dsh-effort-slider

Switching models in DeepSeek Harness silently resets reasoning effort: when the target model has no reasoning route configured, the level falls back to undefined, which is equivalent to enable_thinking: false, and thinking is quietly turned off. dsh-effort-slider closes that gap by injecting a draggable effort slider into the conversation input dock, scoped to the current session so other chats stay untouched.

Three core capabilities stand out. First, the slider renders only the tiers the active model declares, auto-hiding itself for single-tier or zero-tier models. Second, it remembers the last level you explicitly chose for each provider/model pair, then carries and clamps that value onto the target model ladder on a switch, guaranteeing it can never silently demote thinking to no-thinking. Third, if you have never picked a level it honestly shows a default state rather than pretending a value was set. The whole bundle is a static module with zero build tooling and zero dependencies, and can optionally back dsh-bailian-models.

Built for Harness users who frequently switch between multiple models and want precise, persistent control over reasoning depth. Drag once, switch freely, and your effort level follows.