Introduction¶
In the agent development workflow of DeepSeek Harness (DSH), dataset preparation and processing often take up a lot of time. From cleaning raw data to selecting high-quality samples and checking data diversity, and then splitting training and validation sets, as well as performing subsequent annotation work, manual handling is not only inefficient but also error-prone. The following section introduces a plugin that integrates these dataset curation workflows into the DSH workflow.
What It Is¶
satan9394/dsh-dataset-curation is a workflow-type DSH plugin. It was developed by the maintainer satan9394, with the goal of addressing the need for dataset organization, cleaning, quality filtering, diversity handling, data splitting, and annotation in agent development. The design of this plugin was inspired by the wshobson/agents project.
Core Features¶
The plugin currently provides the following verified features:
* Dataset organization
* Data cleaning
* Quality filtering
* Diversity handling
* Data splitting
* Data annotation
Installation and Enablement¶
No official installation command is currently provided. Please refer to the GitHub repository for the latest installation method.
Typical Usage¶
No specific usage examples are currently provided.
Use Cases and Notes¶
This plugin is suitable for developers who need to batch-process and standardize datasets. When using it, please note that the plugin runs with the permissions of the current DSH process. Since license information is not currently specified, be sure to review the source code and license terms before installation.
Conclusion¶
In summary, this tool incorporates the tedious process of dataset curation into the DSH workflow, making it well suited to agent developers who frequently need to process datasets. For more details and source code, please visit GitHub or the DSH Community Catalog.