DeepSeek Harness (DSH) has a plugin ecosystem that allows extended functionality. When developing agents, as more tools and MCP plugins are added, system prompts, tool lists, and skill catalogs can become bloated, affecting model performance. In addition, DSH’s native subagent configuration is relatively simple, making fine-grained control difficult. dsh-agent-studio aims to solve these problems.

What It Is

dsh-agent-studio is a graphical plugin for DeepSeek Harness, maintained by thissensen and released under the MIT license. The core value of the plugin lies in precisely configuring an Agent’s prompts, tools, skills, subagents, and fallback models through a visual interface, supporting preset-based binding and takeover, and being compatible with dsh-context.

Core Features

  • Visual Assembly: Supports configuring fallback models, PTC mode, and runtime snapshot toggles.
  • Fine-Grained Configuration: Supports visually and precisely configuring an Agent’s prompts, tools, skills, subagents, and fallback models.
  • Independent Subagents: Subagents support independent configuration of models, prompts, tools, skills, and thinking intensity.
  • Configuration Management: Supports configuration import/export and is compatible with dsh-context.

Installation and Enablement

Install it using npm:

dsh plugin --profile web add dsh-agent-studio

After installation, DSH must be restarted once, and the panel can be viewed under Settings → Agent Center.

Typical Usage

  1. Open Agent Center in settings.
  2. Create prompt sets, tool sets, and skill sets.
  3. Create an agent, attach the sets to the agent, and configure the model and background mode.
  4. Bind the agent to a preset.
  5. Save the configuration.

Notes

  • Data Isolation: The plugin’s configuration is stored in its own data, does not modify the host’s existing preset definitions, and restoring the original state is achieved by uninstalling.
  • Dependency Environment: DSH must be restarted after installation for changes to take effect.
  • Installation Methods: Supports offline installation (by downloading the .tgz file) and installation from source code.

This plugin is suitable for developers who need fine-grained control over agent components, allowing them to quickly build and debug complex agent workflows through a visual interface. For more details, refer to GitHub.