Preface

In the DSH ecosystem, plugins are the primary way to extend capabilities. For agent developers, a common question is: requirements usually start as a single sentence of natural language, but when they land in code, they need to be refined, broken down, executed, and merged, and every step’s state must be visible.

dsh-task-kanban is a DSH web plugin that adds a “Task Kanban” to the DSH web main interface: it refines requirements into multi-phase implementation plans, implements them serially in git worktree, and automatically merges them into the main branch after all phases are complete.

Below is an introduction to its positioning, core capabilities, installation method, and typical usage.

What is it

The npm package name for dsh-task-kanban is @fonlan/dsh-task-kanban, maintained by fonlan, with an MIT license.

Its core value is managing the “Requirement → Plan → Execution → Merge” chain within the DSH web main interface and displaying task status via swimlanes.

GitHub repository address:

https://github.com/fonlan/dsh-task-kanban

Core Features

Kanban Swimlanes

The task kanban visualizes cards across five swimlanes:

Requirement / Queue / Running / Completed / Merged

Cards display the following information:

  • Title
  • Requirement Summary
  • Model
  • Refinement Skill
  • plan status badge
  • Phase
  • Time

Cards also support Stop, Retry, and Delete.

Refining Requirements into Phase Plans

After creating a task, the plugin automatically enters the refinement session.

In this session, the requirement is decomposed into a phase plan for serial execution, known as plan. Subsequent execution revolves around these phases.

Skill-Driven Requirement Refinement

The plugin supports refining requirements via Skills.

Type / at the beginning of the requirement input box when creating a new task to pop up the list of available Skills; you can also directly type /skill-name to specify the Skill to use for refinement.

Example:

/grill-me help me plan…

The plugin injects the full instructions of that Skill into the refinement session and enters interactive refinement mode.

If the Skill is unknown or unavailable, a clear error will be given, and you can retry.

Queue and Serial Execution

Planned cards can be dragged into the queue and picked up for execution in FIFO order.

Execution constraints are as follows:

  • Each phase uses a separate session.
  • Each phase advances strictly in serial within the same git worktree.
  • The same workspace maintains single concurrency.
  • Cross-workspace concurrency is allowed based on settings.

Auto-Merge Main Branch

When all phases are complete, the plugin performs the following actions:

  1. Merge the plan branch into main or master.
  2. Clean up worktree.
  3. Clean up branches.

Plugin Settings

The settings entry is located at:

Sidebar Settings → Plugin Config → "Task Kanban"

Here you can configure:

  • Global parallel worker count.
  • Default model, reasoning intensity, and Agent presets for the requirement refinement session.
  • Default model, reasoning intensity, and Agent presets for the Phase implementation session.

Specific meanings are as follows:

  • Leaving the model empty means following the default path.
  • Reasoning intensity only lists the reasoning levels supported by the selected model.
  • Agent presets are fixed after the session starts.

Installation and Enablement

Install via Published Package

First, confirm the directory you want to install to. The directory running the install command will serve as the default workspace root directory.

Run in the target workspace root directory:

dsh plugin --profile web add @fonlan/dsh-task-kanban

Verified materials show the README provides an example for installing a specific version:

dsh plugin --profile web add @fonlan/dsh-task-kanban@0.1.0

However, within the same batch of verified materials, there is a conflict between the README installation instructions and the version description in package.json: the README states the current released version is 0.1.0, while the version in package.json is 0.2.3. Therefore, this article does not conclude on a specific version as the current released version based on these materials.

Install from Local Source

If installing from a local repository, you need to first build the lib/ artifacts.

Run in the repository directory:

pnpm build
dsh plugin --profile web add .

After installation, the plugin is mounted to the web profile via cordis.patch.yml. Usually, after restarting the web profile, the “Task Kanban” entry will appear.

Typical Usage

Below is a reproducible usage path.

1. Create a Task

Create a task in the task kanban and enter the requirement.

If no Skill is specified, the system will enter a normal refinement session.

2. Specify Refinement Skill

If you want to use a specific Skill to refine the requirement, type / at the beginning of the requirement input box and select a Skill; you can also directly type:

/skill-name requirement description

For example:

/grill-me help me plan…

The plugin will inject the full instructions of that Skill and enter interactive refinement mode.

3. Put Planned Cards into Queue

After the requirement refinement is complete, drag the planned cards into the queue.

The queue will pick up execution in FIFO order.

4. View Execution Status

During execution, you can observe card status changes in the kanban:

Requirement / Queue / Running / Completed / Merged

On the card, you can see the Model, Refinement Skill, plan status badge, Phase, and Time, and it supports Stop, Retry, and Delete.

5. Wait for Auto-Merge

After all phases are complete, the plugin will automatically merge the plan branch into main or master and clean up the worktree and branches.

6. Adjust Session Defaults

Enter:

Sidebar Settings → Plugin Config → "Task Kanban"

You can configure the global parallel worker count, as well as the default model, reasoning intensity, and Agent presets for the two types of sessions:

  • Requirement Refinement Session
  • Phase Implementation Session

Suitable Scenarios and Notes

Suitable for Who

This plugin is quite suitable for the following scenarios:

  • Using the DSH web main interface to manage agent tasks.
  • Wanting to split natural language requirements into multi-phase plans.
  • Wanting to execute serially in git worktree and automatically merge into the main branch upon completion.
  • Wanting to track the “Requirement / Queue / Running / Completed / Merged” status in the kanban.

Usage Notes

Before use, pay attention to the following points:

  • The plugin runs with the permissions of the current DSH process; check the source code and license before installing.
  • The plugin will automatically merge into main or master; confirm your branch strategy before execution.
  • The same workspace maintains single concurrency; cross-workspace parallelism is constrained by settings.
  • Leaving the model empty means following the default path.
  • Reasoning intensity only lists the reasoning levels supported by the selected model.
  • Agent presets are fixed after the session starts.
  • Unknown or unavailable Skills will give a clear error and can be retried.
  • The web profile needs to be restarted after modifying the server side.
  • Client-side changes go through the web HMR path and take effect when pnpm run dev:web is started.

Conclusion

The value of dsh-task-kanban lies in putting requirement refinement, phase execution, status tracking, and auto-merge into the same DSH web kanban, reducing the overhead of manually switching and tracking across multiple sessions.

Repository address:

https://github.com/fonlan/dsh-task-kanban

If you need to view the plugin directory, refer to the corresponding DSH plugin directory site; the DSH plugin directory is a community directory site and should not be understood as the official app store of DeepSeek or Huanshan.