dsh-reasoning-effort: A Codex-style Reasoning Effort Slider for DeepSeek Harness

dsh-reasoning-effort is a reasoning-effort slider plugin (DSH plugin) for DeepSeek Harness, maintained by HanaAyane under the MIT license, written in TypeScript. It brings Codex-style model and reasoning-effort controls directly into the DSH web interface, solving the problem of quickly adjusting reasoning effort during conversations. The plugin is enabled by default, stays in sync with DSH's /model command, and automatically adapts to the reasoning.efforts exposed by the current model. It also includes a fun "Big Fat Fish" running animation slider for a livelier UI.

What core features does the dsh-reasoning-effort DSH plugin support?

The plugin offers a truly responsive drag experience: the button moves continuously with the pointer and snaps to valid levels on release. It supports dark and light themes with blue-purple-black and blue-white gradients respectively, deeper blue for higher effort. Effects (waves, shockwaves, pixel radiation, particles, trails) only emit to the left. It stays in sync with DSH state, reading/writing the same session model directory as the /model command. On update failure, it automatically rolls back to the last confirmed version.

How do I install dsh-reasoning-effort?

Use the following commands. It's recommended to use #main for the latest code, or pin to #v0.6.2. After installation, restart the DSH Web Host and refresh the page.

dsh plugin --profile web add github:hanaayane/dsh-reasoning-effort#main
dsh --profile web --dump-config

Where do the reasoning-effort levels come from?

The slider levels come entirely from the reasoning.efforts exposed by the current model in the DSH model directory; the count, names, and order are determined by the model, and the plugin adapts automatically. For example, DeepSeek models typically expose off / high / max, while GLM coding models (like GLM-5.2) expose five levels. If the model provides no levels, the menu shows a notice.

How does dsh-reasoning-effort provide level guidance for custom providers?

For DSH built-in routes, the plugin is strictly read-only and never modifies anything. Only for models you declare yourself in llm-pi-ai in settings.yaml does the plugin offer guidance, including suggested levels, a copyable YAML entry, and the config file path. The built-in knowledge base currently includes GLM-5.2 and Kimi K3; you can also append custom entries under the plugin's namespace.

What is the Big Fat Fish slider?

The plugin enables an eight-frame running figure as the slider button by default; you can turn it off in Settings → General → Appearance. The Big Fat Fish only changes the button appearance, not snapping, keyboard control, or model selection. With "reduce motion" enabled, it stays on a stable frame. The "Reasoning Effort Slider" master switch on the same page can temporarily disable the entire enhanced control, restoring the native DSH model selector.

See more DSH plugins at dsh-plugin.org