Skip to content
  • Home
  • Plugin manual
  • dsh-orchestrator

zibo2025/dsh-orchestrator

50Last commit Aug 14, 2026

dsh-orchestrator DSH plugin

dsh-orchestrator provides a master agent that decomposes tasks and dispatches them to multiple worker agents. Workers can communicate with each other natively via agent inbox, without file mailboxes or polling. Each worker can be configured with its own model and thinking effort.

skills-workflowsremote-execution

How to install the dsh-orchestrator DSH plugin

dsh plugin --profile <该名字> add dsh-orchestrator

This source command needs manual review. Copying does not run it.

dsh-orchestrator DSH plugin data source

dsh-orchestrator DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-orchestrator DSH plugin can do

  • Spawn persistent worker agents with custom provider, model, maxTokens, and effort (thinking intensity).
  • Send messages between any agents in the grid (master-to-worker, worker-to-master, worker-to-worker).
  • Broadcast a message to all online agents in the grid simultaneously.
  • List all agents in the grid, including labels, parent, and master flags.

Where the dsh-orchestrator DSH plugin fits

  • Decompose a complex research task into multiple sub-tasks, each handled by a specialized worker agent.
  • Create a fact-checking workflow: one worker researches a topic, another worker verifies the conclusions.
  • Parallel execution of independent tasks, with results aggregated by the master agent.

Who the dsh-orchestrator DSH plugin is for

  • Users of DeepSeek Harness who want to build multi-agent workflows.
  • Developers seeking a native orchestration layer without external dependencies.

dsh-orchestrator DSH plugin limitations

  • Turn-based: messages become the target agent's next turn and cannot interrupt ongoing calls.
  • Workers leave the registry after completion (agent_send returns 'not online' is expected); re-spawn if needed.
  • The grid is confined to the current session's agent tree; agents from other sessions are invisible.

dsh-orchestrator DSH plugin: from the repository README

Quoted from the zibo2025/dsh-orchestrator README, the upstream source of the dsh-orchestrator DSH plugin. Copyright remains with the original authors.

> **orchestrator** 意为「编排器 / 协调者」；读音：美 /ˈɔːrkəstreɪtər/ · 英 /ˈɔːkɪstreɪtə(r)/，重音在第三音节，近似「奥克斯**特雷**特尔」。 面向 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) 的原生全互联智能体编排插件： 一个主智能体负责任务分解与分派，多个 worker 负责实际执行，**任意智能体之间都能原生互发消息** （主 → 子、子 → 主、子 ↔ 子），并支持为每个 worker 单独指定模型与思考强度。 没有文件信箱、没有提示词约定、没有轮询——消息直接走 harness 自带的智能体收件箱。 ```mermaid flowchart LR U([用户需求]) --> M[主智能体<br/>任务分解与分派] subgraph G[worker 网格 · 任意智能体之间互通] W1[worker A<br/>独立模型与思考强度] W2[worker B<br/>独立模型与思考强度] W3[worker …] end M -- agent_spawn --> G G -- 回报 --> M M --> R([汇总结果给用户]) W1 <--> W2 W2 <--> W3 ``` ## 一键安装 把下面这句话发给你的 DSH 智能体，它会全自动完成安装： > 请自动安装 dsh-orchestrator 插件：运行 `ls $DSH_HOME/profiles` 找到我的 profile 名，然后执行 > `dsh plugin --profile <该名字> add dsh-orchestrator`；若失败，加 `--registry=https://registry.npmjs.org` > 重试一次；完成后提醒我重启 dsh。 装完**重启 `dsh`** 生效（运行中的进程不会热加载新安装的包）。 ## 手动安装 | 方式 | 命令 | | --- | --- | | npm（推荐） | `dsh plugin --profile <name> add dsh-orchestrator` | | GitHub 源码 | `dsh plugin --profile <name> add github:zibo2025/dsh-orchestrator#v0.1.3` | - `<name>` 是你的 profile 名，即 `$DSH_HOME/profiles/` 下的目录名（例如 `web`）。 - 国内镜像未同步时用官方源：`dsh plugin --pro

Read the full READMERepository license: MIT

dsh-orchestrator DSH plugin questions

How do I install dsh-orchestrator?

You can install it by sending the command to your DSH agent: 'dsh plugin --profile <name> add dsh-orchestrator'. Replace <name> with your profile name (e.g., 'web'). If the npm registry fails, retry with --registry=https://registry.npmjs.org. After installation, restart dsh for the changes to take effect.

How do I spawn a worker with a specific model?

Use the agent_spawn tool and specify the 'model' parameter along with provider, maxTokens, and effort. For example: agent_spawn with label 'researcher', model 'gpt-4', effort 'high'. The worker will be created with those settings.

Can workers communicate with each other?

Yes, workers can send messages to each other using the agent_send tool. This works both from worker to master and between workers. The message goes directly to the target agent's inbox and is processed in the next turn.

What does 'turn-based' mean in limitations?

It means that when you send a message to an agent, that message becomes the agent's next turn of input. The agent cannot interrupt its current ongoing call. This is a design constraint of the underlying DeepSeek Harness architecture.

Why is my worker not visible after it finishes?

Workers automatically leave the registry after they complete their task. If you need to communicate with them again, you must re-spawn them. Agent_send will return 'not online' if the target is no longer registered.