limuyang2/agent-team
Runs a team of independent root agents in DSH, each with its own model, context and permissions, coordinated by one Leader.
Agent Team builds teams of independent AI agents inside DeepSeek Harness. The design point, argued at length in the README, is that members are NOT subagents: each is an independent root agent with its own model, session, context, permissions, reasoning mode and tool activity, and only the team tasks, messages and shared Workspace form the collaboration layer. The README's stated reason is cost and focus: a parent/subagent workflow reuses much of the parent runtime, so a small commit-message task can still run on the expensive model with a broad tool catalog and a growing context, whereas a team can give each role only the Skills and MCP servers it needs. Documented example teams assign a strong model as Architecture Leader to design and verify, and a fast model as Commit Assistant with read-only permission to read git status and diffs and write a Conventional Commit message. Members work in their own conversations while sharing the same Workspace.
Install
dsh plugin --profile web add @limuyang2/dsh-agent-teamPublished on npm as @limuyang2/dsh-agent-team, currently 0.1.4, which is the line the README documents. Requires pnpm on PATH because Harness uses it to manage profile plugins.
Compatibility
Requires Node.js 22.19.0+ or 24.0.0+, and DeepSeek Harness 0.1.1-rc.2. A built-in assistant drafts each member's long-term instructions from a plain description of the role, collecting missing settings first and only creating the assistant after you confirm.
Details
- Repo: limuyang2/agent-team
- Category: Agent Capabilities
- Stars: 46
- Version: 0.1.4 (npm)
Alternatives
Contexera/dsh-agent-team · xieshuaix/dsh-agent-swarm · markelayan/agents-in-the-loop