@nicearrack/dsh-translator

Translator

有道风格的划词翻译，适用于 DSH Web GUI。在任意位置选择文本，点击悬浮按钮，即可通过 harness 自带的 LLM 获取翻译——无需 API 密钥，也不使用第三方端点。

目标 Profile

插件会安装到这里；不确定时保持 web。

npx -y @deepseek-ai/dsh plugin --profile web add @nicearrack/dsh-translator@0.2.1

Selecting text and the floating translate button

兼容性与来源证明

Translator 以 @nicearrack/dsh-translator 发布，当前版本为 0.2.1。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。

DSH 兼容范围: *
运行环境: web
发布来源: npm
Registry 更新时间: 2026/9/21

版本

0.2.1stable

2026/8/28

0.2.0stable

2026/8/24

0.1.1stable

2026/8/22

查看其余 1 个版本收起版本

0.1.0stable

2026/8/21

README

dsh-translator

Youdao-style word-selection translation for DeepSeek Harness (DSH).

Select any text — a conversation message, the sidebar, an input field — and a floating translate button appears at the selection's top-right corner. Click it to get a translation card. Translations are produced by the harness's own LLM: no API keys, no third-party endpoints, your selected text never leaves the model pipeline you already configured.

English	中文 README

Usage

Select text → click the floating translate button → the translation card appears. Drag the card by its header to move it, or pin it so it can't be dismissed by accident. Click the translation (or the source text, or an error message) to copy it. The card's footer shows the translation direction, the model, the reasoning effort, and the token consumption.

Features

Word-selection trigger anywhere in the app; floating button follows the selection, theme-aware (light/dark)
Automatic direction: text in your configured primary language → English; anything else → the primary language
Primary language configurable: 简体中文 / 繁體中文 / 日本語 / 한국어 / Русский (default 简体中文)
Config card in DSH settings (Settings → Plugins → 划词翻译): primary language, model (pick one, or leave it empty to follow the session default), reasoning effort (off/low/high/max), timeout, max tokens, temperature — each field shows "overridden / restore default", with staged edits and save/discard
Translation card: source, direction badge, result, retry on failure; the footer shows model · reasoning effort · tokens
Click to copy: translation, source text, or error message — with a "已复制 / Copied" hint
Draggable card: drag it by its header; it stays where you put it
Pinnable card: locks drag, Esc and click-outside until you close it — one card at a time, no nesting
Streaming-robust: works on live streaming output (no flicker), and the card stays put when the content scrolls

Screenshots

Select text → the floating translate button appears at the selection's top-right corner:

Selecting text and the floating translate button

Click it → the translation card opens (English → Chinese):

Translation card EN → ZH

Chinese → English direction is automatic:

Translation card ZH → EN

Configure it in DSH settings → Plugins → 划词翻译:

Plugin configuration card

Install

From npm:

dsh plugin --profile web add @nicearrack/dsh-translator@latest

From a local checkout:

dsh plugin --profile web add /path/to/dsh-translator

From git:

dsh plugin --profile web add github:nicearrack/dsh-translator

Git installs run the package's prepare build script. pnpm ≥ 10 requires one-time authorization: add allowBuilds: { "@nicearrack/dsh-translator": true } to the profile's pnpm-workspace.yaml if the first install is refused.

Update

# update to the newest release
dsh plugin --profile web update @nicearrack/dsh-translator

# or pin a specific version
dsh plugin --profile web update @nicearrack/dsh-translator@0.2.1

Fresh releases: npm registry metadata can lag after a publish (a few minutes, occasionally longer). If @latest still resolves to an older version, pin the exact one — @nicearrack/dsh-translator@0.2.1 — or check npm view @nicearrack/dsh-translator versions.

After updating, restart the DSH instance (dsh --profile web) so the new host bundle loads; the client bundle picks up on the next page refresh.

Uninstall

dsh plugin --profile web remove @nicearrack/dsh-translator

Restart the instance (dsh --profile web) for the removal to take effect.

License

MIT