NanmiCoder/dsh-agent-teams

AgentTeams plugin for DeepSeek Harness

What it solves

dsh-agent-teams is a plugin for DeepSeek Harness (DSH) that allows a single LLM session to act as a "captain" to coordinate a team of specialized sub-agents. It solves the problem of managing complex goals by splitting them into dependency-aware tasks and delegating them to durable, reusable agents who can communicate directly with each other and the captain.

How it works

  1. Captain-led Delegation: The current session becomes the captain and creates a team with role-specific members.
  2. Task Scheduling: The goal is broken down into tasks with explicit dependencies. A shared scheduler atomically assigns ready tasks to idle members.
  3. Durable Sub-agents: Members are backed by continuable sub-agents that can be woken up for focused follow-up turns.
  4. Communication: Members use a durable mailbox system to send direct messages to teammates or the captain without needing a relay.
  5. State Management: Team state is stored on disk, allowing the Web UI to provide a live activity panel showing roles, tasks, and team history.
  6. Agent Creation: By default, the plugin snapshots the captain's LLM provider and model to create members, though heterogeneous teams (using different models/providers) can be configured.

Who it’s for

Developers and users of DeepSeek Harness who need to coordinate multiple AI agents to solve complex, multi-step problems that require specialized roles and structured task management.

Highlights

  • Dependency-aware tasks: Tasks cannot be claimed until their dependencies are finished.
  • Live activity panel: A dedicated Web UI for monitoring team roles, current work, and task dependencies.
  • Automatic reuse: Idle members are automatically reused for ready tasks, with safe takeover mechanisms to handle interrupted turns.
  • Direct messaging: Durable mailbox messages between agents.
  • Zero-interaction member creation: Automatically mirrors the captain's model and provider settings for seamless setup.

Related

  • Dispatch
    DeepSeek Harness – Open‑Source Agent Framework with Plugin‑First Architecture

    DeepSeek Harness is an open‑source, plugin‑first agent framework that lets developers hot‑reload, dynamically enable, and cleanly dispose plugins, offering traceable execution and a web UI for building LLM‑driven agents.

  • Project
    Aisland-SJL/dsh-worktable

    An agent-project workbench for DeepSeek Harness that provides a dockable split-screen workspace and a real-time control room to monitor multiple AI agent projects.

  • Project
    0xsline/awesome-deepseek-harness

    A curated ecosystem of plugins, tools, and infrastructure for the DeepSeek Harness (DSH) to extend agent orchestration, memory, and context management.

  • Project
    deepseek-ai/deepseek-harness

    DeepSeek Harness is an open-source agent harness with a plugin-based architecture for building and managing AI agents.

  • Project
    xmanrui/dsh-im

    A plugin for DeepSeek Harness that connects AI agents to nine different IM platforms and provides a remote connection to a public AI Office for task management.