@sakka6868/dsh-voice-input 0.1.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 sakka6868
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,85 @@
+# @sakka6868/dsh-voice-input
+
+> DeepSeek Harness Web UI 的「语音输入」插件 —— 点一下麦克风（或按 `Ctrl+Shift+M`），边说边把话实时变成输入框里的文字，停止后自动校对。
+
+[![license](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
+![language](https://img.shields.io/badge/plain%20javascript-ES2020-yellow)
+![dependencies](https://img.shields.io/badge/dependencies-0-brightgreen)
+
+## ✨ 核心亮点
+
+- **实时转录** — 使用浏览器原生 Web Speech API（Edge 中后端即**微软语音服务**），说话时文字逐字出现在输入框，无录音文件、无本地模型、无等待；
+- **零依赖零配置** — 不需要 API Key、不下载模型、不占用 CPU；
+- **自动校对** — 停止说话后静默调用当前会话模型校对：修正错别字、同音/近音转录偏差、繁体转简体、理顺标点；**你手动改过输入框就不会覆盖**；
+- **快捷键** — `Ctrl+Shift+M` 一键开始/停止语音输入；
+- **位置贴合** — 麦克风按钮固定在发送按钮左侧（flex `order:1`，与内置控件同排）；
+- **无打扰 UI** — 录音过程不弹出任何面板/浮层，输入框里的文字就是全部反馈；
+- **干净可逆** — 只写输入框草稿，是否发送完全由你决定；卸载即移除按钮、快捷键与样式。
+
+## 安装与使用
+
+### 方式一：静态插件（npm 包，推荐）
+
+```powershell
+dsh plugin --profile web add @sakka6868/dsh-voice-input
+```
+
+然后在 `~/.dsh/profiles/web/cordis.patch.yml` 注册插件行：
+
+```yaml
+- insert:
+    - id: dsh-voice-input
+      name: '@sakka6868/dsh-voice-input'
+```
+
+重启 DSH web 服务后，输入栏发送按钮左侧出现麦克风按钮。
+
+### 方式二：Dynamic 模式（开发/尝鲜）
+
+在 DSH 的 Cordis 动态插件面板：
+
+1. 新建插件（idPrefix 如 `voice`）；
+2. `code.host` 粘贴 `lib/index.js` 中 `proofread` 相关逻辑（或改用 `harness.handle('voice.proofread', ...)` 形态）；
+3. `code.client` 粘贴 `lib/client.js` 中组件与状态机逻辑（把 `fetch` 换回 `host.call`、把 `<style>` 注入换回 `styles.insert`、把原生定时器换回 `ctx.timeout`）；
+4. 激活并授权。
+
+## 使用说明
+
+| 操作 | 效果 |
+| --- | --- |
+| 点击 🎤 或按 `Ctrl+Shift+M` | 开始语音识别（首次会请求麦克风权限），文字实时写入输入框 |
+| 再次点击红色停止钮或 `Ctrl+Shift+M` | 结束识别，静默校对后回填（未改动则替换，改动过则保留你的版本） |
+| 停顿约 1.5 秒 | 自动结束识别 |
+| 按原有发送按钮 | 发送输入框中的文字 |
+
+识别语言固定为 `zh-CN`（可在 `lib/client.js` 的 `createRecognition()` 中修改 `rec.lang`）。
+
+## 工作原理
+
+```
+浏览器 Web Speech API (实时识别, Edge→微软语音服务)
+        │  interim/final 结果
+        ▼
+conversation.input.right 麦克风按钮 → inputActions.setDraft(文字)  ← 实时写入输入框
+        │  停止 / 停顿 1.5s
+        ▼
+POST /dsh-voice-input/proofread  (Host 半, webServer 路由)
+        │  ctx.get('llm') + ctx.get('agentDefaultModel')
+        ▼
+校对文本 → 回填输入框（仅在草稿未被用户修改时）
+```
+
+- **Client 半**（`lib/client.js`）：module-loader factory 包，注册 `conversation.input.right`（按钮）与 `conversation.input.dock`（错误提示），快捷键监听挂在 `document`，卸载时清理。
+- **Host 半**（`lib/index.js`）：一个 `webServer` exact POST 路由，用会话默认模型做一次低温度校对调用；任何服务缺失或调用失败都退化为“原样返回转写文本”，不会打断输入。
+
+## 兼容性与限制
+
+- **Edge / Chrome**：完整可用（Web Speech API 需要联网，识别在浏览器厂商的语音服务端完成）；
+- **Firefox**：不支持 `SpeechRecognition`，按钮会提示改用 Edge/Chrome；
+- **Safari**：部分版本支持 `webkitSpeechRecognition`，`continuous` 行为可能不同；
+- 离线环境无法识别（语音服务在云端）；
+- 校对依赖当前会话是否配置了可用模型；未配置时保留原始转写。
+
+## 许可
+
+[MIT](LICENSE)

package/lib/client.js

@@ -0,0 +1,266 @@
+// dsh-voice-input — static client half (module-loader factory bundle).
