Introduction

When managing models using DeepSeek Harness (DSH), creating or editing model configurations often requires specifying which input types the model actually supports. dsh-model-capability addresses this scenario: it provides an input type selector on the model settings page in Harness, allowing capability declarations to be directly embedded in the model configuration.

DSH’s plugin-based approach allows extending web-side capabilities by scenario; the community directory is an independent site, not the official DeepSeek/Huafang app store. Below is an introduction to the plugin’s positioning, capabilities, installation method, and notes.

What is it

dsh-model-capability is a DeepSeek Harness plugin used to explicitly select the input types supported by a model when creating or editing a model in Harness.

Verified information includes:

  • Plugin Name: dsh-model-capability
  • Maintainer: WJZ-P
  • License: MIT
  • Version: 1.0.1
  • Currently targeting: DeepSeek Harness 0.1.0-rc.5 Web Client discovery path
  • Extension Point Used: settings.models.model.fields

The core problem it solves is: whether a model supports image input cannot rely solely on the model name or guesswork; it should be expressed through the input type declaration in the model configuration.

Core Features

The verified functionality points are introduced below.

  1. Displays an input type selector within Harness’s built-in model editing area.
  2. Selectable options are Text, Image, or Text and Image.
  3. Preserves the provider’s default capabilities and does not write input explicitly.
  4. Uses the settings.models.model.fields slot without taking over the form and persistence logic.
  5. Reuses Harness design variables and automatically follows light/dark themes.
  6. Displays corresponding field text based on Harness’s current language.

Input Types and Configuration Writing

The selection options correspond to the input field in the model configuration. The specific mapping is as follows.

When “Inherit Provider Default” is selected, input is omitted.

When “Text” is selected, it writes:

["text"]

When “Text + Image” is selected, it writes:

["text", "image"]

When “Image” is selected, it writes:

["image"]

Please note here: The plugin only updates the input field of the current model row via the model settings slot publicly exposed by Harness; form handling and persistence are still the responsibility of the Harness side.

Installation and Activation

To install the plugin in the DSH Web profile, use the following command. This command installs the plugin to the web profile.

dsh plugin --profile web add dsh-model-capability

After installation, when creating or editing a model on the model settings page, you will see the input type selector.

Typical Usage

  1. Open the model settings page in Harness.
  2. Create or edit a custom model.
  3. Expand the Advanced Configuration for the model entry.
  4. Select the capabilities the model actually supports in the Input Type field.
  5. Save the model configuration.

If the model supports images, “Text + Image” is usually selected.

Applicable Scenarios and Notes

This plugin is suitable for the following scenarios:

  • You need to maintain custom models in Harness and need to explicitly declare input types.
  • You want model capabilities to be consistent with Harness’s model settings rather than relying on extra explanations.
  • You only wish to add a field in the native model editing area without changing the form submission or persistence flow.

Pre-use considerations:

  • The current version targets the Web Client discovery path for DeepSeek Harness 0.1.0-rc.5; compatibility should be confirmed before using other versions.
  • Development environment requires Node.js ^22.19.0 || >=24.0.0 and pnpm.
  • Official @deepseek-ai/* packages all use peerDependencies.
  • The plugin runs with the permissions of the current dsh process; source code and license should be checked before installation.

Conclusion

The value of dsh-model-capability is focused: clearly stating “Text”, “Image”, or “Text + Image” in the Harness model configuration reduces the issue of inconsistencies between declared model capabilities and actual capabilities.

GitHub Repository: https://github.com/WJZ-P/dsh-model-capabilities. The verified materials did not provide a usable directory page URL, so this article only provides the GitHub link.