hive-mind-advanced
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseHive Mind Advanced Skill
Hive Mind高级技能
Master the advanced Hive Mind collective intelligence system for sophisticated multi-agent coordination using queen-led architecture, Byzantine consensus, and collective memory.
掌握高级Hive Mind集体智能系统,该系统采用女王主导架构、拜占庭共识与集体记忆,实现复杂的多Agent协调。
Overview
概述
The Hive Mind system represents the pinnacle of multi-agent coordination in Claude Flow, implementing a queen-led hierarchical architecture where a strategic queen coordinator directs specialized worker agents through collective decision-making and shared memory.
Hive Mind系统是Claude Flow中多Agent协调的顶尖方案,采用女王主导的分层架构,由战略型女王协调者通过集体决策与共享记忆指挥专业的工作Agent。
Core Concepts
核心概念
Architecture Patterns
架构模式
Queen-Led Coordination
- Strategic queen agents orchestrate high-level objectives
- Tactical queens manage mid-level execution
- Adaptive queens dynamically adjust strategies based on performance
Worker Specialization
- Researcher agents: Analysis and investigation
- Coder agents: Implementation and development
- Analyst agents: Data processing and metrics
- Tester agents: Quality assurance and validation
- Architect agents: System design and planning
- Reviewer agents: Code review and improvement
- Optimizer agents: Performance enhancement
- Documenter agents: Documentation generation
Collective Memory System
- Shared knowledge base across all agents
- LRU cache with memory pressure handling
- SQLite persistence with WAL mode
- Memory consolidation and association
- Access pattern tracking and optimization
女王主导协调
- 战略型女王Agent统筹高层目标
- 战术型女王管理中层执行
- 自适应型女王根据性能动态调整策略
工作Agent专业化
- 研究员Agent:分析与调研
- 编码Agent:实现与开发
- 分析师Agent:数据处理与指标分析
- 测试Agent:质量保证与验证
- 架构师Agent:系统设计与规划
- 评审Agent:代码评审与优化
- 优化师Agent:性能提升
- 文档师Agent:文档生成
集体记忆系统
- 所有Agent共享的知识库
- 带内存压力处理的LRU缓存
- 开启WAL模式的SQLite持久化
- 内存整合与关联
- 访问模式追踪与优化
Consensus Mechanisms
共识机制
Majority Consensus
Simple voting where the option with most votes wins.
Weighted Consensus
Queen vote counts as 3x weight, providing strategic guidance.
Byzantine Fault Tolerance
Requires 2/3 majority for decision approval, ensuring robust consensus even with faulty agents.
多数共识
简单投票制,得票最多的选项获胜。
加权共识
女王的投票权重为普通Agent的3倍,提供战略导向。
拜占庭容错
决策需获得2/3多数通过,即使存在故障Agent也能确保稳健共识。
Getting Started
快速开始
1. Initialize Hive Mind
1. 初始化Hive Mind
bash
undefinedbash
undefinedBasic initialization
基础初始化
npx claude-flow hive-mind init
npx claude-flow hive-mind init
Force reinitialize
强制重新初始化
npx claude-flow hive-mind init --force
npx claude-flow hive-mind init --force
Custom configuration
自定义配置
npx claude-flow hive-mind init --config hive-config.json
undefinednpx claude-flow hive-mind init --config hive-config.json
undefined2. Spawn a Swarm
2. 生成Agent集群
bash
undefinedbash
undefinedBasic spawn with objective
带目标的基础生成
npx claude-flow hive-mind spawn "Build microservices architecture"
npx claude-flow hive-mind spawn "Build microservices architecture"
Strategic queen type
战略型女王
npx claude-flow hive-mind spawn "Research AI patterns" --queen-type strategic
npx claude-flow hive-mind spawn "Research AI patterns" --queen-type strategic
Tactical queen with max workers
战术型女王+最大工作Agent数
npx claude-flow hive-mind spawn "Implement API" --queen-type tactical --max-workers 12