+//
+// Adapted from the dynamic plugin form for the STATIC plugin shape:
+//   - registered via window.__ModuleLoader__.load({ id, factory });
+//   - React arrives through require('react');
+//   - the dynamic plugin's host.call('voice.proofread') is replaced by
+//     fetch('/dsh-voice-input/proofread');
+//   - styles use the official per-plugin <style data-plugin-css> pattern
+//     instead of the dynamic styles Builtin;
+//   - native browser timers replace the dynamic ctx.interval/ctx.timeout.
+//
+// Component logic, state machine, slot ids, button placement (flex order 1)
+// and CSS rules are unchanged from the dynamic form.
+window.__ModuleLoader__.load({
+  id: 'dsh-voice-input',
+  factory: (require) => {
+    var module = { exports: {} };
+    var exports = module.exports;
+    var React = require('react');
+
+    const STYLE_TAG_ID = 'dsh-voice-input';
+    const PROOFREAD_URL = '/dsh-voice-input/proofread';
+
+    const CSS = [
+      '.vc-cell{order:1}',
+      '.vc-btn{width:32px;height:32px;border-radius:50%;border:1px solid var(--dsw-alias-border-l2,rgba(127,127,127,.4));background:transparent;color:var(--dsw-alias-label-secondary,#666);display:inline-flex;align-items:center;justify-content:center;cursor:pointer;padding:0;transition:background .15s ease,color .15s ease,border-color .15s ease}',
+      '.vc-btn:hover{background:var(--dsw-alias-bg-layer-2,rgba(0,0,0,.06))}',
+      '.vc-btn:disabled{opacity:.45;cursor:default}',
+      '.vc-btn.vc-rec{border-color:transparent;background:var(--dsw-alias-state-error-primary,#e5484d);color:#fff}',
+      '.vc-btn.vc-rec:hover{background:var(--dsw-alias-state-error-primary,#e5484d);filter:brightness(1.1)}',
+      '.vc-spin{width:14px;height:14px;border-radius:50%;border:2px solid var(--dsw-alias-border-l2,rgba(127,127,127,.4));border-top-color:var(--dsw-alias-brand-primary,#4c6ef5);animation:vc-spin .7s linear infinite}',
+      '@keyframes vc-spin{to{transform:rotate(360deg)}}',
+      '.vc-dock-line{display:inline-flex;align-items:center;gap:8px;padding:8px 14px;border-radius:999px;background:var(--dsw-alias-bg-layer-1,#fff);border:1px solid var(--dsw-alias-border-l1,rgba(127,127,127,.25));box-shadow:0 2px 10px rgba(0,0,0,.06);font-size:12px;color:var(--dsw-alias-label-secondary,#666);max-width:560px}',
+      '.vc-error{color:var(--dsw-alias-state-error-primary,#e5484d)}',
+    ].join('\n');
+
+    let styleTag = document.querySelector('style[data-plugin-css="' + STYLE_TAG_ID + '"]');
+    if (styleTag === null) {
+      styleTag = document.createElement('style');
+      styleTag.dataset.pluginCss = STYLE_TAG_ID;
+      styleTag.textContent = CSS;
+      document.head.appendChild(styleTag);
+    }
+
+    const svgBase = { width: 16, height: 16, viewBox: '0 0 24 24', fill: 'none', stroke: 'currentColor', strokeWidth: 1.8, strokeLinecap: 'round', strokeLinejoin: 'round' };
+
+    function MicIcon() {
+      return React.createElement('svg', svgBase,
+        React.createElement('path', { d: 'M12 1a3 3 0 0 0-3 3v8a3 3 0 0 0 6 0V4a3 3 0 0 0-3-3z' }),
+        React.createElement('path', { d: 'M19 10v2a7 7 0 0 1-14 0v-2' }),
