优化语音输入/普通文本输入

2026-09-03 13:36:40 +08:00
parent 009d8ca73e
commit 32d179143f
26 changed files with 2318 additions and 53 deletions
@@ -416,6 +416,15 @@
    return [...unique.values()];
  }

+  function parsedCommandSourceLabel(source) {
+    return {
+      direct_control: "设备控件",
+      saved_mode_command_flow: "模式卡片",
+      local_rule: "本地规则",
+      speech_semantic: "语音语义",
+    }[source] || "云端模型";
+  }
+
  function renderParsedCanCommands(result, { message = "" } = {}) {
    const panel = $("#v2CanCommandPanel");
    const list = $("#v2CanCommandList");
@@ -429,7 +438,7 @@
        <li data-can-channel="${escapeHtml(item.channel)}" data-delivery-state="${escapeHtml(item.deliveryState)}">
          <span>
            <b class="v2-can-command-channel">${escapeHtml(item.channel)}</b>
-            ${item.parseSource ? `<b class="v2-can-command-source" data-parse-source="${escapeHtml(item.parseSource)}">${item.parseSource === "direct_control" ? "设备控件" : item.parseSource === "saved_mode_command_flow" ? "模式卡片" : item.parseSource === "local_rule" ? "本地规则" : "云端模型"}</b>` : ""}
+            ${item.parseSource ? `<b class="v2-can-command-source" data-parse-source="${escapeHtml(item.parseSource)}">${parsedCommandSourceLabel(item.parseSource)}</b>` : ""}
            <b class="v2-can-command-delivery" data-delivery-state="${escapeHtml(item.deliveryState)}" title="${escapeHtml(`${item.deliveryTransport} · ${item.deliveryLabel}`)}">${escapeHtml(item.deliveryLabel)}</b>
            <i class="v2-can-command-category">${escapeHtml(item.category)}</i>
          </span>
@@ -19,10 +19,10 @@
    <script src="/vendor/three/0.147.0/FBXLoader.js" defer></script>
    <script src="/network.js?v=android-pad-2" defer></script>
    <script src="/app.js?v=dual-seat-114" defer></script>
-    <script src="/speech-input.js?v=sensevoice-6" defer></script>
+    <script src="/speech-input.js?v=sensevoice-7" defer></script>
    <script type="module" src="/gesture-control.js?v=android-pad-9"></script>
    <script src="/pwa.js?v=android-pad-3" defer></script>
-    <script src="/console-v2.js?v=delivery-state-54" defer></script>
+    <script src="/console-v2.js?v=delivery-state-55" defer></script>
    <script src="/self-extension.js?v=gscu-tcp-41" defer></script>
  </head>
  <body class="console-v2">
@@ -1,4 +1,4 @@
-const CACHE_VERSION = "cognitive-agent-pad-v227";
+const CACHE_VERSION = "cognitive-agent-pad-v228";
 const SHELL_CACHE = `${CACHE_VERSION}-shell`;
 const RUNTIME_CACHE = `${CACHE_VERSION}-runtime`;
 const APP_SHELL = [
@@ -10,11 +10,11 @@ const APP_SHELL = [
  "/gesture-control.js?v=android-pad-9",
  "/pwa.js?v=android-pad-3",
  "/console-v2.css?v=delivery-state-50",
-  "/console-v2.js?v=delivery-state-54",
+  "/console-v2.js?v=delivery-state-55",
  "/self-extension.css?v=gscu-tcp-42",
  "/self-extension.js?v=gscu-tcp-41",
  "/speech-input.css?v=sensevoice-5",
-  "/speech-input.js?v=sensevoice-6",
+  "/speech-input.js?v=sensevoice-7",
  "/speech-audio-worklet.js?v=sensevoice-4",
  "/assets/pets/self-extension-cat.png?v=3",
  "/assets/pets/self-extension-cat-entrance.png?v=3",
@@ -380,15 +380,21 @@
    const { input } = elements();
    if (!input) throw new Error("没有找到一句话编排输入框");
    const cleanText = String(result.clean_text || result.text || "").trim();
+    const peripheralEligible = result.peripheral_routing?.execution_eligible === true;
+    const inputText = peripheralEligible
+      ? String(result.routing_text || result.raw_text || cleanText).trim()
+      : cleanText;
    const commands = Array.isArray(result.commands) ? result.commands : [];
-    if (!cleanText || commands.length === 0) {
+    if (!inputText || (commands.length === 0 && !peripheralEligible)) {
      window.dispatchEvent(new CustomEvent("cognitive-agent:speech-result", { detail: result }));
      return false;
    }
    speech.applyingResult = true;
-    input.value = cleanText.slice(0, input.maxLength || 1000);
+    input.value = inputText.slice(0, input.maxLength || 1000);
    input.dataset.speechUncertain = String(result.uncertain === true);
-    input.dataset.speechExecutionEligible = String(result.execution_eligible === true);
+    input.dataset.speechExecutionEligible = String(
+      result.execution_eligible === true || peripheralEligible,
+    );
    input.dataset.speechSemanticStatus = String(result.semantic_status || "");
    input.dataset.speechValue = input.value.trim();
    input.dataset.speechUtteranceId = String(result.utterance_id || "");
@@ -423,6 +429,10 @@
        if (!applied) {
          setPhase("success", `未形成可执行控制指令 · ${latency} ms`);
          announce("已保留原始转录，但没有激活控制指令");
+        } else if (result.peripheral_routing?.execution_eligible === true) {
+          const count = Number(result.peripheral_routing.action_count || 1);
+          setPhase("success", `已识别为 ${count} 条外设控制指令 · ${latency} ms`);
+          announce("语音已匹配当前 NFC 外设的已学习控制指令");
        } else if (result.semantic_status === "recognized_unsupported") {
          setPhase("success", `已识别，但当前车辆能力不支持 · ${latency} ms`);
          announce("已完整保留用户语义，但当前车辆能力不支持执行");
@@ -0,0 +1,30 @@
+# 设备能力卡片实验 Demo
+
+这是一个完全独立的静态实验，不导入 Cognitive Agent 的现有前端或后端模块，
+也不会调用 API、执行器或 CAN 通道。
+
+## 运行
+
+```bash
+cd experiments/device-capability-card-demo
+python3 -m http.server 8877 --bind 127.0.0.1
+```
+
+浏览器访问：<http://127.0.0.1:8877/>
+
+## 测试
+
+```bash
+cd experiments/device-capability-card-demo
+npm test
+```
+
+## 演示内容
+
+- 实时修改设备名称、ID、领域、别名、目标范围和执行通道；
+- 启用或停用能力，修改各项功能的自然语言说法；
