Introduction

When debugging agents or demonstrating features in DeepSeek Harness (DSH) Web mode, showing interface content to the model often requires screenshots. The traditional approach relies on system screenshot tools and then manually uploads the image to the chat box, making the workflow cumbersome. The dsh-screenshot plugin provides a screenshot experience similar to WeChat PC client for the DSH composer, integrating screenshot capture, cropping, and attachment sending into a single workflow.

Plugin Overview

This is a screenshot plugin designed for the DeepSeek Harness Web chat interface, maintained by developer XEsx630. It adds a new button to the composer toolbar, allowing users to capture a screen region or take a screenshot after hiding the current window, and then insert the processed image directly into the conversation without leaving the DSH interface.

Core Features

  • Toolbar Integration: Adds a “Screenshot” button to the composer toolbar.
  • Dual-Mode Screenshots: Supports “Capture Any Screen Region” (full-screen mask with drag selection) and “Screenshot After Hiding Current Window” modes.
  • Server-Side Processing: Server-side PowerShell handles image cropping, without relying on browser canvas, thus avoiding decoding timing issues.
  • Automatic Path Injection: When sending a message, it automatically injects text containing the absolute image path for the model to read.
  • MCP Compatibility: Supports reading image content through the MCP tool (vision).

Installation and Enablement

Use the official installation command to add the plugin. After installation, the dsh.bundle manifest automatically mounts the dsh-screenshot entry into the profile. Restart dsh web and refresh the browser to use it.

dsh plugin --profile web add github:XEsx630/dsh-screenshot

Typical Usage

  1. Click the “Screenshot” button in the DSH composer toolbar.
  2. Choose “Capture Any Screen Region” or “Screenshot After Hiding Current Window”.
  3. Drag the mouse to select the screenshot area. After confirmation, an image card appears above the input box.
  4. Enter your text and click send. The model receives text containing the absolute image path and uses it to call the vision MCP to read the image content.

Dependencies and Limitations

  • Environment Dependencies: Requires a Windows environment, PowerShell 5.1, and the System.Windows.Forms/System.Drawing libraries.
  • DSH Version: Depends on the DSH 0.1.x web profile.
  • Network and Security: All routes are limited to loopback + same-origin. Screenshot/cropping scripts are executed only locally, do not upload any data, and image content is only written to the $DSH_HOME/uploads directory.

Summary

This plugin uses local PowerShell to complete image processing, ensuring the stability of the screenshot feature. It is suitable for local debugging or demonstration scenarios. For more details, visit the GitHub repository.