Preface

The core logic of DeepSeek Harness (DSH) is “everything is a plugin.” When building complex Agent systems, using prompt concatenation directly often makes it difficult to control lifecycle and state management. dsh-j-space fully integrates J-Space Cognition Suite SV1 into DSH, positioning it as a first-class Agent preset.

This plugin solves the pain point of introducing complex cognitive frameworks into a DSH environment. Through Scope isolation and state externalization, it enables large models to participate in more persistent and controllable engineering and reasoning tasks.

Plugin Overview

  • Name: dsh-j-space
  • Author: AnonyJcy
  • Type: Workflow plugin / Agent preset
  • License: MIT
  • Positioning: A native DeepSeek Harness Agent preset and standalone plugin package for J-Space Cognition Suite SV1.

Core Features

The plugin decomposes the J-Space system into the following core capabilities:

  1. Plug and Play: After deployment, the J-Space Cognition Suite preset option automatically appears in the DSH session creation menu.
  2. Full Lifecycle Scope Isolation: Compliant with the Cordis specification, tools and cognitive skills are strictly confined to J-Space Agent sessions to avoid polluting other presets.
  3. Persistent Controller and State Externalization: Includes a persistent controller (control.py), a host adapter, and a lightweight ledger. It maintains .jspace/ in the task workspace to enable goal tracking, checkpoint recovery, and a code semantic map.
  4. 13 Cognitive and Engineering Modules: In addition to basic reasoning, it integrates modules such as capacity control, deep reasoning, and directed focus, along with the new SV1 modules cyber (security analysis), epistemics (belief verification), orchestration (multi-agent orchestration), and repository (code repository semantic map).
  5. Model Decoupling and Dynamic Routing: Seamlessly adapts to various base models (DeepSeek-Chat, DeepSeek-Reasoner, Claude, GPT, etc.) and dynamically resolves routing.

Installation and Configuration

It is currently compatible with DSH 0.2.0, and the plugin version is 1.2.3.

1. Installation

It is recommended to install it using DSH’s built-in plugin management command. DSH loads cordis.patch.yml through dsh.bundle.patch in the package to register the preset.

# 安装到实际使用的 Web profile
dsh plugin --profile web add @anonyjcy/dsh-j-space@latest

Note:
- Running only npm install into a regular project directory will not register the preset. You must use the command above.
- After installation, restart the DSH Web process and create a new session.

2. Configure Subagent Model Selection

The J-Space preset requires the host to load the @deepseek-ai/dsh-tool-subagent/model-selection-settings plugin by default.

In the DSH Web interface, go to Settings -> Plugins -> Subagent, and enable Allow Agent to Select Models for Subagent. Then check the models to allow in the model routing list. After creating a new session, J-Space can invoke these subagent models.

Usage

After installation, the steps are as follows:

  1. In the DSH Web interface, click New Session.
  2. In the Agent Preset dropdown, select J-Space Cognition Suite.
  3. In the model selector below, choose any compatible model (such as deepseek-chat or deepseek-reasoner) and start the task.

Notes

  1. Host Dependency: By default, mounting this preset depends on the @deepseek-ai/dsh-tool-subagent/model-selection-settings plugin included with the DSH Web application. If the host does not load this plugin, the preset cannot be mounted.
  2. Deprecation of Legacy Configuration: The configuration method for the legacy standalone Cordis plugin (CLI install command) has been deprecated. Be sure to use the current bundle installation method.
  3. Permissions and Security: The plugin runs with the current DSH process permissions. Before installation, it is recommended to review the source code and license.

Conclusion

dsh-j-space provides DeepSeek Harness users with a mature integration point for cognitive frameworks. It is not just a prompt template, but a system that includes state management, tool invocation, and modularized cognition.

  • Plugin Directory: https://www.skillhub.cn/plugins/AnonyJcy/dsh-j-space
  • GitHub Repository: https://github.com/AnonyJcy/dsh-j-space