+- 根据能力定义自动生成设备功能卡片；
+- 将自然语言转换为结构化控制命令；
+- 展示设备匹配、功能匹配、范围解析、参数校验和执行器绑定轨迹；
+- 对数值越界、未命中设备、功能停用等情况执行失败关闭；
+- 所有执行载荷均带有 `demo_only: true`。
@@ -0,0 +1,288 @@
+import {
+  DEFAULT_DESCRIPTOR,
+  buildManualCommand,
+  cloneDescriptor,
+  parseInstruction,
+  splitAliases,
+} from "./parser-engine.js";
+
+const $ = (selector, root = document) => root.querySelector(selector);
+const $$ = (selector, root = document) => [...root.querySelectorAll(selector)];
+
+let descriptor = cloneDescriptor();
+let latestPayload = null;
+
+const categoryLabels = {
+  vehicle_electrical: "车端电器",
+  seat: "座椅设备",
+  peripheral: "扩展外设",
+};
+
+const transportLabels = {
+  native_can: "原生 CAN",
+  domain_runtime: "领域执行器",
+  gscu_tcp: "红外网关",
+  mqtt: "MQTT",
+};
+
+const exampleTemplates = [
+  (name) => `开启${name}`,
+  (name) => `把${name}温度调到4度`,
+  (name) => `将${name}设为速冷模式`,
+  (name) => `关闭${name}`,
+  (name) => `把${name}调到零下10度`,
+];
+
+function escapeHtml(value) {
+  return String(value ?? "")
+    .replaceAll("&", "&amp;")
+    .replaceAll("<", "&lt;")
+    .replaceAll(">", "&gt;")
+    .replaceAll('"', "&quot;")
+    .replaceAll("'", "&#039;");
+}
+
+function fillIdentityForm() {
+  $("#deviceName").value = descriptor.device.name;
+  $("#deviceId").value = descriptor.device.id;
+  $("#deviceDomain").value = descriptor.device.domain;
+  $("#deviceCategory").value = descriptor.device.category;
+  $("#deviceAliases").value = descriptor.device.aliases.join("、");
+  $("#deviceScope").value = descriptor.device.defaultScope;
+  $("#deviceTransport").value = descriptor.device.transport;
+}
+
+function syncIdentityFromForm() {
+  descriptor.device.name = $("#deviceName").value.trim() || "未命名设备";
+  descriptor.device.id = $("#deviceId").value.trim() || "unnamed_device";
+  descriptor.device.domain = $("#deviceDomain").value.trim() || "custom";
+  descriptor.device.category = $("#deviceCategory").value;
+  descriptor.device.aliases = splitAliases($("#deviceAliases").value);
+  descriptor.device.defaultScope = $("#deviceScope").value;
+  descriptor.device.targetScopes = [descriptor.device.defaultScope];
+  descriptor.device.transport = $("#deviceTransport").value;
+  descriptor.device.targetDevice = descriptor.device.id;
+  renderAll({ preserveResult: true });
+}
+
+function renderCapabilityEditor() {
+  $("#capabilityEditor").innerHTML = descriptor.capabilities.map((capability, capabilityIndex) => `
+    <section class="capability-definition" data-capability-index="${capabilityIndex}">
+      <div class="capability-title">
+        <div>
+          <strong>${escapeHtml(capability.label)}</strong>
+          <code>${escapeHtml(capability.id)}</code>
+        </div>
+        <label class="toggle-label">
+          <input type="checkbox" data-capability-enabled ${capability.enabled !== false ? "checked" : ""} />
+          启用
+        </label>
+      </div>
+      <div class="intent-list">
+        ${capability.intents.map((intent, intentIndex) => `
+          <label class="intent-alias-field">
+            <span>${escapeHtml(intent.label)}</span>
+            <input
+              data-intent-index="${intentIndex}"
+              value="${escapeHtml(intent.phrases.join("、"))}"
+              aria-label="${escapeHtml(capability.label + intent.label)}的自然语言说法"
+            />
+          </label>
+        `).join("")}
+      </div>
+    </section>
+  `).join("");
+}
+
+function renderExamples() {
+  $("#exampleCommands").innerHTML = exampleTemplates.map((makeExample) => {
+    const utterance = makeExample(descriptor.device.name);
+    return `<button type="button" data-example="${escapeHtml(utterance)}">${escapeHtml(utterance)}</button>`;
+  }).join("");
+}
+
+function controlMarkup(capability) {
+  const ui = capability.ui;
+  if (ui.type === "select") {
+    return `
+      <label class="card-control">
+        <span>${escapeHtml(capability.label)}</span>
+        <select data-manual-capability="${escapeHtml(capability.id)}">
+          ${ui.options.map(([value, label]) => `<option value="${escapeHtml(value)}"${String(value) === String(ui.value) ? " selected" : ""}>${escapeHtml(label)}</option>`).join("")}
+        </select>
+      </label>
+    `;
+  }
+  return `
+    <label class="card-control">
+      <span>${escapeHtml(capability.label)} <output>${escapeHtml(ui.value)}${escapeHtml(ui.unit || "")}</output></span>
+      <input
+        type="range"
+        min="${ui.minimum}"
+        max="${ui.maximum}"
+        step="${ui.step}"
+        value="${ui.value}"
+        data-manual-capability="${escapeHtml(capability.id)}"
+        data-unit="${escapeHtml(ui.unit || "")}"
+      />
+    </label>
+  `;
+}
+
+function renderDeviceCard() {
+  const enabledCapabilities = descriptor.capabilities.filter((item) => item.enabled !== false);
+  const category = categoryLabels[descriptor.device.category] || "自定义设备";
+  const transport = transportLabels[descriptor.device.transport] || descriptor.device.transport;
+  const icon = descriptor.device.category === "seat" ? "◫" : descriptor.device.category === "peripheral" ? "⌁" : "❄";
+  $("#deviceCardPreview").innerHTML = `
+    <header class="device-card-header">
+      <div class="device-identity">
+        <span class="device-icon" aria-hidden="true">${icon}</span>
+        <div>
+          <h3>${escapeHtml(descriptor.device.name)}</h3>
+          <p>${escapeHtml(category)} · ${escapeHtml(transport)} · ${escapeHtml(descriptor.device.defaultScope)}</p>
+        </div>
+      </div>