+        React.createElement('line', { x1: 12, y1: 19, x2: 12, y2: 23 }))
+    }
+
+    function StopIcon() {
+      return React.createElement('svg', { width: 13, height: 13, viewBox: '0 0 24 24', fill: 'currentColor' },
+        React.createElement('rect', { x: 5, y: 5, width: 14, height: 14, rx: 3 }))
+    }
+
+    /** Host-side proofread; any failure degrades to the raw transcript. */
+    function proofreadRemote(text) {
+      return fetch(PROOFREAD_URL, {
+        method: 'POST',
+        headers: { 'content-type': 'application/json' },
+        body: JSON.stringify({ text: text }),
+      }).then((res) => res.json()).then((reply) => {
+        if (reply && typeof reply.text === 'string' && reply.text.trim().length > 0) return reply.text.trim()
+        return text
+      }).catch(() => text)
+    }
+
+    function apply(ctx) {
+      const slots = ctx.get('slots')
+      if (slots === undefined) return
+
+      const TRANSITIONS = {
+        IDLE: { start: { to: 'REQUESTING', effect: 'startRec' } },
+        REQUESTING: { started: { to: 'RECORDING' }, denied: { to: 'ERROR', effect: 'recError' } },
+        RECORDING: { stop: { to: 'IDLE', effect: 'stopRec' }, ended: { to: 'IDLE', effect: 'autoFinalize' } },
+        ERROR: { dismiss: { to: 'IDLE' } },
+      }
+
+      const store = {
+        state: 'IDLE',
+        errorText: '',
+        sessionId: null,
+        inputActions: null,
+        baseDraft: '',
+        finalText: '',
+        currentDraft: null,
+        stopDraft: null,
+      }
+      const listeners = []
+      let recognition = null
+      let speechEndTimer = null
+
+      function notify() { listeners.forEach((fn) => { try { fn() } catch (_e) { /* noop */ } }) }
+      function set(next, extra) { if (extra !== undefined) Object.assign(store, extra); store.state = next; notify() }
+      function dispatch(event) {
+        const row = TRANSITIONS[store.state] && TRANSITIONS[store.state][event]
+        if (row === undefined) return false
+        if (row.effect !== undefined && EFFECTS[row.effect] !== undefined) EFFECTS[row.effect]()
+        else set(row.to)
+        return true
+      }
+
+      function createRecognition() {
+        const SR = typeof SpeechRecognition !== 'undefined' ? SpeechRecognition : (typeof webkitSpeechRecognition !== 'undefined' ? webkitSpeechRecognition : null)
+        if (SR === null) return null
+        const rec = new SR()
+        rec.lang = 'zh-CN'
+        rec.continuous = true
+        rec.interimResults = true
+        return rec
+      }
+
+      function setDraft(text) {
+        const a = store.inputActions
+        if (a && typeof a.setDraft === 'function') { try { a.setDraft(text) } catch (_e) { /* noop */ } }
+      }
+
+      function finishSilently() {
+        set('IDLE')
+        const text = store.finalText.trim()
+        store.stopDraft = typeof store.currentDraft === 'string' ? store.currentDraft : text
+        if (text.length === 0) return
+        proofreadRemote(text).then((cleaned) => {
+          // Only replace the draft when the user did not touch it meanwhile.
