Introduction

When using agents in DSH, requirements understanding, solution presentation, delivery acceptance, and token costs usually require extra constraints. dsh-ui-three-body is a DeepSeek Harness (DSH) plugin that injects the “Open Wisdom” (Kai Zhi) kernel into agents, providing a floating UI and toggles in the settings page. Below is an introduction to its core capabilities, installation methods, and usage notes.

What is this

dsh-ui-three-body is a DSH plugin maintained by EternalNight996, licensed under MIT. It calls itself the “Tamer (Taming Ground)”: translating human language for agents.

The plugin injects the “Open Wisdom” kernel, covering first principles, requirement analysis, minimal communication, and minimal tokens. The README states no modification of DSH source code, zero third-party memory frameworks, and one-command installation; runtime environment requires Node >=22.

Core Features

Open Wisdom Kernel and Five Strategies

The plugin provides the “Open Wisdom” kernel, emphasizing first principles, requirement analysis, minimal communication, and minimal tokens. The Five Strategies of the Schemer are:

  1. Expand · Clarify
  2. Diminish Dimensions · Propose Solutions
  3. Face the Wall · Contract
  4. Water Drop · Execute
  5. Gather · Deliver

Scale Routing

Provides S/M/L scale routing, determining the level through three questions, and supports upgrading/downgrading on the fly.

Quality Gate

The quality gate includes:

  • Real Verification Loop
  • Acceptance = Executable Assertions
  • Self-evaluation meets standards
  • Human review + rollback + stop if outside contract

UI Preview Before Delivery

UI preview before delivery supports 4 templates / 1 template / markdown only, and provides HTML templates and design tokens.

Floating Schemer

The floating schemer provides 11 dynamic skins, supporting mouse-tracking eyes, random blinking, ghost mode, progress display above the head, and Three-Body dialogue.

Settings and Toggles

After installation, go to “Settings → Three-Body” to configure:

  • Schemer Activity
  • AI Mode
  • Floating Schemer
  • Size
  • Skin
  • Ghost Mode
  • Target Display
  • Three-Body Dialogue
  • Kernel Level
  • Language
  • Tone
  • Self-reference
  • Reference to User
  • Requirement Analysis Tool
  • Kernel Override

AI mode acts as the master token switch; when turned off, it results in zero kernel tokens per round. Kernel levels support minimal / balanced / full. When the requirement analysis tool beast_analyze is enabled, an extra model request is made for each call.

Installation and Enablement

First, install the DSH CLI:

npm i -g @deepseek-ai/dsh

Then install the plugin. npm method:

dsh plugin --profile web add dsh-ui-three-body

If you want to use the npm official registry:

npm_config_registry=https://registry.npmjs.org/ dsh plugin --profile web add dsh-ui-three-body

If you want to use a GitHub source:

dsh plugin --profile web add github:EternalNight996/dsh-ui-three-body

If you want to use a local path:

dsh plugin --profile web add F:/MyApp/eternal/dsh-ui-three-body

The README suggests using pnpm workspace to manage the profile and warns that npm install might error with EUNSUPPORTEDPROTOCOL. If you operate within the profile directory, you can use:

cd ~/.dsh/profiles/web && pnpm add dsh-ui-three-body@latest

After installation, restart dsh web. The README installation instructions state the floating schemer appears on the right side of the screen, while package.json describes it as a floating cute pet in the top-left; the documentation is inconsistent, so please refer to the actual interface before use. The floating schemer supports click-to-open menus and long-press dragging.

If a desktop shell is needed, the README recommends using it in conjunction with dsh-desktop.

Typical Usage

  1. Install the plugin and restart dsh web.

  2. Open “Settings → Three-Body”, select AI mode based on the task; when AI mode is off, there will be zero kernel tokens per round.

  3. Select the kernel level (minimal / balanced / full) based on the task.

  4. If the task involves UI, use UI preview before delivery: 4 templates, 1 template, or markdown only, and check the HTML templates and design tokens.

  5. Pay attention to the quality gate during execution: Real Verification Loop, Acceptance = Executable Assertions, Self-evaluation meets standards, Human review + rollback + stop if outside contract.

Applicable Scenarios and Notes

Suitable for scenarios in DSH web/profile where you want to add requirement breakdown, solution finalization, execution constraints, and delivery acceptance to agents.

Notes:

  • The plugin runs with the permissions of the current dsh process; you should check the source code and license before installing.
  • The license is MIT; Node requires >=22.
  • Dependencies are @deepseek-ai/schemastery ^3.18.1; peerDependencies include @deepseek-ai/cordis ^4.0.1, @deepseek-ai/dsh-tools ^0.1.0-rc.7, @deepseek-ai/dsh-llm ^0.1.0-rc.7.
  • Voice real-time interaction and visual real-time interaction are marked as planned/pending in the README; this article does not introduce them as implemented features.
  • The floating position is inconsistent between package.json and the README; please refer to the actual interface before use.

The DSH Community Directory is an independent site and is not equivalent to the DeepSeek official app store.

  • Directory Page: https://www.skillhub.cn/plugins/EternalNight996/dsh-ui-three-body
  • GitHub: https://github.com/EternalNight996/dsh-ui-three-body