npx claude-flow hive-mind spawn "Implement API" --queen-type tactical --max-workers 12
Adaptive queen with consensus
自适应型女王+拜占庭共识
npx claude-flow hive-mind spawn "Optimize system" --queen-type adaptive --consensus byzantine
npx claude-flow hive-mind spawn "Optimize system" --queen-type adaptive --consensus byzantine
Generate Claude Code commands
生成Claude Code命令
npx claude-flow hive-mind spawn "Build full-stack app" --claude
undefinednpx claude-flow hive-mind spawn "Build full-stack app" --claude
undefined3. Monitor Status
3. 监控状态
bash
undefinedbash
undefinedCheck hive mind status
检查Hive Mind状态
npx claude-flow hive-mind status
npx claude-flow hive-mind status
Get detailed metrics
获取详细指标
npx claude-flow hive-mind metrics
npx claude-flow hive-mind metrics
Monitor collective memory
监控集体内存
npx claude-flow hive-mind memory
undefinednpx claude-flow hive-mind memory
undefinedAdvanced Workflows
高级工作流
Session Management
会话管理
Create and Manage Sessions
bash
undefined创建与管理会话
bash
undefinedList active sessions
列出活跃会话
npx claude-flow hive-mind sessions
npx claude-flow hive-mind sessions
Pause a session
暂停会话
npx claude-flow hive-mind pause <session-id>
npx claude-flow hive-mind pause <session-id>
Resume a paused session
恢复暂停的会话
npx claude-flow hive-mind resume <session-id>
npx claude-flow hive-mind resume <session-id>
Stop a running session
停止运行中的会话
npx claude-flow hive-mind stop <session-id>
**Session Features**
- Automatic checkpoint creation
- Progress tracking with completion percentages
- Parent-child process management
- Session logs with event tracking
- Export/import capabilitiesnpx claude-flow hive-mind stop <session-id>
**会话特性**
- 自动创建检查点
- 带完成百分比的进度追踪
- 父子进程管理
- 带事件追踪的会话日志
- 导出/导入功能Consensus Building
共识构建
The Hive Mind builds consensus through structured voting:
javascript
// Programmatic consensus building
const decision = await hiveMind.buildConsensus(
'Architecture pattern selection',
['microservices', 'monolith', 'serverless']
);
// Result includes:
// - decision: Winning option
// - confidence: Vote percentage
// - votes: Individual agent votesConsensus Algorithms
- Majority - Simple democratic voting
- Weighted - Queen has 3x voting power
- Byzantine - 2/3 supermajority required
Hive Mind通过结构化投票达成共识:
javascript
// 程序化共识构建
const decision = await hiveMind.buildConsensus(
'Architecture pattern selection',
['microservices', 'monolith', 'serverless']
);
// 返回结果包括:
// - decision: 胜出选项
// - confidence: 投票占比
// - votes: 单个Agent的投票情况共识算法
- 多数投票 - 简单民主投票
- 加权投票 - 女王拥有3倍投票权
- 拜占庭共识 - 需2/3超级多数通过
Collective Memory
集体记忆
Storing Knowledge
javascript
// Store in collective memory
await memory.store('api-patterns', {
rest: { pros: [...], cons: [...] },
graphql: { pros: [...], cons: [...] }
}, 'knowledge', { confidence: 0.95 });Memory Types
- : Permanent insights (no TTL)
knowledge - : Session context (1 hour TTL)
context - : Task-specific data (30 min TTL)
task - : Execution results (permanent, compressed)
result - : Error logs (24 hour TTL)
error - : Performance metrics (1 hour TTL)
metric - : Decision records (permanent)
consensus - : System configuration (permanent)
system
Searching and Retrieval
javascript
// Search memory by pattern
