dsh-sampling-sliders

Tweaking temperature or maxTokens across multiple providers in DeepSeek Harness usually means editing a config file and restarting — a small friction that compounds quickly. dsh-sampling-sliders collapses that into a single "Sampling" button next to the input bar: click it and a lightweight panel slides out with two sliders (temperature 0-2, maxTokens 512-32768), each with an override toggle. Leave a toggle off and the model default applies, so there is zero collateral. Because the interception fires before provider routing, one setting takes effect on the official DeepSeek endpoint and every third-party provider you have connected — no per-provider setup needed.

Two modes cover different workflows. Hot mode keeps values alive for the current session only and auto-clears on the next launch, ideal for quick A/B experiments where you want to feel how different settings shift output style. Persistent mode writes values into the settings namespace so they survive restarts, making them handy as a fixed sampling baseline for a particular class of task. On startup, any leftover hot-mode state is purged automatically to prevent accidental bleed-through.

Who benefits most: DSH power users who switch between providers frequently and want to retune sampling on the fly without touching a config file; experimentation-driven users who want to feel the difference between temperature settings at a glance. One button, two sliders — sampling parameters moved from a hidden config line to your fingertips.