better-reasoning-slider

中文版

A DeepSeek Harness WebUI plugin that replaces the composer's model/effort dropdown with a compact, theme-aware control:

• Model picker — click the model pill to switch models.
• Reasoning effort slider — drag to adjust effort (e.g. low / high / max) with visible tick nodes on the track.

The slider writes through the same per-session ModelDirectory as the official @deepseek-ai/dsh-client-ui-model-selection plugin, so the /model popup and the composer stay consistent.

Features

• Official-style composer trigger: model name + current effort + chevron.
• Floating popup that stays inside the viewport.
• Model groups and descriptions from the existing model directory.
• Reasoning effort as a continuous slider with snap-to-node behavior.
• Nodes are rendered directly on the slider track and align with the thumb.
• Theme adaptive: uses DSH design tokens (--dsw-specific-menu, --dsw-alias-*, --dsw-shadow-lv3), so it follows light/dark Harness themes.
• No configuration file needed.

Install

From GitHub (recommended)

dsh plugin --profile web add github:vvvspec/better-reasoning-slider

Or with a full URL:

dsh plugin --profile web add https://github.com/vvvspec/better-reasoning-slider

Restart dsh web after installing.

Local development

dsh plugin --profile web add file:/absolute/path/to/better-reasoning-slider

Restart dsh web.

Usage

1. Open the Harness WebUI.
2. In the composer, click the model pill (e.g. the current model name).
3. A popup opens with:
   • The model list (grouped by provider).
   • The Reasoning slider below the model list.
4. Drag the slider thumb to the desired effort level.
   • The thumb snaps to the nearest available node when you release.
   • The selected effort is written to the current session and stays consistent with /model.

How it works

• The plugin shadows the official conversation.input.model slot.
• It uses the same modelDirectories service as the official model selector.
• The slider maps a 0–1000 draft range to the model's available reasoning.efforts list.
• On release, it commits the nearest effort through ModelDirectory.select().

Requirements

• DeepSeek Harness Web profile.
• A model that exposes reasoning.efforts to see the slider. Models without reasoning efforts still show the model picker.

License

MIT