const results = await memory.search('api*', {
type: 'knowledge',
minConfidence: 0.8,
limit: 50
});
// Get related memories
const related = await memory.getRelated('api-patterns', 10);
// Build associations
await memory.associate('rest-api', 'authentication', 0.9);存储知识
javascript
// 存储到集体内存
await memory.store('api-patterns', {
rest: { pros: [...], cons: [...] },
graphql: { pros: [...], cons: [...] }
}, 'knowledge', { confidence: 0.95 });内存类型
- : 永久洞察(无过期时间)
knowledge - : 会话上下文(1小时过期)
context - : 任务特定数据(30分钟过期)
task - : 执行结果(永久存储,已压缩)
result - : 错误日志(24小时过期)
error - : 性能指标(1小时过期)
metric - : 决策记录(永久存储)
consensus - : 系统配置(永久存储)
system
搜索与检索
javascript
undefinedTask Distribution
按模式搜索内存
Automatic Worker Assignment
The system intelligently assigns tasks based on:
- Keyword matching with agent specialization
- Historical performance metrics
- Worker availability and load
- Task complexity analysis
javascript
// Create task (auto-assigned)
const task = await hiveMind.createTask(
'Implement user authentication',
priority: 8,
{ estimatedDuration: 30000 }
);Auto-Scaling
javascript
// Configure auto-scaling
const config = {
autoScale: true,
maxWorkers: 12,
scaleUpThreshold: 2, // Pending tasks per idle worker
scaleDownThreshold: 2 // Idle workers above pending tasks
};const results = await memory.search('api*', {
type: 'knowledge',
minConfidence: 0.8,
limit: 50
});
Integration Patterns
获取相关记忆
With Claude Code
—
Generate Claude Code spawn commands directly:
bash
npx claude-flow hive-mind spawn "Build REST API" --claudeOutput:
javascript
Task("Queen Coordinator", "Orchestrate REST API development...", "coordinator")
Task("Backend Developer", "Implement Express routes...", "backend-dev")
Task("Database Architect", "Design PostgreSQL schema...", "code-analyzer")
Task("Test Engineer", "Create Jest test suite...", "tester")const related = await memory.getRelated('api-patterns', 10);
With SPARC Methodology
建立关联
bash
undefinedawait memory.associate('rest-api', 'authentication', 0.9);
undefinedUse hive mind for SPARC workflow
任务分配
npx claude-flow sparc tdd "User authentication" --hive-mind
自动工作Agent分配
系统基于以下因素智能分配任务:
- 与Agent专长的关键词匹配
- 历史性能指标
- 工作Agent的可用性与负载
- 任务复杂度分析
javascript
undefinedSpawns:
创建任务(自动分配)
- Specification agent
—
- Architecture agent
—
- Coder agents
—
- Tester agents
—
- Reviewer agents
—
undefinedconst task = await hiveMind.createTask(
'Implement user authentication',
priority: 8,
{ estimatedDuration: 30000 }
);
**自动扩缩容**
```javascriptWith GitHub Integration
配置自动扩缩容
bash
undefinedconst config = {
autoScale: true,
maxWorkers: 12,
scaleUpThreshold: 2, // 每个空闲工作Agent对应的待处理任务数
scaleDownThreshold: 2 // 待处理任务数之上的空闲工作Agent数
};
undefinedRepository analysis with hive mind
集成模式
—
与Claude Code集成
npx claude-flow hive-mind spawn "Analyze repo quality" --objective "owner/repo"
直接生成Claude Code生成命令:
bash
npx claude-flow hive-mind spawn "Build REST API" --claude输出:
javascript
Task("Queen Coordinator", "Orchestrate REST API development...", "coordinator")
Task("Backend Developer", "Implement Express routes...", "backend-dev")
Task("Database Architect", "Design PostgreSQL schema...", "code-analyzer")
Task("Test Engineer", "Create Jest test suite...", "tester")PR review coordination
与SPARC方法论集成
