orchestrator

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Orchestrator - Automated Multi-Agent Coordinator

Orchestrator - 自动化多Agent协调器

When to use

适用场景

Complex feature requires multiple specialized agents working in parallel
User wants automated execution without manually spawning agents
Full-stack implementation spanning backend, frontend, mobile, and QA
User says "run it automatically", "run in parallel", or similar automation requests

复杂功能需要多个专业Agent并行工作
用户希望自动执行任务，无需手动启动Agent
涉及后端、前端、移动端和QA的全栈实现
用户提出“自动运行”、“并行运行”或类似自动化需求

When NOT to use

不适用场景

Simple single-domain task -> use the specific agent directly
User wants step-by-step manual control -> use workflow-guide
Quick bug fixes or minor changes

简单的单领域任务 → 直接使用特定Agent即可
用户希望分步手动控制 → 使用工作流向导（workflow-guide）
快速bug修复或小改动

Important

注意事项

This skill orchestrates CLI subagents via

gemini -p "..." --approval-mode=yolo

. It uses MCP Memory tools as a shared state bus. Each subagent runs as an independent process.

本技能通过

gemini -p "..." --approval-mode=yolo

编排CLI子Agent，使用MCP Memory工具作为共享状态总线。每个子Agent作为独立进程运行。

Configuration

配置

Setting	Default	Description
MAX_PARALLEL	3	Max concurrent subagents
MAX_RETRIES	2	Retry attempts per failed task
POLL_INTERVAL	30s	Status check interval
MAX_TURNS (impl)	20	Turn limit for backend/frontend/mobile
MAX_TURNS (review)	15	Turn limit for qa/debug
MAX_TURNS (plan)	10	Turn limit for pm

设置项	默认值	描述
MAX_PARALLEL	3	最大并发子Agent数量
MAX_RETRIES	2	每个失败任务的重试次数
POLL_INTERVAL	30s	状态检查间隔
MAX_TURNS (impl)	20	后端/前端/移动端的交互轮次上限
MAX_TURNS (review)	15	QA/调试的交互轮次上限
MAX_TURNS (plan)	10	产品经理（pm）的交互轮次上限

Memory Configuration

内存配置

Memory provider and tool names are configurable via

mcp.json

json

{
  "memoryConfig": {
    "provider": "serena",
    "basePath": ".serena/memories",
    "tools": {
      "read": "read_memory",
      "write": "write_memory",
      "edit": "edit_memory"
    }
  }
}

内存提供商和工具名称可通过

mcp.json

配置：

json

{
  "memoryConfig": {
    "provider": "serena",
    "basePath": ".serena/memories",
    "tools": {
      "read": "read_memory",
      "write": "write_memory",
      "edit": "edit_memory"
    }
  }
}

Workflow Phases

工作流阶段

PHASE 1 - Plan: Analyze request -> decompose tasks -> generate session ID PHASE 2 - Setup: Use memory write tool to create

orchestrator-session.md

task-board.md

PHASE 3 - Execute: Spawn agents by priority tier (never exceed MAX_PARALLEL) PHASE 4 - Monitor: Poll every POLL_INTERVAL; handle completed/failed/crashed agents PHASE 4.5 - Verify: Run

../_shared/verify.sh {agent-type} {workspace}

per completed agent PHASE 5 - Collect: Read all

result-{agent}.md

, compile summary, cleanup progress files

See

resources/subagent-prompt-template.md

for prompt construction. See

resources/memory-schema.md

for memory file formats.