+          if (store.state === 'IDLE' && store.currentDraft === store.stopDraft) setDraft(cleaned)
+        })
+      }
+
+      const EFFECTS = {
+        startRec() {
+          const rec = createRecognition()
+          if (rec === null) { set('ERROR', { errorText: '当前浏览器不支持语音识别（请使用 Edge 或 Chrome）' }); return }
+          recognition = rec
+          store.finalText = ''
+          store.baseDraft = typeof store.currentDraft === 'string' ? store.currentDraft : ''
+          store.stopDraft = null
+          rec.onstart = () => dispatch('started')
+          rec.onresult = (e) => {
+            if (store.state !== 'RECORDING' && store.state !== 'REQUESTING') return
+            if (speechEndTimer !== null) { clearTimeout(speechEndTimer); speechEndTimer = null }
+            let interim = ''
+            for (let i = e.resultIndex; i < e.results.length; i += 1) {
+              const r = e.results[i]
+              if (r.isFinal) store.finalText += r[0].transcript
+              else interim += r[0].transcript
+            }
+            const base = store.baseDraft.trim()
+            const body = (store.finalText + interim).trim()
+            setDraft(base.length > 0 ? base + ' ' + body : body)
+          }
+          rec.onspeechend = () => {
+            if (speechEndTimer !== null) clearTimeout(speechEndTimer)
+            speechEndTimer = setTimeout(() => {
+              speechEndTimer = null
+              if (store.state === 'RECORDING') { try { rec.stop() } catch (_e) { /* noop */ } }
+            }, 1500)
+          }
+          rec.onerror = (e) => {
+            const err = e && e.error
+            if (err === 'not-allowed' || err === 'service-not-allowed') {
+              set('ERROR', { errorText: '麦克风权限被拒绝，请在浏览器设置中允许' })
+            } else if (err === 'no-speech') {
+              if (store.state === 'RECORDING' || store.state === 'REQUESTING') set('IDLE')
+            } else if (err === 'aborted') {
+              /* expected on manual stop */
+            } else if (store.state === 'RECORDING' || store.state === 'REQUESTING') {
+              set('ERROR', { errorText: '语音识别出错：' + (err || 'unknown') })
+            }
+          }
+          rec.onend = () => {
+            if (speechEndTimer !== null) { clearTimeout(speechEndTimer); speechEndTimer = null }
+            recognition = null
+            if (store.state === 'RECORDING') { dispatch('ended'); return }
+            if (store.state === 'REQUESTING') set('IDLE')
+          }
+          set('REQUESTING')
+          try { rec.start() } catch (_e) { set('ERROR', { errorText: '语音识别启动失败' }) }
+        },
+        recError() { /* errorText already set */ },
+        stopRec() {
+          if (recognition !== null) { try { recognition.stop() } catch (_e) { /* noop */ } }
+          finishSilently()
+        },
+        autoFinalize() {
+          finishSilently()
+        },
+      }
+
+      function handleKey(e) {
+        if (!e) return
+        const key = String(e.key || '').toLowerCase()
+        const code = String(e.code || '').toLowerCase()
+        const isM = key === 'm' || code === 'keym'
+        if (!(e.ctrlKey && e.shiftKey && isM)) return
+        if (typeof store.sessionId !== 'string' || store.sessionId.length === 0) return
+        if (e.preventDefault) { try { e.preventDefault() } catch (_e) { /* noop */ } }
+        if (store.state === 'IDLE' || store.state === 'ERROR') dispatch('start')
+        else if (store.state === 'RECORDING') dispatch('stop')
+      }
+
+      ctx.effect(() => () => {
+        if (speechEndTimer !== null) { clearTimeout(speechEndTimer); speechEndTimer = null }
+        if (recognition !== null) { try { recognition.abort() } catch (_e) { /* noop */ } recognition = null }
+        listeners.length = 0
+        store.state = 'IDLE'
+      })
+      document.addEventListener('keydown', handleKey)
+      ctx.effect(() => () => document.removeEventListener('keydown', handleKey))
+
+      function useStore() {
+        const [tick, force] = React.useState(0)
+        React.useEffect(() => {
+          const cb = () => force((n) => n + 1)