+      <span class="connection-state"><i></i>实验就绪</span>
+    </header>
+    <div class="readout-grid">
+      ${descriptor.readouts.map((readout) => `<div><span>${escapeHtml(readout.label)}</span><strong>${escapeHtml(readout.value)}</strong></div>`).join("")}
+    </div>
+    <div class="card-controls">
+      ${enabledCapabilities.length ? enabledCapabilities.map(controlMarkup).join("") : "<p>没有已启用功能。</p>"}
+    </div>
+    <p class="card-footnote">功能控件只生成结构化演示数据；执行映射为虚拟引用，不连接现有服务。</p>
+  `;
+}
+
+function compactDescriptor() {
+  return {
+    version: descriptor.version,
+    device: descriptor.device,
+    capabilities: descriptor.capabilities.map((capability) => ({
+      id: capability.id,
+      property: capability.property,
+      target_scopes: descriptor.device.targetScopes,
+      parameters: capability.parameters,
+      intents: capability.intents,
+      risk_level: capability.risk,
+      confirmation_policy: capability.confirmationPolicy,
+      executor: capability.executor,
+      enabled: capability.enabled,
+    })),
+  };
+}
+
+function renderDescriptor() {
+  $("#descriptorPreview").textContent = JSON.stringify(compactDescriptor(), null, 2);
+}
+
+function renderAll({ preserveResult = false } = {}) {
+  renderCapabilityEditor();
+  renderExamples();
+  renderDeviceCard();
+  renderDescriptor();
+  if (!preserveResult) renderResult(null);
+}
+
+function resultPresentation(result) {
+  if (!result) return { status: "idle", title: "等待解析", message: "输入一句话，查看设备、功能和参数的匹配过程。" };
+  if (result.source === "card_control") {
+    return {
+      status: "success",
+      title: "已生成卡片控制载荷",
+      message: `已生成 ${result.commands.length} 条结构化演示指令；实验页面不会发送到车辆。`,
+    };
+  }
+  if (result.semantic_status === "recognized_executable") {
+    return { status: "success", title: "已匹配可执行功能", message: `${result.clean_text} 当前仅生成演示载荷。` };
+  }
+  if (result.semantic_status === "ambiguous" || result.semantic_status === "incomplete") {
+    return { status: "warning", title: "需要补充信息", message: result.issues.join("；") };
+  }
+  return { status: "error", title: "未生成控制指令", message: result.issues.join("；") || "没有匹配到设备能力。" };
+}
+
+function renderResult(result) {
+  const presentation = resultPresentation(result);
+  $("#resultBanner").dataset.status = presentation.status;
+  $("#resultTitle").textContent = presentation.title;
+  $("#resultMessage").textContent = presentation.message;
+  $("#traceList").innerHTML = result?.trace?.length
+    ? result.trace.map((item) => `<li data-state="${escapeHtml(item.state)}"><strong>${escapeHtml(item.stage)}</strong>${escapeHtml(item.detail)}</li>`).join("")
+    : "<li><strong>尚无轨迹</strong>解析后将在这里显示每一步决策。</li>";
+  latestPayload = result;
+  $("#payloadOutput").textContent = JSON.stringify(result || {}, null, 2);
+  $("#copyPayload").disabled = !result;
+}
+
+function parseCurrentUtterance() {
+  renderResult(parseInstruction($("#utterance").value, descriptor));
+}
+
+function bindEvents() {
+  ["deviceName", "deviceId", "deviceDomain", "deviceCategory", "deviceAliases", "deviceScope", "deviceTransport"]
+    .forEach((id) => $("#" + id).addEventListener("input", syncIdentityFromForm));
+
+  $("#capabilityEditor").addEventListener("input", (event) => {
+    const section = event.target.closest("[data-capability-index]");
+    if (!section) return;
+    const capability = descriptor.capabilities[Number(section.dataset.capabilityIndex)];
+    if (event.target.matches("[data-capability-enabled]")) {
+      capability.enabled = event.target.checked;
+    } else if (event.target.matches("[data-intent-index]")) {
+      capability.intents[Number(event.target.dataset.intentIndex)].phrases = splitAliases(event.target.value);
+    }
+    renderDeviceCard();
+    renderDescriptor();
+  });
+
+  $("#exampleCommands").addEventListener("click", (event) => {
+    const button = event.target.closest("[data-example]");
+    if (!button) return;
+    $("#utterance").value = button.dataset.example;
+    parseCurrentUtterance();
+  });
+
+  $("#parseCommand").addEventListener("click", parseCurrentUtterance);
+  $("#utterance").addEventListener("keydown", (event) => {
+    if ((event.ctrlKey || event.metaKey) && event.key === "Enter") parseCurrentUtterance();
+  });
+
+  $("#deviceCardPreview").addEventListener("input", (event) => {
+    if (!event.target.matches("[data-manual-capability]")) return;
+    const output = event.target.closest(".card-control")?.querySelector("output");
+    if (output) output.textContent = `${event.target.value}${event.target.dataset.unit || ""}`;
+    const values = Object.fromEntries($$("[data-manual-capability]", $("#deviceCardPreview"))
+      .map((input) => [input.dataset.manualCapability, input.value]));
+    const result = {
+      source: "card_control",
+      semantic_status: "recognized_executable",
+      execution_eligible: false,
+      note: "实验页面不允许真实执行",
+      commands: buildManualCommand(descriptor, values),
+    };
+    renderResult(result);
+  });
+
+  $("#copyPayload").addEventListener("click", async () => {
+    if (!latestPayload) return;
+    await navigator.clipboard.writeText(JSON.stringify(latestPayload, null, 2));
+    const button = $("#copyPayload");
+    button.textContent = "已复制";
+    window.setTimeout(() => { button.textContent = "复制"; }, 1200);
+  });
+
+  $("#resetDemo").addEventListener("click", () => {
+    descriptor = cloneDescriptor(DEFAULT_DESCRIPTOR);
+    fillIdentityForm();
+    $("#utterance").value = "把冰箱温度调到4度";
