Preface¶

The core idea of DeepSeek Harness (DSH) is “everything is a plugin.” When developing or debugging agents, developers need to intuitively understand the specific process of plugin and model invocations. toolfolk-for-dsh visualizes this process as a 3D digital office, mapping each plugin stage to a specific role and action.

Overview¶

Visualize the DSH plugin and model invocation process as a 3D digital office — plugins are employees, workstations are positions, and invocations are tasks.
- Author: 浊客er
- Category: Fun Dress-Up

Core Features¶

The plugin maps the plugin and model invocation process to concrete elements in a 3D scene:
- Six fixed positions: Workstations correspond to roles, including information retrieval, writing and editing, web investigation, session memory, command execution, and collaborative scheduling.
- Thought-stream bubbles: Bubbles appear above employees’ heads while they work, displaying the original model inference text.
- Delivery sheet: A slightly wrinkled A4 sheet appears after the response is complete, with a red seal reading “Reviewed”.
- Employee log: A fixed header is shown in the bottom-left corner, listing invocation events and deliveries in reverse chronological order.
- Report records: Located in the top-right corner, collapsed by default; when expanded, it shows all historical responses.
- Light/dark theme: Follows the host’s theme setting.
- Central coordinator robot: The Coordinator node at the center of the scene represents the model itself.

Installation and Activation¶

Prerequisite: official host @deepseek-ai/dsh@0.1.5-rc.1 (host React 18.3.1).

Install the plugin using npx (no authorization required):

npx -y @deepseek-ai/dsh plugin --profile web add toolfolk-for-dsh

Start the host:

npx @deepseek-ai/dsh web --no-open

Open the address shown in the startup output, and switch to “3D Studio” in the view switcher. The room may appear empty the first time; send a message in the conversation, and the employees will immediately take up their positions.

Environment and Notes¶

• Node version: Node 22 is required.
• Local paths: When installing using a local file path, the path must not contain spaces.
• Privacy and permissions: The plugin only reads host events, does not connect to the network, does not upload any data, and writes state only locally (<user home>/.dsh/studio/state.).
• Compatibility: Compatible with the official @deepseek-ai/dsh@0.1.5-rc.1.

References¶

• GitHub: https://github.com/zkl22492-star/Toolfolk-for-DSH
• Catalog: https://www.skillhub.cn/plugins/zkl22492-star/Toolfolk-for-DSH