阶段1 - 规划：分析请求→分解任务→生成会话ID 阶段2 - 准备：使用内存写入工具创建

orchestrator-session.md

和

task-board.md

阶段3 - 执行：按优先级层级启动Agent（绝不超过MAX_PARALLEL上限） 阶段4 - 监控：每隔POLL_INTERVAL轮询一次；处理已完成/失败/崩溃的Agent 阶段4.5 - 验证：针对每个已完成的Agent运行

../_shared/verify.sh {agent-type} {workspace}

阶段5 - 汇总：读取所有

result-{agent}.md

文件，编译汇总报告，清理进度文件

提示词构建可参考

resources/subagent-prompt-template.md

。内存文件格式可参考

resources/memory-schema.md

。

Memory File Ownership

内存文件归属

File	Owner	Others
`orchestrator-session.md`	orchestrator	read-only
`task-board.md`	orchestrator	read-only
`progress-{agent}.md`	that agent	orchestrator reads
`result-{agent}.md`	that agent	orchestrator reads

文件	归属方	其他角色权限
`orchestrator-session.md`	编排器	只读
`task-board.md`	编排器	只读
`progress-{agent}.md`	对应Agent	编排器可读取
`result-{agent}.md`	对应Agent	编排器可读取

Verification Gate (PHASE 4.5)

验证环节（阶段4.5）

After each agent completes, run automated verification before accepting the result:

bash

bash .agent/skills/_shared/verify.sh {agent-type} {workspace}

PASS (exit 0): Accept result, advance to next task
FAIL (exit 1): Treat as failure → enter Retry Logic with verify output as error context
This is mandatory. Never skip verification even if the agent reports success.

每个Agent完成任务后，需先运行自动化验证才能接受结果：

bash

bash .agent/skills/_shared/verify.sh {agent-type} {workspace}

通过（退出码0）：接受结果，推进至下一任务
失败（退出码1）：判定为任务失败→进入重试逻辑，将验证输出作为错误上下文
此环节为必填项，即使Agent报告成功也绝不能跳过验证

Retry Logic

重试逻辑

1st retry: Wait 30s, re-spawn with error context (include verify.sh output)
2nd retry: Wait 60s, add "Try a different approach"
Final failure: Report to user, ask whether to continue or abort

第1次重试：等待30秒，携带错误上下文重新启动Agent（包含verify.sh输出）
第2次重试：等待60秒，添加“尝试不同方法”提示
最终失败：向用户报告情况，询问是否继续或终止任务

Serena Memory (CLI Mode)

Serena内存（CLI模式）

See

../_shared/memory-protocol.md

详情请参考

../_shared/memory-protocol.md

。

References

参考资料

Prompt template:
```
resources/subagent-prompt-template.md
```
Memory schema:
```
resources/memory-schema.md
```
Config:
```
config/cli-config.yaml
```

Scripts:

scripts/spawn-agent.sh

scripts/parallel-run.sh

Task templates:
```
templates/
```
Skill routing:
```
../_shared/skill-routing.md
```
Verification:
```
../_shared/verify.sh
```
API contracts:
```
../_shared/api-contracts/
```
Context loading:
```
../_shared/context-loading.md
```
Difficulty guide:
```
../_shared/difficulty-guide.md
```
Reasoning templates:
```
../_shared/reasoning-templates.md
```
Clarification protocol:
```
../_shared/clarification-protocol.md
```
Context budget:
```
../_shared/context-budget.md
```
Lessons learned:
```
../_shared/lessons-learned.md
```

提示词模板：
```
resources/subagent-prompt-template.md
```
内存 schema：
```
resources/memory-schema.md
```
配置文件：
```
config/cli-config.yaml
```

脚本文件：

scripts/spawn-agent.sh

、

scripts/parallel-run.sh

任务模板：
```
templates/
```
技能路由：
```
../_shared/skill-routing.md
```
验证脚本：
```
../_shared/verify.sh
```
API 契约：
```
../_shared/api-contracts/
```
上下文加载：
```
../_shared/context-loading.md
```
难度指南：
```
../_shared/difficulty-guide.md
```
推理模板：
```
../_shared/reasoning-templates.md
```
澄清协议：
```
../_shared/clarification-protocol.md
```
上下文预算：
```
../_shared/context-budget.md
```
经验总结：
```
../_shared/lessons-learned.md
```