npx claude-flow hive-mind spawn "Review PR #123" --queen-type tactical
undefinedbash
undefinedPerformance Optimization
使用Hive Mind执行SPARC工作流
Memory Optimization
—
The collective memory system includes advanced optimizations:
LRU Cache
- Configurable cache size (default: 1000 entries)
- Memory pressure handling (default: 50MB)
- Automatic eviction of least-used entries
Database Optimization
- WAL (Write-Ahead Logging) mode
- 64MB cache size
- 256MB memory mapping
- Prepared statements for common queries
- Automatic ANALYZE and OPTIMIZE
Object Pooling
- Query result pooling
- Memory entry pooling
- Reduced garbage collection pressure
npx claude-flow sparc tdd "User authentication" --hive-mind
Performance Metrics
生成以下Agent:
—
- 规格制定Agent
—
- 架构设计Agent
—
- 编码Agent
—
- 测试Agent
—
- 评审Agent
javascript
// Get performance insights
const insights = hiveMind.getPerformanceInsights();
// Includes:
// - asyncQueue utilization
// - Batch processing stats
// - Success rates
// - Average processing times
// - Memory efficiencyundefinedTask Execution
与GitHub集成
Parallel Processing
- Batch agent spawning (5 agents per batch)
- Concurrent task orchestration
- Async operation optimization
- Non-blocking task assignment
Benchmarks
- 10-20x faster batch spawning
- 2.8-4.4x speed improvement overall
- 32.3% token reduction
- 84.8% SWE-Bench solve rate
bash
undefinedConfiguration
使用Hive Mind分析仓库
Hive Mind Config
—
javascript
{
"objective": "Build microservices",
"name": "my-hive",
"queenType": "strategic", // strategic | tactical | adaptive
"maxWorkers": 8,
"consensusAlgorithm": "byzantine", // majority | weighted | byzantine
"autoScale": true,
"memorySize": 100, // MB
"taskTimeout": 60, // minutes
"encryption": false
}npx claude-flow hive-mind spawn "Analyze repo quality" --objective "owner/repo"
Memory Config
PR评审协调
javascript
{
"maxSize": 100, // MB
"compressionThreshold": 1024, // bytes
"gcInterval": 300000, // 5 minutes
"cacheSize": 1000,
"cacheMemoryMB": 50,
"enablePooling": true,
"enableAsyncOperations": true
}npx claude-flow hive-mind spawn "Review PR #123" --queen-type tactical
undefinedHooks Integration
性能优化
—
内存优化
Hive Mind integrates with Claude Flow hooks for automation:
Pre-Task Hooks
- Auto-assign agents by file type
- Validate objective complexity
- Optimize topology selection
- Cache search patterns
Post-Task Hooks
- Auto-format deliverables
- Train neural patterns
- Update collective memory
- Analyze performance bottlenecks
Session Hooks
- Generate session summaries
- Persist checkpoint data
- Track comprehensive metrics
- Restore execution context
集体记忆系统包含高级优化:
LRU缓存
- 可配置缓存大小(默认:1000条记录)
- 内存压力处理(默认:50MB)
- 自动淘汰最少使用的记录
数据库优化
- WAL(预写日志)模式
- 64MB缓存大小
- 256MB内存映射
- 常用查询的预编译语句
- 自动ANALYZE与OPTIMIZE
对象池化
- 查询结果池化
- 内存条目池化
- 降低垃圾回收压力
Best Practices
性能指标
1. Choose the Right Queen Type
—
Strategic Queens - For research, planning, and analysis
bash
npx claude-flow hive-mind spawn "Research ML frameworks" --queen-type strategicTactical Queens - For implementation and execution
bash
npx claude-flow hive-mind spawn "Build authentication" --queen-type tacticalAdaptive Queens - For optimization and dynamic tasks
bash
npx claude-flow hive-mind spawn "Optimize performance" --queen-type adaptivejavascript
