dsh-reasoning-effort

When multiple providers are connected to DeepSeek Harness, each model may use different protocol wire values, and reasoning-effort configuration tends to be scattered with no unified entry point. dsh-reasoning-effort solves this by auto-discovering every provider already present in settings and filling in discrete reasoning levels from off to max for any model that lacks a declaration, with no need to type provider names or maintain a whitelist.

The core experience centers on a provider-grouped, collapsible model selector and a chunky-track, discrete-step reasoning slider. Dragging to the highest tier triggers a purple glow, edge flicker, and trailing effect that keeps signalling the current state in real time. Model switches and slider movements are staged on the front end first and only committed to the backend on release or selector close, preventing stale-state flicker. Settings are always appended; existing reasoningEfforts values and unrelated fields are never overwritten.

The plugin suits users juggling several providers who want a single place to manage reasoning intensity, as well as advanced users on custom gateways that need self-defined wire-value mappings. It adds no new credentials, requests, or telemetry-it simply completes the settings contract, leaving concrete model control to Harness itself.