+          listeners.push(cb)
+          return () => { const i = listeners.indexOf(cb); if (i >= 0) listeners.splice(i, 1) }
+        }, [])
+        return store
+      }
+
+      function VoiceButton(props) {
+        const s = useStore()
+        if (props) {
+          if (typeof props.sessionId === 'string') store.sessionId = props.sessionId
+          if (props.inputActions && typeof props.inputActions.setDraft === 'function') store.inputActions = props.inputActions
+          if (typeof props.useInput === 'function') {
+            try {
+              const inputState = props.useInput()
+              if (inputState && typeof inputState.draft === 'string') store.currentDraft = inputState.draft
+            } catch (_e) { /* noop */ }
+          }
+        }
+        const st = s.state
+        if (st === 'IDLE' || st === 'ERROR') {
+          return React.createElement('button', { className: 'vc-btn vc-cell', title: '语音输入（Ctrl+Shift+M）', onClick: () => dispatch('start') }, React.createElement(MicIcon, null))
+        }
+        if (st === 'RECORDING') {
+          return React.createElement('button', { className: 'vc-btn vc-rec vc-cell', title: '停止（Ctrl+Shift+M）', onClick: () => dispatch('stop') }, React.createElement(StopIcon, null))
+        }
+        return React.createElement('span', { className: 'vc-spin vc-cell', style: { display: 'inline-flex', margin: 6 } })
+      }
+
+      function VoiceDock() {
+        const s = useStore()
+        if (s.state !== 'ERROR') return null
+        return React.createElement('div', { className: 'vc-dock-line vc-error' }, s.errorText || '出错了')
+      }
+
+      slots.inject('conversation.input.right', () => slots.register(
+        { name: 'conversation.input.right', id: 'voice-record', order: 10, label: '语音输入' },
+        (props) => React.createElement(VoiceButton, props),
+      ))
+      slots.inject('conversation.input.dock', () => slots.register(
+        { name: 'conversation.input.dock', id: 'voice-dock', order: 0, label: '语音识别错误' },
+        (props) => React.createElement(VoiceDock, props),
+      ))
+    }
+
+    exports.apply = apply;
+    exports.inject = ['slots'];
+
+    return module.exports;
+  },
+});

package/lib/index.js

@@ -0,0 +1,135 @@
+// dsh-voice-input — static host half (cordis plugin, ESM).
+//
+// Adapted from the dynamic plugin form for the STATIC plugin shape:
+//   - the dynamic plugin's harness.handle('voice.proofread') RPC is replaced
+//     by a webServer exact POST route (/dsh-voice-input/proofread);
+//   - the proofread prompt, stream collection and degradation semantics are
+//     kept identical to the dynamic form: any missing service, missing model
+//     or failed call degrades to "return the raw transcript unchanged".
+//
+// The route is deliberately small: the Client half owns speech recognition
+// (browser Web Speech API) and only asks the Host to proofread one string.
+export const name = 'dsh-voice-input'
+
+/** Hard dependency: the host HTTP carrier that owns the route table. */
+export const inject = ['webServer']
+
+const PROOFREAD_SYSTEM = [
+  '你是专业中文校对助手。请对用户提供的语音转写文本进行校对：',
+  '1. 修正错别字；',
+  '2. 修正因读音相同或相近导致的转录偏差；',
+  '3. 将繁体字统一转换为简体中文；',
+  '4. 理顺标点符号与断句；',
+  '5. 使表达通顺自然。',
+  '要求：不改变原意、不增删实质信息、不添加解释或评论；只输出校对后的文本本身。',
+].join('')
+
+/** Collect the request body as UTF-8 text (bounded to 256 KiB). */
+function readBody(req) {
+  return new Promise((resolve, reject) => {
+    const chunks = []
+    let size = 0
+    req.on('data', (chunk) => {
+      size += chunk.length
+      if (size > 262144) {
+        reject(new Error('body too large'))
+        req.destroy()
+        return
+      }
+      chunks.push(chunk)
+    })
+    req.on('end', () => resolve(Buffer.concat(chunks).toString('utf8')))
+    req.on('error', reject)
+  })
+}
+
+function sendJson(res, status, data) {
+  if (res.headersSent) return
+  res.writeHead(status, { 'content-type': 'application/json; charset=utf-8' })
+  res.end(JSON.stringify(data))
+}
+
+/**
+ * Proofread one transcript with the session's default model.
+ * Never fails the caller: on any degradation the raw text is returned.