+    renderAll();
+    parseCurrentUtterance();
+  });
+}
+
+fillIdentityForm();
+renderAll();
+bindEvents();
+parseCurrentUtterance();
@@ -0,0 +1,149 @@
+<!doctype html>
+<html lang="zh-CN">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <meta name="color-scheme" content="light dark" />
+    <title>设备能力卡片实验室</title>
+    <link rel="stylesheet" href="./styles.css" />
+  </head>
+  <body>
+    <main class="demo-shell">
+      <header class="page-header">
+        <div>
+          <p class="eyebrow">独立实验 · 不连接车辆</p>
+          <h1>设备能力卡片实验室</h1>
+          <p class="page-intro">修改设备名称、别名和功能说法，观察自然语言如何匹配为结构化控制指令。</p>
+        </div>
+        <button class="secondary-button" id="resetDemo" type="button">恢复示例</button>
+      </header>
+
+      <section class="safety-note" aria-label="实验边界">
+        <span aria-hidden="true">◇</span>
+        <p><strong>安全边界</strong> 本页面只在浏览器内生成演示数据，不导入当前项目模块，也不会调用控制接口或发送 CAN 指令。</p>
+      </section>
+
+      <div class="workspace-grid">
+        <aside class="configuration-panel" aria-labelledby="configurationTitle">
+          <div class="section-heading">
+            <span>01</span>
+            <h2 id="configurationTitle">设备定义</h2>
+          </div>
+
+          <div class="field-grid">
+            <label class="field">
+              <span>设备名称</span>
+              <input id="deviceName" autocomplete="off" />
+            </label>
+            <label class="field">
+              <span>设备 ID</span>
+              <input id="deviceId" autocomplete="off" spellcheck="false" />
+            </label>
+            <label class="field">
+              <span>所属领域</span>
+              <input id="deviceDomain" autocomplete="off" spellcheck="false" />
+            </label>
+            <label class="field">
+              <span>设备类型</span>
+              <select id="deviceCategory">
+                <option value="vehicle_electrical">车端电器</option>
+                <option value="seat">座椅设备</option>
+                <option value="peripheral">扩展外设</option>
+              </select>
+            </label>
+            <label class="field field-wide">
+              <span>设备别名</span>
+              <input id="deviceAliases" autocomplete="off" placeholder="使用顿号或逗号分隔" />
+              <small>解析时优先匹配最长别名；避免使用“灯、开关、设备”等过宽名称。</small>
+            </label>
+            <label class="field">
+              <span>默认目标</span>
+              <select id="deviceScope">
+                <option value="vehicle">全车</option>
+                <option value="driver">主驾</option>
+                <option value="passenger">副驾</option>
+                <option value="rear_row">后排</option>
+              </select>
+            </label>
+            <label class="field">
+              <span>执行通道</span>
+              <select id="deviceTransport">
+                <option value="native_can">原生 CAN</option>
+                <option value="domain_runtime">领域执行器</option>
+                <option value="gscu_tcp">红外网关</option>
+                <option value="mqtt">MQTT</option>
+              </select>
+            </label>
+          </div>
+
+          <div class="section-heading compact-heading">
+            <span>02</span>
+            <h2>功能定义</h2>
+          </div>
+          <div id="capabilityEditor" class="capability-editor"></div>
+
+          <details class="schema-details">
+            <summary>查看卡片声明</summary>
+            <pre id="descriptorPreview" aria-label="设备卡片声明"></pre>
+          </details>
+        </aside>
+
+        <section class="experiment-panel">
+          <div class="pipeline" aria-label="匹配流程">
+            <span>设备别名</span><i aria-hidden="true">→</i>
+            <span>功能说法</span><i aria-hidden="true">→</i>
+            <span>参数校验</span><i aria-hidden="true">→</i>
+            <span>执行载荷</span>
+          </div>
+
+          <section class="preview-section" aria-labelledby="previewTitle">
+            <div class="section-heading">
+              <span>03</span>
+              <h2 id="previewTitle">功能卡片</h2>
+            </div>
+            <article class="device-card" id="deviceCardPreview"></article>
+          </section>
+
+          <section class="parser-section" aria-labelledby="parserTitle">
+            <div class="section-heading">
+              <span>04</span>
+              <h2 id="parserTitle">指令解析</h2>
+            </div>
+
+            <div class="example-row" id="exampleCommands" aria-label="示例指令"></div>
+
+            <label class="command-composer">
+              <span class="sr-only">自然语言控制指令</span>
+              <textarea id="utterance" rows="2" maxlength="180">把冰箱温度调到4度</textarea>
+              <button class="primary-button" id="parseCommand" type="button">解析指令</button>
+            </label>
+
+            <div class="result-banner" id="resultBanner" data-status="idle" aria-live="polite">
+              <span class="result-dot" aria-hidden="true"></span>
+              <div>
+                <strong id="resultTitle">等待解析</strong>
+                <p id="resultMessage">输入一句话，查看设备、功能和参数的匹配过程。</p>
+              </div>
+            </div>
+
+            <div class="result-grid">
+              <section class="trace-panel" aria-labelledby="traceTitle">
+                <h3 id="traceTitle">匹配轨迹</h3>
+                <ol id="traceList"></ol>
+              </section>
+              <section class="payload-panel" aria-labelledby="payloadTitle">
+                <div class="payload-heading">
+                  <h3 id="payloadTitle">结构化结果</h3>
