agent-team: Multi-Agent Team Collaboration DSH plugin for DeepSeek Harness
limuyang2/agent-team
Build teams of independent agents in DeepSeek Harness, each with its own model, skills, and context, collaborating in a shared workspace.
Let each agent in DeepSeek Harness independently configure models, tools, and permissions, coordinated by a Leader over a shared workspace for specialized division of labor.
dsh plugin --profile web add github:limuyang2/agent-team- Category
- Skills & Agents
- Platform
- DSH-Plugin
- Author
- limuyang2
- Distribution
- Plugin
agent-team Key Features
agent-team Repository
agent-team Repo Summary
Agent Team is a team collaboration plugin (DSH plugin) for DeepSeek Harness, maintained by limuyang2, licensed under MIT, current version 0.1.2. It addresses the problem of a single agent handling all tasks with mixed models, tools, and contexts. By making each member an independent root agent with its own model, session, context, permissions, and tools, while sharing a common workspace, it enables specialized division of labor. Core capabilities include: independently configuring models and providers for each member, assigning Skills and MCP tools by role, isolating sessions and contexts, setting least-privilege permissions, and coordinating task assignment and result verification through a Leader.
What core features does the Agent Team DSH plugin support?
The core feature of Agent Team is building teams of independent agents: each member is an independent root agent, not inheriting parent configuration, and can freely choose different models and providers (e.g., GPT, GLM, DeepSeek Flash). Members have isolated sessions and context windows, preventing context bloat. The team collaborates through tasks, messages, and a shared workspace. The Leader is responsible for planning, assigning, and verifying, while members focus on execution. Additionally, each member can load Skills and MCP servers on demand and set runtime permissions, implementing least-privilege principles.
How to install the Agent Team DSH plugin?
Installing the Agent Team plugin is straightforward. Simply run the following command in your DeepSeek Harness environment:
dsh plugin --profile web add github:limuyang2/agent-team
After installation, the plugin automatically registers with DeepSeek Harness, and you can create teams and configure members in the interface.
How to configure teams and members in the Agent Team DSH plugin?
When configuring a team, you need to specify a Leader and several members. Each member can independently set model, Skills, MCP tools, permissions, and reasoning mode. For example, you can have an Architecture Leader use GPT, a Coding Agent use GLM with development tools, and a Commit Assistant use DeepSeek Flash with read-only permissions. Configuration is done through the DeepSeek Harness interface or configuration files; refer to the repository documentation for details.
Which DeepSeek Harness versions and platforms are compatible with the Agent Team DSH plugin?
Agent Team is built on the official DeepSeek Harness SDK, depending on packages like @deepseek-ai/dsh-agent and @deepseek-ai/dsh-client-runtime, so it is compatible with the DeepSeek Harness Web platform. It is written in TypeScript, installable via npm, with the current version 0.1.2, last updated on 2026-08-19.
What scenarios is the Agent Team DSH plugin suitable for?
Agent Team is suitable for complex tasks requiring multi-role collaboration, such as software development teams: architecture planning, coding implementation, and commit message generation. It allows each task to use the most appropriate model and tools, reducing cost and context confusion. If you want each agent to maintain independent configuration rather than inheriting parent settings, this plugin is ideal.