+ */
+async function proofread(ctx, text) {
+  if (text === '') return { ok: true, text: '', skipped: true }
+  const llm = ctx.get('llm')
+  if (llm === undefined) return { ok: true, text: text, skipped: true }
+  const model = ctx.get('agentDefaultModel')
+  let provider = ''
+  let modelId = ''
+  try {
+    if (model !== undefined) {
+      const sel = model.currentSelection()
+      if (sel && typeof sel.provider === 'string') provider = sel.provider
+      if (sel && typeof sel.model === 'string') modelId = sel.model
+    }
+  } catch (_e) { /* fall through to directory lookup */ }
+  if (provider === '') {
+    try {
+      const ps = llm.listProviders()
+      if (ps && ps.length > 0 && ps[0] && typeof ps[0].id === 'string') provider = ps[0].id
+    } catch (_e) { /* noop */ }
+  }
+  if (provider !== '' && modelId === '') {
+    try {
+      const ms = await llm.listModels(provider)
+      if (ms && ms.length > 0 && ms[0] && typeof ms[0].id === 'string') modelId = ms[0].id
+    } catch (_e) { /* noop */ }
+  }
+  if (provider === '' || modelId === '') return { ok: true, text: text, skipped: true }
+  try {
+    const stream = llm.stream({
+      provider: provider,
+      model: modelId,
+      system: PROOFREAD_SYSTEM,
+      temperature: 0,
+      maxTokens: 4000,
+      messages: [{
+        id: 'voice-proof-user',
+        role: 'user',
+        content: [{ type: 'text', text: text }],
+        source: { kind: 'plugin', plugin: 'dsh-voice-input' },
+      }],
+    })
+    let out = ''
+    for await (const chunk of stream) {
+      if (chunk && chunk.type === 'text-delta' && typeof chunk.text === 'string') out += chunk.text
+      else if (chunk && chunk.type === 'block-end' && chunk.block && chunk.block.type === 'text' && typeof chunk.block.text === 'string') out = chunk.block.text
+    }
+    const cleaned = out.trim()
+    return cleaned.length > 0 ? { ok: true, text: cleaned, skipped: false } : { ok: true, text: text, skipped: true }
+  } catch (err) {
+    return {
+      ok: true,
+      text: text,
+      skipped: true,
+      error: (err && err.message) ? String(err.message) : String(err),
+    }
+  }
+}
+
+export function apply(ctx) {
+  ctx.effect(() => ctx.webServer.register({
+    kind: 'exact',
+    path: '/dsh-voice-input/proofread',
+    handler: async (req, res) => {
+      if (req.method !== 'POST') {
+        sendJson(res, 405, { ok: false, error: 'method not allowed' })
+        return
+      }
+      let payload
+      try {
+        payload = JSON.parse(await readBody(req))
+      } catch (_e) {
+        sendJson(res, 400, { ok: false, error: 'invalid JSON body' })
+        return
+      }
+      const text = (payload && typeof payload.text === 'string') ? payload.text.trim() : ''
+      const result = await proofread(ctx, text)
+      sendJson(res, 200, result)
+    },
+  }), 'dsh-voice-input: /dsh-voice-input/proofread route')
+}

package/package.json

@@ -0,0 +1,56 @@
+{
+  "name": "@sakka6868/dsh-voice-input",
+  "version": "0.1.0",
+  "description": "DeepSeek Harness Web UI 语音输入插件：发送按钮左侧的麦克风按钮，调用浏览器 Web Speech API（Edge 中即微软语音服务）实时把语音转成文字写入输入框，停止后自动校对（错别字/同音偏差/繁体转简体/标点）。支持 Ctrl+Shift+M 快捷键，零依赖、无需 API Key。",
+  "license": "MIT",
+  "type": "module",
+  "main": "./lib/index.js",
+  "exports": {
+    ".": {
+      "default": "./lib/index.js"
+    },
+    "./client": {
+      "default": "./lib/client.js"
+    },
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "lib",
+    "README.md",
+    "LICENSE"
+  ],
+  "engines": {
+    "node": ">=20"
+  },
+  "dsh": {
+    "client": {
+      "platform": "web",
+      "immediately": false,
+      "inject": [
+        "@deepseek-ai/dsh-client-runtime",
+        "@deepseek-ai/dsh-client-ui-slots"
+      ]
+    }
+  },
+  "peerDependencies": {
+    "@deepseek-ai/cordis": "^4.0.1",
+    "@deepseek-ai/dsh-client-runtime": ">=0.1.0-rc.6",
+    "@deepseek-ai/dsh-client-ui-slots": ">=0.1.0-rc.6",
+    "react": "^18.2.0"
+  },
+  "keywords": [
+    "dsh",
+    "deepseek-harness",
+    "cordis",
+    "plugin",
+    "client-plugin",
+    "voice",
+    "speech-to-text",
+    "stt",
+    "dictation",
+    "web-speech-api"
+  ],
+  "publishConfig": {
+    "access": "public"
+  }
+}