+                  <button id="copyPayload" class="text-button" type="button" disabled>复制</button>
+                </div>
+                <pre id="payloadOutput">{}</pre>
+              </section>
+            </div>
+          </section>
+        </section>
+      </div>
+    </main>
+
+    <script type="module" src="./app.js"></script>
+  </body>
+</html>
@@ -0,0 +1,8 @@
+{
+  "name": "device-capability-card-demo",
+  "private": true,
+  "type": "module",
+  "scripts": {
+    "test": "node parser-engine.test.mjs"
+  }
+}
@@ -0,0 +1,353 @@
+export const DEFAULT_DESCRIPTOR = {
+  version: 1,
+  device: {
+    id: "cabin_refrigerator",
+    name: "车载冰箱",
+    aliases: ["冰箱", "车载冷藏箱", "冷藏箱"],
+    domain: "refrigerator",
+    category: "vehicle_electrical",
+    targetDevice: "refrigerator",
+    targetScopes: ["vehicle"],
+    defaultScope: "vehicle",
+    transport: "native_can",
+    channel: "实验通道",
+  },
+  readouts: [
+    { key: "power", label: "工作状态", value: "制冷中" },
+    { key: "temperature", label: "当前温度", value: "6℃" },
+    { key: "mode", label: "当前模式", value: "节能" },
+  ],
+  capabilities: [
+    {
+      id: "refrigerator.set_power",
+      label: "工作状态",
+      property: "refrigerator.power",
+      operation: "set",
+      enabled: true,
+      risk: "low",
+      confirmationPolicy: "on_ambiguity",
+      parameters: [
+        { name: "state", type: "string", required: true, allowedValues: ["on", "off"] },
+      ],
+      intents: [
+        { label: "开启", phrases: ["打开", "开启", "启动", "开启制冷"], values: { state: "on" } },
+        { label: "关闭", phrases: ["关闭", "关掉", "停止", "停止制冷"], values: { state: "off" } },
+      ],
+      ui: {
+        type: "select",
+        value: "on",
+        options: [["off", "关闭"], ["on", "开启"]],
+      },
+      executor: { binding: "NativeCanRuntime", mappingRef: "demo.refrigerator.power" },
+    },
+    {
+      id: "refrigerator.set_temperature",
+      label: "目标温度",
+      property: "refrigerator.temperature",
+      operation: "set",
+      enabled: true,
+      risk: "medium",
+      confirmationPolicy: "on_ambiguity",
+      parameters: [
+        { name: "temperature_c", type: "number", unit: "celsius", required: true, minimum: 0, maximum: 15 },
+      ],
+      intents: [
+        {
+          label: "设置温度",
+          phrases: ["温度", "调到", "设置为", "制冷到", "冷藏到"],
+          extract: { parameter: "temperature_c", kind: "number" },
+        },
+      ],
+      ui: { type: "range", value: 4, minimum: 0, maximum: 15, step: 1, unit: "℃" },
+      executor: { binding: "NativeCanRuntime", mappingRef: "demo.refrigerator.temperature" },
+    },
+    {
+      id: "refrigerator.set_mode",
+      label: "工作模式",
+      property: "refrigerator.mode",
+      operation: "set",
+      enabled: true,
+      risk: "low",
+      confirmationPolicy: "on_ambiguity",
+      parameters: [
+        { name: "mode", type: "string", required: true, allowedValues: ["eco", "fast_cool"] },
+      ],
+      intents: [
+        { label: "节能", phrases: ["节能模式", "省电模式", "节能"], values: { mode: "eco" } },
+        { label: "速冷", phrases: ["速冷模式", "快速制冷", "速冷"], values: { mode: "fast_cool" } },
+      ],
+      ui: {
+        type: "select",
+        value: "eco",
+        options: [["eco", "节能"], ["fast_cool", "速冷"]],
+      },
+      executor: { binding: "NativeCanRuntime", mappingRef: "demo.refrigerator.mode" },
+    },
+  ],
+};
+
+const SCOPE_ALIASES = {
+  driver: ["主驾", "驾驶位", "司机位"],
+  passenger: ["副驾", "副驾驶", "乘客位"],
+  rear_row: ["后排", "第二排"],
+  vehicle: ["全车", "整车", "车内"],
+};
+
+const POLITE_WORDS = /^(?:请|麻烦|拜托|帮我|请帮我|给我)+/;
+const END_POLITE_WORDS = /(?:谢谢|可以吗|好吗|行吗)[。！!？?]*$/;
+
+export function cloneDescriptor(value = DEFAULT_DESCRIPTOR) {
+  return JSON.parse(JSON.stringify(value));
+}
+
+export function splitAliases(value) {
+  return [...new Set(String(value || "")
+    .split(/[,，、;；|]/)
+    .map((item) => item.trim())
+    .filter(Boolean))];
+}
+
+export function normalizeText(value) {
+  return String(value || "")
+    .normalize("NFKC")
+    .trim()
+    .replace(POLITE_WORDS, "")
+    .replace(END_POLITE_WORDS, "")
+    .toLowerCase()
+    .replace(/[\s，,。；;！!？?、]/g, "");
+}
+
+function longestContained(normalizedText, phrases) {
+  return [...new Set(phrases.map(normalizeText).filter(Boolean))]
+    .filter((phrase) => normalizedText.includes(phrase))
+    .sort((left, right) => right.length - left.length)[0] || "";
+}
+
+function chineseInteger(token) {
+  if (!token) return null;
+  if (/^\d+$/.test(token)) return Number(token);
+  const digits = { 零: 0, 〇: 0, 一: 1, 二: 2, 两: 2, 三: 3, 四: 4, 五: 5, 六: 6, 七: 7, 八: 8, 九: 9 };
+  if (token.includes("十")) {
+    const [left, right] = token.split("十", 2);
+    const tens = left ? digits[left.at(-1)] : 1;
+    const ones = right ? digits[right.at(-1)] : 0;
+    return Number.isFinite(tens) && Number.isFinite(ones) ? tens * 10 + ones : null;
+  }
+  if ([...token].every((char) => Object.hasOwn(digits, char))) {
+    return Number([...token].map((char) => digits[char]).join(""));
+  }
+  return null;
+}
+
+function extractNumber(text) {
+  const belowZero = String(text).match(/零下(\d+(?:\.\d+)?|[零〇一二两三四五六七八九十]+)/);
+  if (belowZero) {
+    const value = /^\d/.test(belowZero[1]) ? Number(belowZero[1]) : chineseInteger(belowZero[1]);
+    return value === null ? null : -value;
+  }
+  const numeric = String(text).match(/-?\d+(?:\.\d+)?/);
+  if (numeric) return Number(numeric[0]);