undefined2. Leverage Consensus
获取性能洞察
Use consensus for critical decisions:
- Architecture pattern selection
- Technology stack choices
- Implementation approach
- Code review approval
- Release readiness
const insights = hiveMind.getPerformanceInsights();
3. Utilize Collective Memory
包含:
—
- asyncQueue利用率
—
批量处理统计
—
- 成功率
—
- 平均处理时间
—
- 内存效率
Store Learnings
javascript
// After successful pattern implementation
await memory.store('auth-pattern', {
approach: 'JWT with refresh tokens',
pros: ['Stateless', 'Scalable'],
cons: ['Token size', 'Revocation complexity'],
implementation: {...}
}, 'knowledge', { confidence: 0.95 });Build Associations
javascript
// Link related concepts
await memory.associate('jwt-auth', 'refresh-tokens', 0.9);
await memory.associate('jwt-auth', 'oauth2', 0.7);undefined4. Monitor Performance
任务执行
bash
undefined并行处理
- 批量生成Agent(每批5个)
- 并发任务编排
- 异步操作优化
- 非阻塞任务分配
基准测试
- 批量生成速度提升10-20倍
- 整体速度提升2.8-4.4倍
- Token使用量减少32.3%
- SWE-Bench解决率达84.8%
Regular status checks
配置
—
Hive Mind配置
npx claude-flow hive-mind status
javascript
{
"objective": "Build microservices",
"name": "my-hive",
"queenType": "strategic", // strategic | tactical | adaptive
"maxWorkers": 8,
"consensusAlgorithm": "byzantine", // majority | weighted | byzantine
"autoScale": true,
"memorySize": 100, // MB
"taskTimeout": 60, // minutes
"encryption": false
}Track metrics
内存配置
npx claude-flow hive-mind metrics
javascript
{
"maxSize": 100, // MB
"compressionThreshold": 1024, // bytes
"gcInterval": 300000, // 5 minutes
"cacheSize": 1000,
"cacheMemoryMB": 50,
"enablePooling": true,
"enableAsyncOperations": true
}Analyze memory usage
Hooks集成
npx claude-flow hive-mind memory
undefinedHive Mind与Claude Flow hooks集成以实现自动化:
任务前Hooks
- 按文件类型自动分配Agent
- 验证目标复杂度
- 优化拓扑选择
- 缓存搜索模式
任务后Hooks
- 自动格式化交付物
- 训练神经模式
- 更新集体记忆
- 分析性能瓶颈
会话Hooks
- 生成会话摘要
- 持久化检查点数据
- 追踪全面指标
- 恢复执行上下文
5. Session Management
最佳实践
—
1. 选择合适的女王类型
Checkpoint Frequently
javascript
// Create checkpoints at key milestones
await sessionManager.saveCheckpoint(
sessionId,
'api-routes-complete',
{ completedRoutes: [...], remaining: [...] }
);Resume Sessions
bash
undefined战略型女王 - 适用于研究、规划与分析
bash
npx claude-flow hive-mind spawn "Research ML frameworks" --queen-type strategic战术型女王 - 适用于实现与执行
bash
npx claude-flow hive-mind spawn "Build authentication" --queen-type tactical自适应型女王 - 适用于优化与动态任务
bash
npx claude-flow hive-mind spawn "Optimize performance" --queen-type adaptiveResume from any previous state
2. 利用共识机制
npx claude-flow hive-mind resume <session-id>
undefined在关键决策中使用共识:
- 架构模式选择
- 技术栈选型
- 实现方案
- 代码评审批准
- 发布就绪性
Troubleshooting
3. 利用集体记忆
Memory Issues
—
High Memory Usage
bash
undefined存储经验
javascript
undefinedRun garbage collection
成功实现模式后
npx claude-flow hive-mind memory --gc
await memory.store('auth-pattern', {
approach: 'JWT with refresh tokens',
pros: ['Stateless', 'Scalable'],
cons: ['Token size', 'Revocation complexity'],
implementation: {...}
}, 'knowledge', { confidence: 0.95 });
**建立关联**
```javascriptOptimize database
关联相关概念
npx claude-flow hive-mind memory --optimize
await memory.associate('jwt-auth', 'refresh-tokens', 0.9);
await memory.associate('jwt-auth', 'oauth2', 0.7);
undefinedExport and clear
4. 监控性能
npx claude-flow hive-mind memory --export --clear
