In the DeepSeek Harness (DSH) ecosystem, complex tasks are often plagued by context drift, lack of independent validation, and rework that fails to converge. The dsh-expert-team plugin solves these issues by building a role-based multi-agent team. It transforms fragmented agent collaboration into a structured delivery workflow, providing small and mid-sized teams and individual contractors with a complete “digital engineering department.”

What This Is

Maintained by yangdcm, the core goal of this plugin is to allow users to use a single natural-language objective to automatically assemble and direct a multi-agent expert team consisting of 12 roles. The team collaborates through a 9-stage gated pipeline; ultimately, implementers directly modify workspace code, and the entire process remains auditable as reviewable artifacts.

Core Features

  • 12-role multi-agent team: Includes roles such as product manager, architect, technical research, UI/UX, frontend/backend, data, security audit, code review, testing, DevOps, and documentation.
  • 9-stage gated pipeline: Progresses through clarification, research, design, specification review, plan confirmation, implementation, review, testing, and delivery to ensure orderly task execution.
  • Zero runtime dependencies and zero build steps: The plugin declares dependencies: {}, requires no build tools or install hooks, and is ready to use immediately after installation.
  • Direct workspace modification and process traceability: Implementers directly modify code in the user’s workspace, while all process artifacts (such as SPEC.md and PLAN.md) are saved in <workspace>/team/<run-id>/.
  • Quality gates enforced by code: The plugin code—not prompts—intercepts invalid operations, such as preventing incomplete tasks from being marked complete and requiring quality issues to be decided by designated roles.

Installation and Enablement

  1. Run the installation command in the terminal:
dsh plugin --profile web add @yangdcm/dsh-expert-team
  1. Restart dsh web. When the plugin loads, it automatically sets up the “Expert Team Mode” preset.

Typical Usage

  1. Switch to “Expert Team Mode”.
  2. Enter a command in the input box, for example:
    /team Build a payment module with login
  3. Use the persistent status bar above the input box (showing active roles) or /team canvas to open the full-screen canvas and view team status.
  4. Use /team status to check the current task status.

Use Cases and Considerations

  • Use cases: Internal tools and product iteration for small and mid-sized teams, independent contracting / freelance delivery, and complete projects built by independent developers.
  • Environment requirements: Requires DeepSeek Harness >= 0.1.5-rc.1 (web profile) and Node.js >= 20.
  • Configuration and restart: Some settings (such as the memory backend) require restarting dsh web after changes before they take effect.
  • Data storage: Process artifacts are saved in <your workspace>/team/<run-id>/, and local data is saved in $DSH_HOME/expert-team/.

With code-level quality gates and a structured delivery workflow, this plugin helps developers turn LLM capabilities into reviewable artifacts. For more details, see the GitHub repository.