+  const chinese = String(text).match(/[零〇一二两三四五六七八九十]+/);
+  return chinese ? chineseInteger(chinese[0]) : null;
+}
+
+function resolveScope(normalizedText, device) {
+  const explicit = Object.entries(SCOPE_ALIASES)
+    .flatMap(([scope, aliases]) => aliases.map((alias) => ({ scope, alias: normalizeText(alias) })))
+    .find(({ alias }) => normalizedText.includes(alias));
+  if (explicit) {
+    if (!device.targetScopes.includes(explicit.scope)) {
+      return { status: "unsupported", scope: explicit.scope, explicit: true, reason: `设备不支持“${explicit.alias}”范围` };
+    }
+    return { status: "resolved", scope: explicit.scope, explicit: true, reason: `显式目标：${explicit.alias}` };
+  }
+  if (device.defaultScope && device.targetScopes.includes(device.defaultScope)) {
+    return { status: "resolved", scope: device.defaultScope, explicit: false, reason: `采用卡片默认目标：${device.defaultScope}` };
+  }
+  if (device.targetScopes.length === 1) {
+    return { status: "resolved", scope: device.targetScopes[0], explicit: false, reason: `能力仅支持唯一目标：${device.targetScopes[0]}` };
+  }
+  return { status: "ambiguous", scope: null, explicit: false, reason: "存在多个目标范围，需要用户确认" };
+}
+
+function validateParameters(capability, parameters) {
+  const issues = [];
+  for (const schema of capability.parameters || []) {
+    const value = parameters[schema.name];
+    if ((value === undefined || value === null || value === "") && schema.required) {
+      issues.push(`缺少必填参数 ${schema.name}`);
+      continue;
+    }
+    if (value === undefined || value === null || value === "") continue;
+    if (schema.type === "number" && typeof value !== "number") {
+      issues.push(`${schema.name} 必须是数字`);
+    }
+    if (schema.minimum !== undefined && value < schema.minimum) {
+      issues.push(`${schema.name}=${value} 低于下限 ${schema.minimum}`);
+    }
+    if (schema.maximum !== undefined && value > schema.maximum) {
+      issues.push(`${schema.name}=${value} 超过上限 ${schema.maximum}`);
+    }
+    if (schema.allowedValues && !schema.allowedValues.includes(value)) {
+      issues.push(`${schema.name}=${value} 不在允许值中`);
+    }
+  }
+  return issues;
+}
+
+function renderCleanText(descriptor, capability, parameters) {
+  const deviceName = descriptor.device.name;
+  if (capability.property.endsWith(".power")) {
+    return `${parameters.state === "on" ? "开启" : "关闭"}${deviceName}。`;
+  }
+  if (capability.property.endsWith(".temperature")) {
+    return `将${deviceName}设置为${parameters.temperature_c}度。`;
+  }
+  if (capability.property.endsWith(".mode")) {
+    const modeLabels = { eco: "节能模式", fast_cool: "速冷模式" };
+    return `将${deviceName}设置为${modeLabels[parameters.mode] || parameters.mode}。`;
+  }
+  return `执行${deviceName}的${capability.label}。`;
+}
+
+function baseResult(rawText, normalizedText, trace) {
+  return {
+    raw_text: rawText,
+    normalized_text: normalizedText,
+    semantic_status: "no_command",
+    execution_eligible: false,
+    clean_text: "",
+    command: null,
+    issues: [],
+    trace,
+  };
+}
+
+export function parseInstruction(rawText, descriptorInput = DEFAULT_DESCRIPTOR) {
+  const descriptor = cloneDescriptor(descriptorInput);
+  const normalizedText = normalizeText(rawText);
+  const trace = [{ stage: "文本整理", state: "success", detail: normalizedText || "输入为空" }];
+  const result = baseResult(String(rawText || ""), normalizedText, trace);
+
+  if (!normalizedText) {
+    result.semantic_status = "incomplete";
+    result.issues.push("请输入自然语言指令");
+    trace.push({ stage: "设备识别", state: "error", detail: "没有可分析的文字" });
+    return result;
+  }
+
+  const devicePhrases = [descriptor.device.name, ...(descriptor.device.aliases || [])];
+  const matchedDevice = longestContained(normalizedText, devicePhrases);
+  if (!matchedDevice) {
+    result.issues.push("指令没有命中设备名称或别名");
+    trace.push({ stage: "设备识别", state: "error", detail: `候选：${devicePhrases.join("、")}` });
+    return result;
+  }
+  trace.push({ stage: "设备识别", state: "success", detail: `“${matchedDevice}” → ${descriptor.device.id}` });
+
+  const candidates = [];
+  for (const capability of descriptor.capabilities.filter((item) => item.enabled !== false)) {
+    for (const intent of capability.intents || []) {
+      const phrase = longestContained(normalizedText, intent.phrases || []);
+      if (!phrase) continue;
+      const parameters = { ...(intent.values || {}) };
+      if (intent.extract?.kind === "number") {
+        const number = extractNumber(rawText);
+        if (number !== null) parameters[intent.extract.parameter] = number;
+      }
+      candidates.push({
+        capability,
+        intent,
+        phrase,
+        parameters,
+        score: matchedDevice.length * 100 + phrase.length,
+      });
+    }
+  }
+
+  if (!candidates.length) {
+    result.semantic_status = "incomplete";
+    result.issues.push("设备已识别，但没有命中已启用的功能说法");
+    trace.push({ stage: "功能识别", state: "error", detail: "请使用卡片中定义的功能说法" });
+    return result;
+  }
+
+  candidates.sort((left, right) => right.score - left.score);
+  const best = candidates[0];
+  const tied = candidates.filter((item) => item.score === best.score);
+  if (tied.length > 1) {
+    result.semantic_status = "ambiguous";
+    result.issues.push(`同时命中多个功能：${tied.map((item) => item.intent.label).join("、")}`);
+    trace.push({ stage: "功能识别", state: "error", detail: "最高匹配分相同，需要补充更明确的动作" });
+    return result;
+  }
+  trace.push({ stage: "功能识别", state: "success", detail: `“${best.phrase}” → ${best.capability.id}` });
