpenAPI/Swagger documentation | API docs |
system-architect |
architect | High-level system design | Architecture |
code-analyzer |
analyzer | Advanced code quality analysis | Code quality |
base-template-generator |
generator | Boilerplate and template creation | Code generation |
Location: .claude/agents/testing/
Comprehensive testing and validation strategies:
| Agent | Type | Description | Testing Approach |
|---|---|---|---|
tdd-london-swarm |
tester | Mock-driven TDD with London School methodology | Unit testing |
production-validator |
validator | Real implementation validation for deployment | Integration testing |
Location: .claude/agents/templates/
Workflow templates and orchestration patterns:
| Agent | Type | Description | Purpose |
|---|---|---|---|
automation-smart-agent |
automation | Intelligent agent coordination | Smart automation |
coordinator-swarm-init |
coordinator | Swarm initialization and topology setup | Swarm setup |
github-pr-manager |
manager | GitHub PR management templates | PR templates |
implementer-sparc-coder |
implementer | SPARC implementation patterns | SPARC coding |
memory-coordinator |
coordinator | Cross-agent memory coordination | Memory management |
migration-plan |
planner | System migration planning | Migration |
orchestrator-task |
orchestrator | Complex task workflow coordination | Task orchestration |
performance-analyzer |
analyzer | Performance analysis templates | Performance |
sparc-coordinator |
coordinator | SPARC methodology coordination | SPARC orchestration |
Location: .claude/agents/analysis/, .claude/agents/architecture/
| Agent | Type | Description | Focus Area |
|---|---|---|---|
analyze-code-quality |
analyzer | Comprehensive code quality analysis | Code quality |
arch-system-design |
architect | System architecture design patterns | System design |
Locations: .claude/agents/data/, .claude/agents/devops/, .claude/agents/documentation/
| Agent | Type | Description | Domain |
|---|---|---|---|
data-ml-model |
developer | Machine learning model development | Data science |
ops-cicd-github |
engineer | GitHub-based CI/CD operations | DevOps |
docs-api-openapi |
documenter | OpenAPI specification generation | API documentation |
spec-mobile-react-native |
developer | React Native mobile development | Mobile development |
Claude-Flow is optimized for concurrent agent deployment. Always use multiple agents in a single message:
// ✅ CORRECT: Concurrent deployment
[Single Message]:
- Task("Agent 1", "full instructions", "agent-type-1")
- Task("Agent 2", "full instructions", "agent-type-2")
- Task("Agent 3", "full instructions", "agent-type-3")
- Task("Agent 4", "full instructions", "agent-type-4")
- Task("Agent 5", "full instructions", "agent-type-5")Task("System architecture", "...", "system-architect")
Task("Backend APIs", "...", "backend-dev")
Task("Frontend mobile", "...", "mobile-dev")
Task("Database design", "...", "coder")
Task("API documentation", "...", "api-docs")
Task("CI/CD pipeline", "...", "cicd-engineer")
Task("Performance testing", "...", "performance-benchmarker")
Task("Production validation", "...", "production-validator")Task("Byzantine consensus", "...", "byzantine-coordinator")
Task("Raft coordination", "...", "raft-manager")
Task("Gossip protocols", "...", "gossip-coordinator")
Task("CRDT synchronization", "...", "crdt-synchronizer")
Task("Security management", "...", "security-manager")
Task("Performance monitoring", "...", "performance-monitor")Task("PR management", "...", "pr-manager")
Task("Code review", "...", "code-review-swarm")
Task("Issue tracking", "...", "issue-tracker")
Task("Release coordination", "...", "release-manager")
Task("Workflow automation", "...", "workflow-automation")Task("Requirements spec", "...", "specification")
Task("Algorithm design", "...", "pseudocode")
Task("System architecture", "...", "architecture")
Task("TDD implementation", "...", "sparc-coder")
Task("London school tests", "...", "tdd-london-swarm")
Task("Iterative refinement", "...", "refinement")
Task("Production validation", "...", "production-validator")- High Priority: Use 3-5 agents max for critical path
- Medium Priority: Use 5-8 agents for complex features
- Large Projects: Use 8+ agents with proper coordination
- Use
memory-coordinatorfor cross-agent state - Implement
swarm-memory-managerfor distributed coordination - Apply
collective-intelligence-coordinatorfor decision-making
- Hierarchical: Best for large-scale projects with clear structure
- Mesh: Ideal for fault-tolerant, high-availability systems
- Adaptive: Perfect for dynamic workloads and optimization
All agents integrate with 87+ MCP tools including:
-
mcp__claude-flow__swarm_init- Swarm initialization -
mcp__claude-flow__agent_spawn- Agent creation -
mcp__claude-flow__task_orchestrate- Task coordination -
mcp__claude-flow__memory_usage- Memory management -
mcp__claude-flow__performance_report- Performance analytics
Agents support pre/post execution hooks for:
- Environment setup and validation
- Resource allocation and cleanup
- Performance monitoring and reporting
- Error handling and recovery
Native GitHub integration through specialized agents:
- Automated PR creation and management
- Intelligent code review workflows
- Issue tracking and board synchronization
- Release coordination and deployment
- Multi-repository management
# Use single agent
claude-flow agent use coder "implement user authentication"
# Use multiple agents concurrently
claude-flow swarm "build REST API" --agents coder,tester,reviewer# Initialize hierarchical swarm
claude-flow hive-mind spawn "build microservices" --topology hierarchical
# Use adaptive coordination
claude-flow swarm "optimize performance" --coordinator adaptive-coordinator# Full SPARC workflow
claude-flow sparc tdd "implement payment system" --agents specification,pseudocode,architecture,refinement- Always use concurrent agent deployment for maximum performance
- Match agent specialization to task requirements for optimal results
- Use coordination agents for complex multi-step workflows
- Apply appropriate topology based on project scale and requirements
- Leverage memory coordination for persistent state across agents
- Monitor performance using dedicated monitoring agents
- Validate production readiness with specialized validation agents
- Hive-Mind Intelligence - Deep dive into collective intelligence
- SPARC Methodology - Test-driven development patterns
- MCP Tools Reference - Complete tool documentation
- GitHub Integration - Repository management automation
- Performance Optimization - System optimization strategies
🚀 Enterprise-Grade AI Orchestration: 64 specialized agents working together to revolutionize development workflows through intelligent coordination and swarm intelligence.