**Low Cache Hit Rate**
```javascript
// Increase cache size in config
{
"cacheSize": 2000,
"cacheMemoryMB": 100
}bash
undefinedPerformance Issues
定期检查状态
Slow Task Assignment
javascript
// Enable worker type caching
// The system caches best worker matches for 5 minutes
// Automatic - no configuration neededHigh Queue Utilization
javascript
// Increase async queue concurrency
{
"asyncQueueConcurrency": 20 // Default: min(maxWorkers * 2, 20)
}npx claude-flow hive-mind status
Consensus Failures
追踪指标
No Consensus Reached (Byzantine)
bash
undefinednpx claude-flow hive-mind metrics
Switch to weighted consensus for more decisive results
分析内存使用
npx claude-flow hive-mind spawn "..." --consensus weighted
npx claude-flow hive-mind memory
undefinedOr use simple majority
5. 会话管理
npx claude-flow hive-mind spawn "..." --consensus majority
undefined频繁创建检查点
javascript
undefinedAdvanced Topics
在关键里程碑创建检查点
Custom Worker Types
—
Define specialized workers in :
.claude/agents/yaml
name: security-auditor
type: specialist
capabilities:
- vulnerability-scanning
- security-review
- penetration-testing
- compliance-checking
priority: highawait sessionManager.saveCheckpoint(
sessionId,
'api-routes-complete',
{ completedRoutes: [...], remaining: [...] }
);
**恢复会话**
```bashNeural Pattern Training
从任意之前的状态恢复
The system trains on successful patterns:
javascript
// Automatic pattern learning
// Happens after successful task completion
// Stores in collective memory
// Improves future task matchingnpx claude-flow hive-mind resume <session-id>
undefinedMulti-Hive Coordination
故障排除
—
内存问题
Run multiple hive minds simultaneously:
bash
undefined高内存占用
bash
undefinedFrontend hive
执行垃圾回收
npx claude-flow hive-mind spawn "Build UI" --name frontend-hive
npx claude-flow hive-mind memory --gc
Backend hive
优化数据库
npx claude-flow hive-mind spawn "Build API" --name backend-hive
npx claude-flow hive-mind memory --optimize
They share collective memory for coordination
导出并清理
undefinednpx claude-flow hive-mind memory --export --clear
**缓存命中率低**
```javascriptExport/Import Sessions
在配置中增大缓存大小
bash
undefined{
"cacheSize": 2000,
"cacheMemoryMB": 100
}
undefinedExport session for backup
性能问题
npx claude-flow hive-mind export <session-id> --output backup.json
任务分配缓慢
javascript
undefinedImport session
启用工作Agent类型缓存
—
系统会缓存最佳工作Agent匹配结果5分钟
—
自动生效 - 无需配置
npx claude-flow hive-mind import backup.json
undefined
**队列利用率高**
```javascriptAPI Reference
增大异步队列并发数
HiveMindCore
—
javascript
const hiveMind = new HiveMindCore({
objective: 'Build system',
queenType: 'strategic',
maxWorkers: 8,
consensusAlgorithm: 'byzantine'
});
await hiveMind.initialize();
await hiveMind.spawnQueen(queenData);
await hiveMind.spawnWorkers(['coder', 'tester']);
await hiveMind.createTask('Implement feature', 7);
const decision = await hiveMind.buildConsensus('topic', options);
const status = hiveMind.getStatus();
await hiveMind.shutdown();{
"asyncQueueConcurrency": 20 // 默认值: min(maxWorkers * 2, 20)
}
undefinedCollectiveMemory
共识失败
javascript
const memory = new CollectiveMemory({
swarmId: 'hive-123',
maxSize: 100,
cacheSize: 1000
});
await memory.store(key, value, type, metadata);
const data = await memory.retrieve(key);
