Preface

If you run multi-subagent task flows in DeepSeek Harness (DSH), you might encounter a specific problem: the default configurations for a subagent’s provider, model, and reasoning effort are scattered across configuration files. To change the default model, you have to manually edit YAML, and after changing it, you need to verify that every subagent has adopted the new value.

dsh-subagent-setting moves this to the Web Settings page: select the default provider/model/reasoning effort on the interface, and newly created subagents take effect immediately. You can also choose to let existing idle subagents follow the new settings.

What is this

dsh-subagent-setting is a DSH plugin maintained by marshfolx, current version 0.1.0, and is licensed under the MIT license. It solves a single problem: providing a visual entry point for default model configuration for subagents, replacing manual YAML editing.

Core Features

According to the plugin’s README, it provides the following capabilities:

  1. Settings Page Configuration: Select provider, model, and reasoning effort from the real-time model directory without manually editing YAML.

  2. Newly Created Subagents Take Effect Immediately: override is applied individually when each subagent is created; newly created subagents directly use the new configuration after creation.

  3. Optional Real-time Propagation: When the setting “Apply setting changes to subagents created earlier” is enabled, configuration changes will update all existing idle subagents, and the new values take effect on their next request; running subagents will take effect from the next step. When this option is disabled, changes only affect subagents created later.

  4. Leave it Blank to Inherit Parent Session: If any of provider/model/effort is left blank, the subagent will use the configuration used by the parent session.

  5. WebView-friendly Dropdown: The settings form uses a fully controlled custom dropdown component instead of the native <select>, ensuring consistent behavior in browsers and the Tauri (WebView2) shell.

Implementation Details

Understanding the implementation helps determine if it fits your environment; the README provides a brief explanation:

  • On the Host side (lib/index.js), the dsh-subagent-setting settings namespace is registered. In the agent/created event, an agent-level agent/request waterfall listener is installed for each agent with origin: 'subagent' in the session header, rewriting the call configuration to the configured route.
  • The listener reads the subagent’s real-time route holder; therefore, applyToIdle only needs to replace that holder to let existing subagents switch to the new configuration on their next request.
  • The Client side (client/client.js) is responsible for rendering the settings area and communicates via the official connection APIs (settings.describe / settings.replace / llm.models).

Installation and Activation

First, install the plugin to the profile, then add it to the bundle list:

dsh plugin --profile web add github:marshfolx/dsh-subagent-setting

After installation, open Settings → Subagent Model to configure the defaults.

Typical Usage

Following the usage flow in the README:

  1. Execute the installation command above to load the plugin into the web profile and add it to the bundle list.
  2. Open Settings → Subagent Model and select the provider, model, and reasoning effort from the model directory.
  3. If you want previously created subagents to follow the setting changes, enable the “Apply setting changes to subagents created earlier” option; otherwise, keep it disabled to let only newly created subagents use the new configuration.

If you don’t want to lock in a specific model, simply leave provider/model/effort blank, and the subagent will inherit the configuration from the parent session.

Suitable Scenarios and Notes

Suitable scenarios: DSH users who manage multiple subagents on the Web, need to frequently adjust default model configurations, or want to free model selection from YAML editing.

A few notes:

  • Real-time propagation only applies to idle subagents; running subagents will only get the new configuration from the next step.
  • The plugin runs with the permissions of the current DSH process; it is recommended to check the source code and license before installing. This project uses the MIT license, and the source code is available on GitHub.
  • The project is currently at version 0.1.0, in its early stages; it is recommended to verify it in your own environment before using it in production.

Conclusion

dsh-subagent-setting does something not complicated, but it changes subagent model configuration from manual YAML editing to a few clicks in the settings page, and implements real-time updates for existing idle subagents through route holder replacement. If your workflow depends on multiple subagents, it is worth a try.

  • Community Directory Page: https://www.skillhub.cn/plugins/marshfolx/dsh-subagent-setting
  • GitHub Repository: https://github.com/marshfolx/dsh-subagent-setting

By the way, DSH’s philosophy is “Everything is a plugin”. The directory page above is an independent community site with no official affiliation with DeepSeek / Hypersphere.