Introduction

In the development of DeepSeek Harness (DSH), when integrating visual recognition capabilities into an agent, it is often necessary to handle detection tasks for specific targets in images. Although general-purpose vision models are powerful, they may have insufficient accuracy or incur high costs for precise detection of specific devices, such as mobile phones. xby-cellphone-detection is a DSH plugin designed for this scenario, aiming to rapidly locate and identify mobile phones in images through a dedicated detection service.

Functionality

The plugin provides the following core capabilities:

  • set_xby_apikey: Sets the API key in the conversation context. This setting is automatically persistent and remains valid even after restarting the DSH process.
  • cellphone_detection: Accepts an image file URL and detects mobile phones in the image. The output includes bounding boxes, confidence scores, and labels for all detected targets.
  • cellphone_detection_for_data_base64: Accepts the BASE64 encoding of an image file and performs the same mobile phone detection logic as above.
  • cellphone_detection_for_data_file: Accepts the local file path of an image file and performs the same mobile phone detection logic as above.

Installation and Prerequisites

Before use, go to the Xiaobenyang official website to register and obtain an API key.

The installation command is as follows:

dsh plugin --profile web add github:xby_skill/xby-cellphone-detection

Notes

This plugin is open-sourced under the MIT license. As a DSH plugin, its execution depends on the permission environment of the current process. Before installation, developers are advised to review the source code to ensure compliance with local security policies. The plugin belongs to the community directory ecosystem and has no official affiliation with DeepSeek or High-Flyer Quant Research.

Conclusion

xby-cellphone-detection addresses the practical pain point of DSH plugin developers integrating mobile phone visual detection. To view the source code or submit issues, visit its GitHub page.