dsh-reasoning

Switching reasoning effort in DeepSeek Harness Web used to mean navigating a two-level menu inside the composer seat with a mouse—slow, and disruptive for anyone who lives on the keyboard. dsh-reasoning wraps the exact same set of effort levels (Off / Low / High / Max, plus a Default row when the adapter has no explicit default) into a single /reasoning slash command: type three characters, press Enter, arrow or type to filter the popup, hit Enter again. No mouse required, three keystrokes to switch.

The plugin stays deliberately thin. It writes no persistent configuration, registers no global shortcuts, and never touches your draft, IME composition, or an in-flight request. It contributes one popupSelect through the official commandUi extension point and submits picks via the same session.selectModel RPC the built-in effort pane uses, so behaviour is identical to the native menu. Picking the already-active effort closes the popup without any request; a rejected selection keeps the previous effort and surfaces the host error inline in the popup.

Built for DeepSeek Harness Web users who prefer a fully keyboard-driven workflow and switch reasoning effort often. All you need is an ordinary session with a model whose adapter advertises reasoning metadata. The command hides itself in subagent sessions (same rule as /model) and leaves the popup empty when no model is selected, so it will never fire a stray request or conflict with other plugins.