Fleet Manager Skill
Version
Version: 1.0.0 Status: OPERATIONAL
Overview
Fleet Manager provides comprehensive resource allocation, tier-based resource management, performance monitoring, and load balancing for distributed agent swarms.
Features
- Agent Registration with Resource Tiers
- Dynamic Resource Allocation
- Real-time Performance Metrics
- Intelligent Load Balancing
- Resource Quota Management
- Health Monitoring
Resource Tiers
Shared Tier (Default)
- CPU: 0.5 cores
- Memory: 512 MB
- Use Case: Lightweight tasks, research, documentation
- Cost: Minimal
Dedicated Tier
- CPU: 2.0 cores
- Memory: 2048 MB
- Use Case: Backend development, testing, code analysis
- Cost: Standard
Premium Tier
- CPU: 4.0 cores
- Memory: 4096 MB
- Use Case: CFN Loop validation, compilation, heavy processing
- Cost: Maximum
Agent Integration Examples
1. Register Agent with Resource Tier
# Register agent with shared tier (default)
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "researcher-1" \
--tier shared
# Register agent with dedicated tier
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "backend-dev-1" \
--tier dedicated
# Register agent with premium tier
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "cfn-validator-1" \
--tier premium
Output:
{
"status": "registered",
"agentId": "backend-dev-1",
"tier": "dedicated",
"resources": {
"cpu": 2.0,
"memory": 2048
},
"timestamp": 1729332000
}
2. Allocate Resources to Agent
# Allocate custom resources
./.claude/skills/fleet-manager/invoke-fleet-allocate.sh \
--agent-id "backend-dev-1" \
--cpu 3.0 \
--memory 3072
# Allocate with priority flag
./.claude/skills/fleet-manager/invoke-fleet-allocate.sh \
--agent-id "cfn-validator-1" \
--cpu 4.0 \
--memory 4096 \
--priority high
Output:
{
"status": "allocated",
"agentId": "backend-dev-1",
"allocated": {
"cpu": 3.0,
"memory": 3072
},
"availability": {
"cpuRemaining": 5.0,
"memoryRemaining": 4096
},
"timestamp": 1729332100
}
3. Get Performance Metrics
# Get metrics for specific agent
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--agent-id "backend-dev-1"
# Get metrics with detailed breakdown
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--agent-id "backend-dev-1" \
--detailed
# Get fleet-wide metrics
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--all
Output:
{
"agentId": "backend-dev-1",
"metrics": {
"cpuUsage": 2.4,
"cpuUtilization": 0.80,
"memoryUsage": 1800,
"memoryUtilization": 0.88,
"taskCompletionRate": 0.95,
"averageResponseTime": 120
},
"health": "healthy",
"timestamp": 1729332200
}
4. Trigger Load Balancing
# Balance resources across multiple agents
./.claude/skills/fleet-manager/invoke-fleet-balance.sh \
--agents "backend-dev-1,backend-dev-2,backend-dev-3"
# Balance with strategy
./.claude/skills/fleet-manager/invoke-fleet-balance.sh \
--agents "cfn-validator-1,cfn-validator-2" \
--strategy round-robin
# Balance entire fleet
./.claude/skills/fleet-manager/invoke-fleet-balance.sh \
--all
Output:
{
"status": "balanced",
"strategy": "round-robin",
"agents": [
{
"agentId": "backend-dev-1",
"beforeLoad": 0.80,
"afterLoad": 0.65,
"adjustment": "reduced"
},
{
"agentId": "backend-dev-2",
"beforeLoad": 0.45,
"afterLoad": 0.60,
"adjustment": "increased"
}
],
"timestamp": 1729332300
}
Integration Patterns
Pattern 1: CFN Loop with Resource Management
# Register all CFN Loop agents with appropriate tiers
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "coder-1" --tier dedicated
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "reviewer-1" --tier dedicated
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "product-owner-1" --tier premium
# Monitor performance during loop
metrics=$(
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--agent-id "coder-1"
)
# If CPU utilization high, trigger rebalancing
cpu_util=$(echo "$metrics" | jq '.metrics.cpuUtilization')
if (( $(echo "$cpu_util > 0.85" | bc -l) )); then
./.claude/skills/fleet-manager/invoke-fleet-balance.sh \
--agents "coder-1,coder-2,coder-3"
fi
Pattern 2: Dynamic Resource Scaling
# Start with shared tier
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "researcher-1" --tier shared
# If task complexity increases, allocate more resources
./.claude/skills/fleet-manager/invoke-fleet-allocate.sh \
--agent-id "researcher-1" \
--cpu 2.0 \
--memory 2048
# Monitor and adjust
metrics=$(
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--agent-id "researcher-1"
)
memory_util=$(echo "$metrics" | jq '.metrics.memoryUtilization')
