Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture, but in task management and agent workflow orchestration, developers often need to manually handle Git branch switching, task status tracking, and code merging. The dsh-kanban plugin integrates projects, Git branches, and AI agents into an automated pipeline: create task → agent processes it in an isolated branch → commit → manual review → automatically merge back to the base branch.

What Is It

This is a local project task kanban plugin designed for DeepSeek Harness (DSH). It runs on the client side of DSH, is maintained by zhouStar7, and uses the MIT license. The core value of the plugin is binding Git operations to agent task states, enabling end-to-end automation from task creation to code merging.

Core Features

The plugin deeply customizes the DSH interface and mainly includes the following capabilities:

  • Interface and Layout: Provides a custom sidebar with two tabs: “Sessions” and “Kanban,” while other slots in the official sidebar remain unchanged. The new session button is injected at the beginning of the workspace header action group. The kanban opens in the right main area, reserves 40px of top spacing, and does not use modal overlays.
  • Views and Toolbar: The left side of the toolbar includes creating new tasks and filtering projects, while the right side provides tab switching between “Kanban” and “Roadmap.” It supports two views: column view and Roadmap (Gantt).
  • Task Flow and State Machine: Task states include todo, running, review, approved, done, and paused. Agents can automatically claim tasks and work in isolated kanban/<id8> branches, including code commits.
  • Code Merging Process: After manual review, the code is merged back to the base branch using --no-ff, and the task branch is automatically deleted.
  • Interaction and Details: Paused agent sessions can be resumed by sending a comment. It supports model/time scheduling for new sessions, Git branch dropdown selection, and / path autocompletion. Each task retains a complete change log (final Agent output).

Installation and Activation

  1. Run the installation command:
dsh plugin --profile web add github:zhouStar7/dsh-kanban
  1. Restart or reload DSH. The sidebar will show “Sessions / Kanban” tabs.

Typical Usage

  1. Create a task in the kanban.
  2. The Agent automatically claims the task and creates a local isolated branch named kanban/<id8>.
  3. The Agent executes the task on the branch, writes code, and commits it.
  4. Manually enter review mode and mark the task status as approved.
  5. The plugin automatically performs the merge operation (--no-ff) and cleans up the branch, so the code is merged back to the base branch.

Use Cases and Notes

  • Use Cases: Suitable for developers who need to deeply integrate AI agent development workflows with Git version control. By managing task status with a kanban, combined with automatic agent claiming and branch isolation, it can significantly reduce the time spent manually switching branches and tracking task status.
  • Notes: The plugin has no telemetry. Its state data is stored in the local DSH storage domain and browser-side data. Git operations are limited to the scope of the selected project. The plugin uses the MIT license, and it is recommended to review the source code before installation.

Conclusion

dsh-kanban connects DSH’s agent capabilities with Git workflows through a kanban view, making it suitable for local developers who want to automate task management and code submission. For related resources, refer to the plugin catalog page or GitHub repository.