+
+  const scope = resolveScope(normalizedText, descriptor.device);
+  trace.push({ stage: "目标解析", state: scope.status === "resolved" ? "success" : "error", detail: scope.reason });
+  if (scope.status !== "resolved") {
+    result.semantic_status = scope.status === "ambiguous" ? "ambiguous" : "recognized_unsupported";
+    result.issues.push(scope.reason);
+    return result;
+  }
+
+  const issues = validateParameters(best.capability, best.parameters);
+  trace.push({
+    stage: "参数校验",
+    state: issues.length ? "error" : "success",
+    detail: issues.length ? issues.join("；") : JSON.stringify(best.parameters),
+  });
+  if (issues.length) {
+    result.semantic_status = Object.keys(best.parameters).length ? "recognized_unsupported" : "incomplete";
+    result.issues.push(...issues);
+    return result;
+  }
+
+  result.semantic_status = "recognized_executable";
+  result.execution_eligible = true;
+  result.clean_text = renderCleanText(descriptor, best.capability, best.parameters);
+  result.command = {
+    domain: descriptor.device.domain,
+    capability: best.capability.id,
+    property: best.capability.property,
+    target_device: descriptor.device.targetDevice,
+    target_scope: scope.scope,
+    operation: best.capability.operation,
+    parameters: best.parameters,
+    risk_level: best.capability.risk,
+    confirmation_policy: best.capability.confirmationPolicy,
+    executor_binding: best.capability.executor.binding,
+    executor_payload: {
+      transport: descriptor.device.transport,
+      mapping_ref: best.capability.executor.mappingRef,
+      capability: best.capability.id,
+      target: scope.scope,
+      parameters: best.parameters,
+      demo_only: true,
+    },
+  };
+  trace.push({ stage: "执行绑定", state: "success", detail: `${best.capability.executor.binding} · 演示载荷，不发送` });
+  return result;
+}
+
+export function buildManualCommand(descriptorInput, values) {
+  const descriptor = cloneDescriptor(descriptorInput);
+  const commands = [];
+  for (const capability of descriptor.capabilities.filter((item) => item.enabled !== false)) {
+    const value = values[capability.id];
+    if (value === undefined) continue;
+    const schema = capability.parameters?.[0];
+    if (!schema) continue;
+    const parsedValue = schema.type === "number" ? Number(value) : value;
+    const parameters = { [schema.name]: parsedValue };
+    const issues = validateParameters(capability, parameters);
+    if (issues.length) continue;
+    commands.push({
+      capability: capability.id,
+      target_scope: descriptor.device.defaultScope,
+      parameters,
+      demo_only: true,
+    });
+  }
+  return commands;
+}
@@ -0,0 +1,34 @@
+import assert from "node:assert/strict";
+import { DEFAULT_DESCRIPTOR, cloneDescriptor, parseInstruction } from "./parser-engine.js";
+
+const successCases = [
+  ["开启车载冰箱", "refrigerator.set_power", { state: "on" }],
+  ["把冰箱温度调到4度", "refrigerator.set_temperature", { temperature_c: 4 }],
+  ["将冷藏箱设为速冷模式", "refrigerator.set_mode", { mode: "fast_cool" }],
+  ["关闭冰箱", "refrigerator.set_power", { state: "off" }],
+];
+
+for (const [utterance, capability, parameters] of successCases) {
+  const result = parseInstruction(utterance, DEFAULT_DESCRIPTOR);
+  assert.equal(result.semantic_status, "recognized_executable", utterance);
+  assert.equal(result.execution_eligible, true, utterance);
+  assert.equal(result.command.capability, capability, utterance);
+  assert.deepEqual(result.command.parameters, parameters, utterance);
+  assert.equal(result.command.executor_payload.demo_only, true, utterance);
+}
+
+const outOfRange = parseInstruction("把冰箱调到零下10度", DEFAULT_DESCRIPTOR);
+assert.equal(outOfRange.semantic_status, "recognized_unsupported");
+assert.equal(outOfRange.execution_eligible, false);
+assert.match(outOfRange.issues.join(" "), /低于下限/);
+
+const noDevice = parseInstruction("把空调调到4度", DEFAULT_DESCRIPTOR);
+assert.equal(noDevice.semantic_status, "no_command");
+assert.equal(noDevice.execution_eligible, false);
+
+const disabled = cloneDescriptor();
+disabled.capabilities.find((item) => item.id === "refrigerator.set_temperature").enabled = false;
+const disabledResult = parseInstruction("把冰箱温度调到4度", disabled);
+assert.equal(disabledResult.semantic_status, "incomplete");
+
+console.log(`device capability card demo: ${successCases.length + 3} parser checks passed`);
@@ -0,0 +1,746 @@
+:root {
+  color-scheme: light dark;
+  --page: #f2f5f2;
+  --surface: #ffffff;
+  --surface-soft: #f7f9f7;
+  --surface-strong: #e8eeea;
+  --ink: #18201c;
+  --muted: #68736c;
+  --line: #d8dfda;
+  --accent: #176b4a;
+  --accent-soft: #dceee5;
+  --accent-ink: #0f5137;
+  --warning: #9a640b;
+  --warning-soft: #fff3d7;
+  --danger: #a33a3a;
+  --danger-soft: #fde9e7;
+  --shadow: 0 18px 55px rgba(32, 50, 40, 0.08);
+  font-family: Inter, "PingFang SC", "Microsoft YaHei", system-ui, sans-serif;
+}
+
+* {
+  box-sizing: border-box;
+}
+
+body {
+  margin: 0;
+  min-width: 320px;
+  background:
+    radial-gradient(circle at 8% 3%, rgba(23, 107, 74, 0.1), transparent 25rem),
+    var(--page);
+  color: var(--ink);
+}
+
+button,
+input,
+select,
+textarea {
+  font: inherit;
+}
+
+button {
+  cursor: pointer;
+}
+
+button:disabled {
+  cursor: not-allowed;
+  opacity: 0.48;
+}
+
+.demo-shell {
+  width: min(1440px, calc(100% - 40px));
+  margin: 0 auto;
+  padding: 38px 0 60px;
+}
+
+.page-header {
+  display: flex;
+  align-items: flex-end;