const results = await memory.search(pattern, options);
const related = await memory.getRelated(key, limit);
await memory.associate(key1, key2, strength);
const stats = memory.getStatistics();
const analytics = memory.getAnalytics();
const health = await memory.healthCheck();未达成共识(拜占庭模式)
bash
undefinedHiveMindSessionManager
切换为加权共识以获得更明确的结果
javascript
const sessionManager = new HiveMindSessionManager();
const sessionId = await sessionManager.createSession(
swarmId, swarmName, objective, metadata
);
await sessionManager.saveCheckpoint(sessionId, name, data);
const sessions = await sessionManager.getActiveSessions();
const session = await sessionManager.getSession(sessionId);
await sessionManager.pauseSession(sessionId);
await sessionManager.resumeSession(sessionId);
await sessionManager.stopSession(sessionId);
await sessionManager.completeSession(sessionId);npx claude-flow hive-mind spawn "..." --consensus weighted
Examples
或使用简单多数投票
Full-Stack Development
—
bash
undefinednpx claude-flow hive-mind spawn "..." --consensus majority
undefinedInitialize hive mind
高级主题
—
自定义工作Agent类型
npx claude-flow hive-mind init
在中定义专业工作Agent:
.claude/agents/yaml
name: security-auditor
type: specialist
capabilities:
- vulnerability-scanning
- security-review
- penetration-testing
- compliance-checking
priority: highSpawn full-stack hive
神经模式训练
npx claude-flow hive-mind spawn "Build e-commerce platform"
--queen-type strategic
--max-workers 10
--consensus weighted
--claude
--queen-type strategic
--max-workers 10
--consensus weighted
--claude
系统会从成功模式中学习:
javascript
undefinedOutput generates Claude Code commands:
自动模式学习
- Queen coordinator
在任务成功完成后执行
- Frontend developers (React)
存储到集体内存
- Backend developers (Node.js)
提升未来任务匹配效率
- Database architects
—
- DevOps engineers
—
- Security auditors
—
- Test engineers
—
- Documentation specialists
—
undefinedundefinedResearch and Analysis
多Hive协调
bash
undefined同时运行多个Hive Mind:
bash
undefinedSpawn research hive
前端Hive
npx claude-flow hive-mind spawn "Research GraphQL vs REST"
--queen-type adaptive
--consensus byzantine
--queen-type adaptive
--consensus byzantine
npx claude-flow hive-mind spawn "Build UI" --name frontend-hive
Researchers gather data
后端Hive
Analysts process findings
—
Queen builds consensus on recommendation
—
Results stored in collective memory
—
undefinednpx claude-flow hive-mind spawn "Build API" --name backend-hive
Code Review
它们共享集体内存以实现协调
bash
undefinedundefinedReview coordination
导出/导入会话
npx claude-flow hive-mind spawn "Review PR #456"
--queen-type tactical
--max-workers 6
--queen-type tactical
--max-workers 6
bash
undefinedSpawns:
导出会话以备份
- Code analyzers
—
- Security reviewers
—
- Performance reviewers
—
- Test coverage analyzers
—
- Documentation reviewers
—
- Consensus on approval/changes
—
undefinednpx claude-flow hive-mind export <session-id> --output backup.json
Skill Progression
导入会话
Beginner
—
- Initialize hive mind
- Spawn basic swarms
- Monitor status
- Use majority consensus
npx claude-flow hive-mind import backup.json
undefinedIntermediate
API参考
—
HiveMindCore
- Configure queen types
- Implement session management
- Use weighted consensus
- Access collective memory
- Enable auto-scaling
javascript
const hiveMind = new HiveMindCore({
