Foreword¶

In DeepSeek Harness (DSH), switching models is typically done through the model entry below the input box or via the /model command. If a model supports multiple reasoning effort levels, the native interface often only displays the model name, with the levels either hidden in the configuration or requiring memorized command parameters. Products like Codex implement “model + effort” as a single slider control, allowing users to drag and switch between levels such as off / high / max.

dsh-reasoning-effort, developed by community maintainer HanaAyane, aims to bring this interaction to the DSH Web interface: while preserving the native model list, it adds a reasoning effort slider that stays synchronized with the session state and the /model command. The plugin is open-sourced on GitHub (MIT license), currently at version 0.6.2 on the main branch, targeting DSH version 0.1.0-rc.6.

What Is This¶

dsh-reasoning-effort is a frontend plugin for DSH that adds a “model + current effort” entry below the input box. Clicking it opens a popup with a reasoning effort slider at the top, where levels automatically adapt based on the reasoning.efforts publicly declared by the current model in the DSH model directory. Below the slider is the native DSH model selector. The plugin is enabled by default and does not add telemetry or extra network requests.

Maintainer: HanaAyane
Catalog page: https://www.skillhub.cn/plugins/HanaAyane/dsh-reasoning-effort
Source code: https://github.com/HanaAyane/dsh-reasoning-effort

Core Features¶

Reasoning Effort Slider¶

The slider levels are entirely derived from the public declarations of the current model. The number of levels, names, and order are determined by the model; the plugin only handles display and submission. Common examples include:

DeepSeek models typically offer off / high / max; GLM coding models (e.g., GLM-5.2) offer five levels: off / low / medium / high / xhigh. If a model declares fewer than two levels or none at all, the menu will prompt “The current model does not provide reasoning effort levels.”

Interaction-wise, the button moves continuously based on pointer position and snaps to the nearest valid level upon release. If an update fails, it automatically rolls back to the last confirmed level. The slider supports both light and dark themes and includes visual effects like waves and shockwaves that emit only to the left.

Synchronization with DSH State¶

The slider and the /model command read and write to the same session model directory. Turning off the “Reasoning Effort Slider” toggle restores the native DSH model selector without needing to uninstall the plugin.

Guidelines for Custom Provider Levels¶

The built-in pi-ai routing level plugin is completely read-only and does not modify anything. For models declared in settings.yaml under llm-pi-ai that are not recognized by the directory, the menu will show a “View Level Declaration Guide,” displaying suggested levels and copyable YAML snippets. The built-in knowledge base currently includes GLM-5.2 and Kimi K3; additional entries can be added under the plugin namespace dsh-reasoning-effort.entries and will take precedence over the built-in ones.

The Fat Fish Slider¶

By default, the plugin replaces the slider button appearance with an eight-frame running character. You can disable the “Fat Fish Slider” under Settings → General Settings → Appearance to revert to a plain white button; this toggle only affects the button’s appearance and does not change the level snapping, keyboard control, or model selection logic.

Installation and Enabling¶

The plugin loads when the Web Host starts. The installation command targets the web profile; after installation, you need to manually restart the DSH Web Host and refresh the page.

When having an Agent perform the installation, you can send it the entire following text:

Please install the dsh-reasoning-effort plugin for the DeepSeek Harness web profile.

Execute only the following two commands without modifying other profiles:
dsh plugin --profile web add github:HanaAyane/dsh-reasoning-effort#main
dsh --profile web --dump-config

After confirming that dsh-reasoning-effort appears in the output, inform me of the installation result.
Do not close or restart the running DSH for me; after installation, remind me to manually restart the DSH Web Host.

For manual installation, execute the following in PowerShell:

dsh plugin --profile web add github:HanaAyane/dsh-reasoning-effort#main
dsh --profile web --dump-config

#main always tracks the latest code (currently consistent with v0.6.2); to pin a version, change #main to #v0.6.2.

To confirm loading:

dsh --profile web --dump-config

The output should include name: dsh-reasoning-effort.

To uninstall:

dsh plugin --profile web remove dsh-reasoning-effort

A restart of the Web Host is also required after uninstallation.

Typical Usage¶

After installation and restart, follow these steps:

1. Create or open a session.
2. Click the button below the input box that shows “model + current effort.”
3. Drag the slider or click the track; upon release, it snaps to the nearest level.
4. Click the model row below the slider to enter the native DSH model list to continue switching models.

If the current model does not declare reasoning levels in the directory (e.g., some newly added GLM-5.3 models), you can add reasoningEfforts under llm-pi-ai in ~/.dsh/settings.yaml:

llm-pi-ai:
  providers:
    zai-coding-cn:
      models:
        - id: glm-5.3
          name: GLM-5.3
          contextWindow: 1000000
          maxTokens: 131072
          reasoningEfforts:
            low: "low"
            high: "high"
            xhigh: "max"
          compat:
            thinkingFormat: "zai"
            supportsReasoningEffort: true

Level names follow the DSH system (off / minimal / low / medium / high / xhigh), and values are the actual reasoning_effort accepted by the endpoint. Final submission is validated by the host; the plugin does not bypass model or deployment capability restrictions.

Use Cases and Notes¶

This plugin is suitable for developers who frequently switch reasoning effort levels in the DSH Web interface and prefer a Codex-style single-hand operation. Custom provider users can reduce the trial-and-error cost of writing YAML manually with the level guidance panel.

A few notes:

• The plugin runs with the current DSH process permissions; it is recommended to read the source code and the MIT license before installation to ensure it aligns with your environment policies.
• Requires Node.js 22.19+; targets DSH packages version 0.1.0-rc.6; upstream is still in developer preview, so UI changes may require synchronous plugin updates.
• If the slider is not visible after installation, check in order: whether the Web Host has been restarted; whether Settings → General Settings → Reasoning Effort Slider is enabled; and whether the current model publicly declares at least two reasoning effort levels.
• The SkillHub catalog page (https://www.skillhub.cn/plugins/HanaAyane/dsh-reasoning-effort) is an independent community site with no official affiliation with DeepSeek or High-Flyer; the DSH ecosystem follows the “everything is a plugin” philosophy, and this plugin is one part focused on the model reasoning UI.

Conclusion¶

dsh-reasoning-effort brings Codex-style model and reasoning effort selection to DSH. The slider levels follow the model directory, the state is synchronized with /model, and custom models have a guidance panel with copyable YAML. If you are already using the DSH Web profile, follow the README to install, restart the Host, and try it out below the input box.

• Catalog page: https://www.skillhub.cn/plugins/HanaAyane/dsh-reasoning-effort
• GitHub: https://github.com/HanaAyane/dsh-reasoning-effort