if (( $(echo "$memory_util > 0.90" | bc -l) )); then
./.claude/skills/fleet-manager/invoke-fleet-allocate.sh \
--agent-id "researcher-1" \
--memory 4096 \
--priority high
fi
Pattern 3: Health-Based Resource Adjustment
# Check agent health
metrics=$(
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--agent-id "backend-dev-1"
)
health=$(echo "$metrics" | jq -r '.health')
if [ "$health" = "degraded" ]; then
# Reduce load on degraded agent
./.claude/skills/fleet-manager/invoke-fleet-balance.sh \
--agents "backend-dev-1,backend-dev-2" \
--strategy offload
fi
Performance Characteristics
Resource Allocation Latency
- Registration: <50ms
- Allocation: <100ms
- Metrics Retrieval: <30ms
- Load Balancing: <200ms
Scalability
- Max Concurrent Agents: 100+
- Max Resource Pool: 400 cores, 400GB RAM
- Metrics Collection Rate: 1Hz per agent
- Balancing Frequency: On-demand or scheduled
Resource Efficiency
- Shared Tier: 95% resource utilization
- Dedicated Tier: 85% resource utilization
- Premium Tier: 75% resource utilization (reserved headroom)
Redis Key Structure
# Agent Registration
fleet:agent:{agentId}:registration -> {tier, cpu, memory, timestamp}
# Resource Allocation
fleet:agent:{agentId}:allocation -> {cpu, memory, priority, timestamp}
# Performance Metrics
fleet:agent:{agentId}:metrics -> {cpuUsage, memoryUsage, health, timestamp}
# Load Balancing State
fleet:balance:{balanceId}:state -> {agents, strategy, status, timestamp}
# Resource Pool
fleet:pool:available -> {totalCpu, totalMemory, allocated, reserved}
Error Handling
Common Errors
Insufficient Resources:
{
"error": "RESOURCE_UNAVAILABLE",
"message": "Requested resources exceed available pool",
"requested": {"cpu": 4.0, "memory": 4096},
"available": {"cpu": 2.0, "memory": 2048}
}
Agent Not Found:
{
"error": "AGENT_NOT_FOUND",
"message": "Agent backend-dev-1 not registered",
"suggestion": "Register agent first using invoke-fleet-register.sh"
}
Invalid Tier:
{
"error": "INVALID_TIER",
"message": "Tier 'ultra' not recognized",
"validTiers": ["shared", "dedicated", "premium"]
}
Best Practices
- Register Before Allocation: Always register agents before allocating custom resources
- Monitor Regularly: Check metrics at least once per iteration for long-running tasks
- Balance Proactively: Trigger load balancing when fleet-wide utilization >70%
- Use Appropriate Tiers: Match tier to task complexity to optimize cost
- Graceful Degradation: Handle resource scarcity by queuing tasks or reducing quality
Testing
Run comprehensive test suite:
./.claude/skills/fleet-manager/test-fleet-manager.sh
Test Coverage
- Agent registration (all tiers)
- Resource allocation (standard and high-priority)
- Metrics collection (individual and fleet-wide)
- Load balancing (multiple strategies)
- Error handling (resource exhaustion, invalid inputs)
- Concurrency (parallel agent operations)
Coordination with Other Skills
Redis Coordination Integration
# Register CFN Loop agents with Fleet Manager
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "coder-1" --tier dedicated
# Start CFN Loop with orchestration
./.claude/skills/redis-coordination/orchestrate-cfn-loop.sh \
--task-id "auth-system" \
--loop3-agents "coder-1,coder-2" \
--loop2-agents "reviewer-1" \
--product-owner "product-owner-1"
# Monitor resource usage during loop
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh \
--agent-id "coder-1"
Agent Spawning Integration
# Register agent in spawning script
./.claude/skills/fleet-manager/invoke-fleet-register.sh \
--agent-id "$AGENT_ID" --tier dedicated
# Spawn agent with Task tool
# ... agent spawning logic ...
# Deallocate resources when agent completes
./.claude/skills/fleet-manager/invoke-fleet-allocate.sh \
--agent-id "$AGENT_ID" --cpu 0 --memory 0
Maintenance
Daily Health Checks
# Check fleet-wide health
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh --all | jq '.agents[] | select(.health != "healthy")'
Weekly Resource Audits
# Identify underutilized agents
./.claude/skills/fleet-manager/invoke-fleet-metrics.sh --all | jq '.agents[] | select(.metrics.cpuUtilization < 0.20)'
Monthly Capacity Planning
# Analyze resource trends
# Export metrics to analytics system for trend analysis
Version History
v1.0.0 (2025-10-19)
- Initial release
- Agent registration with 3-tier resource model
- Dynamic resource allocation
- Real-time performance metrics
- Load balancing with multiple strategies
- Comprehensive test suite