dsh-plugin-reasoning-effort
配置 DeepSeek Harness 的模型输入能力和推理力度
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add dsh-plugin-reasoning-effort@0.1.9Plugin Reasoning Effort 以 dsh-plugin-reasoning-effort 发布，当前版本为 0.1.9。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
A community DeepSeek Harness Web plugin for configuring model capabilities. It adds a settings card for assigning input modalities and reasoning levels to one or more models and keeps model selection in the DSH conversation composer. The model popup opens side-by-side, so the Model and Reasoning rows remain visible while their options are being browsed.
This is an independent community plugin. It is not an official DeepSeek product and does not imply DeepSeek endorsement.
input field (text and image).Off, Minimal, Low, Medium, High, Extra high, Max) in one step.id when the
identifier is also the wire value, or id=wireValue when it differs.llm-pi-ai.providers.<provider>.models[].reasoningEfforts and
llm-pi-ai.providers.<provider>.models[].input, allowing DSH and the provider
integration to validate and send the configured values.Install the precompiled package from npm into the DSH Web profile:
dsh plugin --profile web add dsh-plugin-reasoning-effort
The npm package already contains the JavaScript runtime generated from the TypeScript source, so pnpm does not need to run dependency build scripts during installation.
Restart the DSH Web profile after installation so the Host and Client plugin graph is rebuilt.
To install a local checkout while developing:
git clone https://github.com/LanQin996/dsh-reasoning.git
cd dsh-reasoning
npm install
npm run build
dsh plugin --profile web add file:.
To remove the plugin:
dsh plugin --profile web remove dsh-plugin-reasoning-effort
After restarting DSH Web, open the model selector in the conversation composer. The popup keeps the current Model and Reasoning rows on the left; choose the corresponding list on the right. The Reasoning row is shown only when the selected model has configured levels.
Open:
Settings -> Plugins -> Plugin configuration -> Reasoning Effort
Expand the card and select one or more models:
Text and Image for the selected models. Image
attachments are only sent to models whose input includes image.balanced, thorough=high_reasoning.The card reads the existing llm-pi-ai provider settings and writes model-level
input and reasoningEfforts fields. An unset input keeps the provider or
installed catalog behavior until you explicitly configure it.
For a checkout that should be loaded directly by a DSH installation, use the
included cordis.patch.yml:
dsh web --patch C:/path/to/dsh-reasoning/cordis.patch.yml
The patch inserts the plugin with the id dsh-reasoning-effort.
The repository contains the TypeScript Host entrypoint and browser-side client in
src/index.ts and src/client.ts. TypeScript is compiled into the runtime files in
dist/ before the plugin is loaded:
npm install
npm run typecheck
npm run build
npm pack and npm publish run the prepack hook, which type-checks and compiles
the TypeScript source into dist/. Installing the published package does not execute
that build hook and is compatible with pnpm v10 without an allowBuilds entry.
The client expects the DSH Web runtime and peer packages declared in
package.json. When testing locally, load it through the patch command above
inside a DSH installation that provides those packages.
The plugin does not add a telemetry client, credential flow, or background
network service. It reads the active DSH model directory and llm-pi-ai settings,
then persists only the configured input and reasoningEfforts values through the
DSH settings store.
Copyright (C) 2026 LanQin.
This project is licensed under the GNU Lesser General Public License v3.0
(LGPL-3.0-only).