task-distributor

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Task Distributor

任务分配器

Purpose

用途

Provides expertise in distributing tasks across multi-agent systems efficiently. Specializes in load balancing algorithms, capability-based routing, cost optimization, and ensuring optimal resource utilization across distributed agent pools.

提供多Agent系统任务高效分配的专业能力，专注于负载均衡算法、基于能力的路由、成本优化，以及确保分布式Agent池的资源最优利用。

When to Use

适用场景

Designing task distribution strategies for multi-agent systems
Implementing load balancing across worker pools
Optimizing for cost (token economics) vs speed trade-offs
Building routing logic based on agent capabilities
Managing task queues with priorities and deadlines
Implementing retry and failover strategies
Scaling agent pools dynamically based on demand
Monitoring and optimizing task throughput

为多Agent系统设计任务分配策略
在工作者池中实现负载均衡
优化成本（Token Economics）与速度的权衡
构建基于Agent能力的路由逻辑
管理带优先级和截止期限的任务队列
实现重试与故障转移策略
根据需求动态扩展Agent池
监控并优化任务吞吐量

Quick Start

快速开始

Invoke this skill when:

Designing task distribution strategies for multi-agent systems
Implementing load balancing across worker pools
Optimizing for cost (token economics) vs speed trade-offs
Building routing logic based on agent capabilities
Managing task queues with priorities and deadlines

Do NOT invoke when:

Designing overall agent architecture → use agent-organizer
Implementing individual agent logic → use appropriate domain skill
Handling agent errors and recovery → use error-coordinator
Building workflow orchestration → use workflow-orchestrator

在以下场景调用此技能：

为多Agent系统设计任务分配策略
在工作者池中实现负载均衡
优化成本（Token Economics）与速度的权衡
构建基于Agent能力的路由逻辑
管理带优先级和截止期限的任务队列

请勿在以下场景调用：

设计整体Agent架构 → 使用agent-organizer
实现单个Agent逻辑 → 使用对应领域技能
处理Agent错误与恢复 → 使用error-coordinator
构建工作流编排 → 使用workflow-orchestrator

Decision Framework

决策框架

Distribution Strategy?
├── Uniform Workloads → Round-robin or random distribution
├── Variable Task Complexity → Weighted distribution by capability
├── Cost Sensitive → Route to cheapest capable agent
├── Latency Sensitive → Route to fastest/nearest agent
├── Specialized Tasks → Capability-based routing
└── Burst Traffic → Dynamic scaling + queue management

分配策略？
├── 均匀工作负载 → 轮询或随机分配
├── 可变任务复杂度 → 基于能力的加权分配
├── 成本敏感型 → 路由至成本最低的合格Agent
├── 延迟敏感型 → 路由至最快/最近的Agent
├── 专业化任务 → 基于能力的路由
└── 突发流量 → 动态扩展 + 队列管理

Core Workflows

核心工作流

1. Capability-Based Routing

1. 基于能力的路由

Define capability taxonomy for agents
Tag tasks with required capabilities
Implement capability matching algorithm
Score agents by capability fit and availability
Route to best-matched agent
Track capability utilization for optimization
Adjust routing weights based on performance

定义Agent的能力分类体系
为任务标记所需能力
实现能力匹配算法
根据能力适配度和可用性为Agent评分
路由至最佳匹配的Agent
跟踪能力利用率以优化
根据性能调整路由权重

2. Cost-Optimized Distribution

2. 成本优化分配

Define cost model per agent type (tokens, time, money)
Estimate task cost based on complexity signals
Set budget constraints and optimization targets
Route to minimize cost while meeting SLAs
Implement fallback to higher-cost agents when needed
Track actual vs estimated costs
Refine cost models from historical data

定义每种Agent类型的成本模型（代币、时间、资金）
根据复杂度信号估算任务成本
设置预算约束和优化目标
在满足SLA的前提下最小化成本
必要时实现向更高成本Agent的降级方案
跟踪实际成本与估算成本的差异
根据历史数据优化成本模型

3. Queue Management with Priorities

3. 带优先级的队列管理

Define priority levels and SLA requirements
Implement priority queue with deadline awareness
Set up work stealing for idle agents
Handle starvation of low-priority tasks
Implement backpressure when queue depth exceeds threshold
Monitor queue latency and throughput
Scale agent pool based on queue metrics

定义优先级等级和SLA要求
实现支持截止期限感知的优先级队列
为空闲Agent设置任务窃取机制
处理低优先级任务的饥饿问题
当队列深度超过阈值时实现背压机制
监控队列延迟和吞吐量
根据队列指标扩展Agent池

Best Practices

最佳实践

Implement health checks and remove unhealthy agents from pool
Use exponential backoff with jitter for retries
Track per-agent metrics for informed routing decisions
Implement circuit breakers for failing agent types
Design for graceful degradation under load
Make routing decisions observable for debugging

实现健康检查，将不健康的Agent从池中移除
重试时使用带抖动的指数退避策略
跟踪每个Agent的指标以辅助路由决策
为故障Agent类型实现断路器
设计负载下的优雅降级机制
使路由决策可观测以便调试

Anti-Patterns

反模式

Static assignment → Use dynamic routing based on current state
Ignoring agent health → Remove unhealthy agents from rotation
FIFO only → Implement priority awareness for SLA compliance
Tight coupling → Decouple task producers from agent pool
No backpressure → Implement admission control under overload

静态分配 → 基于当前状态使用动态路由
忽略Agent健康 → 将不健康的Agent从轮换中移除
仅使用FIFO → 实现优先级感知以符合SLA
紧耦合 → 解耦任务生产者与Agent池
无背压机制 → 在过载时实现准入控制