AI Agent Hub
Back to plugins
dsh-reasoning-ruler preview

dsh-reasoning-ruler

Workflow Updated 2026.08.30

Run the following command in DeepSeek Harness:

dsh plugin install zisen123/dsh-reasoning-ruler

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install zisen123/dsh-reasoning-ruler in DeepSeek Harness to install this plugin; the source lives at https://github.com/zisen123/dsh-reasoning-ruler .

About this plugin

Picking a reasoning effort in DSH Web used to mean opening the model popover, scrolling to the right row, then clicking a tier — a long path with sluggish feedback, and the model picker itself often carried small friction like scrollbars and a Refreshing flash. dsh-reasoning-ruler pulls the tier selector out of the popover and turns it into a minimal ruler on the composer: a 2px line that rises left to right, with the selected tier shown as a sliding vertical mark, a ghost mark plus tooltip on hover, and the tier name to the left of the line.

Three experience pillars define the plugin. Tiers are remembered per model in localStorage, so switching back to a model restores the last chosen strength automatically. Switching is optimistic: the UI reacts the instant you click, the RPC settles in the background, and a failure rolls back cleanly. The model picker is also polished to open straight onto the full list, drop the scrollbar, hide the Refreshing-model-list flash, and close on selection, with duplicate tier rows in the popover removed so the ruler stays the single source of truth.

Best suited for developers, researchers, and heavy users who regularly swap models or tune thinking effort inside DSH Web and want tier selection to feel like a glance instead of a menu hunt. The core path rides official slots and the ModelDirectory API, so it is stable across DSH upgrades; the cosmetic layer depends on DOM details and may need an issue filing if a DSH release shifts them. Shipped under MIT as an experimental plugin.

Screenshots

Use Cases

  • Switch models and tune thinking effort often
  • Prefer a one-glance tier control without popovers
  • Keep the last tier per model for quick recall

Best For

  • Frequent DSH Web developers
  • Researchers who fine-tune reasoning strengths
  • Power users seeking smoother interactions