objective: 'Build system',
queenType: 'strategic',
maxWorkers: 8,
consensusAlgorithm: 'byzantine'
});
await hiveMind.initialize();
await hiveMind.spawnQueen(queenData);
await hiveMind.spawnWorkers(['coder', 'tester']);
await hiveMind.createTask('Implement feature', 7);
const decision = await hiveMind.buildConsensus('topic', options);
const status = hiveMind.getStatus();
await hiveMind.shutdown();Advanced
CollectiveMemory
- Byzantine fault tolerance
- Memory optimization
- Custom worker types
- Multi-hive coordination
- Neural pattern training
- Session export/import
- Performance tuning
javascript
const memory = new CollectiveMemory({
swarmId: 'hive-123',
maxSize: 100,
cacheSize: 1000
});
await memory.store(key, value, type, metadata);
const data = await memory.retrieve(key);
const results = await memory.search(pattern, options);
const related = await memory.getRelated(key, limit);
await memory.associate(key1, key2, strength);
const stats = memory.getStatistics();
const analytics = memory.getAnalytics();
const health = await memory.healthCheck();Related Skills
HiveMindSessionManager
- : Basic swarm coordination
swarm-orchestration - : Distributed decision making
consensus-mechanisms - : Advanced memory management
memory-systems - : Structured development workflow
sparc-methodology - : Repository coordination
github-integration
javascript
const sessionManager = new HiveMindSessionManager();
const sessionId = await sessionManager.createSession(
swarmId, swarmName, objective, metadata
);
await sessionManager.saveCheckpoint(sessionId, name, data);
const sessions = await sessionManager.getActiveSessions();
const session = await sessionManager.getSession(sessionId);
await sessionManager.pauseSession(sessionId);
await sessionManager.resumeSession(sessionId);
await sessionManager.stopSession(sessionId);
await sessionManager.completeSession(sessionId);References
示例
—
全栈开发
Skill Version: 1.0.0
Last Updated: 2025-10-19
Maintained By: Claude Flow Team
License: MIT
bash
undefined—
初始化Hive Mind
—
npx claude-flow hive-mind init
—
生成全栈Hive
—
npx claude-flow hive-mind spawn "Build e-commerce platform"
--queen-type strategic
--max-workers 10
--consensus weighted
--claude
--queen-type strategic
--max-workers 10
--consensus weighted
--claude
—
输出会生成Claude Code命令:
—
- 女王协调者
—
- 前端开发(React)
—
- 后端开发(Node.js)
—
- 数据库架构师
—
- DevOps工程师
—
- 安全审计员
—
- 测试工程师
—
- 文档专员
—
undefined—
研究与分析
—
bash
undefined—
生成研究Hive
—
npx claude-flow hive-mind spawn "Research GraphQL vs REST"
--queen-type adaptive
--consensus byzantine
--queen-type adaptive
--consensus byzantine
—
研究员收集数据
—
分析师处理研究结果
—
女王就建议达成共识
—
结果存储到集体内存
—
undefined—
代码评审
—
bash
undefined—
评审协调
—
npx claude-flow hive-mind spawn "Review PR #456"
--queen-type tactical
--max-workers 6
--queen-type tactical
--max-workers 6
—
生成以下Agent:
—
- 代码分析员
—
- 安全评审员
—
- 性能评审员
—
- 测试覆盖率分析员
—
- 文档评审员
—
- 就批准/修改达成共识
—
undefined—
技能进阶
—
初级
—
- 初始化Hive Mind
- 生成基础集群
- 监控状态
- 使用多数共识
—
中级
—
- 配置女王类型
- 实现会话管理
- 使用加权共识
- 访问集体内存
- 启用自动扩缩容
—
高级
—
- 拜占庭容错
- 内存优化
- 自定义工作Agent类型
- 多Hive协调
- 神经模式训练
- 会话导出/导入
- 性能调优
—
相关技能
—
- : 基础集群协调
swarm-orchestration - : 分布式决策
consensus-mechanisms - : 高级内存管理
memory-systems - : 结构化开发工作流
sparc-methodology - : 仓库协调
github-integration
—
参考资料
—