Introduction¶

When building agent applications based on DeepSeek Harness (DSH), recommendation systems or ranking pipelines often involve multiple stages (such as data preparation, candidate retrieval, reranking, etc.). The lack of a structured framework can lead to a loose process that is difficult to reproduce. The following section introduces a DSH skill that provides a pipeline framework.

Plugin Overview¶

satan9394/dsh-recsys-pipeline is a DSH skill plugin intended to provide a six-stage framework for recommendation and ranking pipelines. The framework is inspired by the wshobson/agents project. The plugin is maintained by satan9394 and belongs to the “Workflow” category.

Core Features¶

The plugin provides a structured pipeline framework for handling recommendation/ranking tasks. It divides the recommendation process into six stages (the exact stage definitions should be checked in the source code), helping developers organize task logic.

Installation and Activation¶

No official installation command is currently provided. You can obtain the plugin source code or details in the following ways:
* Directory Page: https://www.skillhub.cn/plugins/satan9394/dsh-recsys-pipeline
* Source Repository: https://github.com/satan9394/dsh-recsys-pipeline

After downloading the source code, add the skill directory to the working directory according to the standard DSH process to enable it.

Applicable Scenarios and Precautions¶

• Audience: DSH developers who need to build recommendation systems or ranking pipelines.
• Permissions and Security: The plugin runs with the permissions of the current DSH process. Before use, be sure to inspect the source code content to ensure there is no malicious code and pay attention to license information (no license is currently specified).

Conclusion¶

If you are looking for a structured recommendation pipeline framework, you can use this plugin as a reference implementation. Specific implementation details need to be adapted by you based on the source code.