+  justify-content: space-between;
+  gap: 24px;
+  margin-bottom: 22px;
+}
+
+.eyebrow {
+  margin: 0 0 8px;
+  color: var(--accent);
+  font-size: 13px;
+  font-weight: 700;
+  letter-spacing: 0.12em;
+}
+
+h1,
+h2,
+h3,
+p {
+  margin-top: 0;
+}
+
+h1 {
+  margin-bottom: 10px;
+  font-size: clamp(30px, 4vw, 48px);
+  line-height: 1.05;
+  letter-spacing: -0.045em;
+}
+
+.page-intro {
+  max-width: 720px;
+  margin-bottom: 0;
+  color: var(--muted);
+  line-height: 1.7;
+}
+
+.secondary-button,
+.primary-button {
+  min-height: 42px;
+  border-radius: 11px;
+  padding: 0 17px;
+  font-weight: 700;
+}
+
+.secondary-button {
+  border: 1px solid var(--line);
+  background: var(--surface);
+  color: var(--ink);
+}
+
+.primary-button {
+  border: 1px solid var(--accent);
+  background: var(--accent);
+  color: #fff;
+}
+
+.safety-note {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  margin-bottom: 20px;
+  border: 1px solid color-mix(in srgb, var(--accent) 20%, var(--line));
+  border-radius: 14px;
+  padding: 12px 16px;
+  background: color-mix(in srgb, var(--accent-soft) 60%, var(--surface));
+  color: var(--accent-ink);
+}
+
+.safety-note > span {
+  display: grid;
+  width: 28px;
+  height: 28px;
+  place-items: center;
+  border-radius: 50%;
+  background: var(--accent);
+  color: #fff;
+}
+
+.safety-note p {
+  margin: 0;
+  line-height: 1.55;
+}
+
+.workspace-grid {
+  display: grid;
+  grid-template-columns: minmax(320px, 0.75fr) minmax(560px, 1.55fr);
+  gap: 20px;
+  align-items: start;
+}
+
+.configuration-panel,
+.preview-section,
+.parser-section {
+  border: 1px solid var(--line);
+  border-radius: 20px;
+  background: var(--surface);
+  box-shadow: var(--shadow);
+}
+
+.configuration-panel {
+  padding: 22px;
+}
+
+.experiment-panel {
+  display: grid;
+  gap: 20px;
+}
+
+.preview-section,
+.parser-section {
+  padding: 22px;
+}
+
+.section-heading {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  margin-bottom: 18px;
+}
+
+.section-heading > span {
+  display: grid;
+  width: 30px;
+  height: 30px;
+  place-items: center;
+  border-radius: 9px;
+  background: var(--accent-soft);
+  color: var(--accent-ink);
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.section-heading h2 {
+  margin: 0;
+  font-size: 19px;
+  letter-spacing: -0.02em;
+}
+
+.compact-heading {
+  margin-top: 28px;
+}
+
+.field-grid {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 14px;
+}
+
+.field {
+  display: grid;
+  gap: 7px;
+}
+
+.field-wide {
+  grid-column: 1 / -1;
+}
+
+.field > span,
+.intent-alias-field > span {
+  color: var(--muted);
+  font-size: 12px;
+  font-weight: 700;
+}
+
+.field small {
+  color: var(--muted);
+  font-size: 11px;
+  line-height: 1.5;
+}
+
+input,
+select,
+textarea {
+  width: 100%;
+  border: 1px solid var(--line);
+  border-radius: 10px;
+  background: var(--surface-soft);
+  color: var(--ink);
+  outline: none;
+  transition: border-color 150ms ease, box-shadow 150ms ease;
+}
+
+input,
+select {
+  min-height: 40px;
+  padding: 0 11px;
+}
+
+textarea {
+  resize: vertical;
+  padding: 12px;
+  line-height: 1.6;
+}
+
+input:focus,
+select:focus,
+textarea:focus {
+  border-color: var(--accent);
+  box-shadow: 0 0 0 3px color-mix(in srgb, var(--accent) 14%, transparent);
+}
+
+.capability-editor {
+  display: grid;
+  gap: 12px;
+}
+
+.capability-definition {
+  border-top: 1px solid var(--line);
+  padding-top: 14px;
+}
+
+.capability-definition:first-child {
+  border-top: 0;
+  padding-top: 0;
+}
+
+.capability-title {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  margin-bottom: 10px;
+}
+
+.capability-title strong {
+  font-size: 14px;
+}
+
+.capability-title code {
+  color: var(--muted);
+  font-size: 11px;
+}
+
+.toggle-label {
+  display: inline-flex;
+  align-items: center;
+  gap: 7px;
+  color: var(--muted);
+  font-size: 12px;
+}
+
+.toggle-label input {
+  width: 16px;
+  min-height: 16px;
+  accent-color: var(--accent);
+}
+
+.intent-list {
+  display: grid;
+  gap: 9px;
+}
+
+.intent-alias-field {
+  display: grid;
+  grid-template-columns: 72px 1fr;
+  align-items: center;
+  gap: 8px;
+}
+
+.intent-alias-field input {
+  min-height: 36px;
+  font-size: 12px;
+}
+
+.schema-details {
+  margin-top: 20px;
+  border-top: 1px solid var(--line);
+  padding-top: 16px;
+}
+
+.schema-details summary {
+  color: var(--accent);
+  font-size: 13px;
+  font-weight: 700;
+  cursor: pointer;
+}
+
+pre {
+  overflow: auto;
+  margin: 12px 0 0;
+  border-radius: 12px;
+  padding: 15px;
+  background: #17231d;
+  color: #d8e6dd;
+  font: 12px/1.65 "SFMono-Regular", Consolas, monospace;
+  white-space: pre-wrap;
+  word-break: break-word;
+}
+
+.pipeline {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 9px;
+  min-height: 54px;
+  padding: 10px 16px;
+  color: var(--muted);
+  font-size: 12px;
+}
+
+.pipeline span {
+  border: 1px solid var(--line);
+  border-radius: 999px;
+  padding: 7px 11px;
+  background: var(--surface);
+  color: var(--ink);
+  font-weight: 700;
+}
+
+.pipeline i {
+  color: var(--accent);
+  font-style: normal;
+}
+
+.device-card {
+  position: relative;
+  overflow: hidden;
+  border: 1px solid color-mix(in srgb, var(--accent) 16%, var(--line));
+  border-radius: 18px;
+  padding: 20px;
+  background:
+    linear-gradient(135deg, color-mix(in srgb, var(--accent-soft) 70%, var(--surface)), var(--surface) 52%);
+}
+
+.device-card::after {
+  position: absolute;
+  right: -50px;
+  top: -70px;
+  width: 180px;
+  height: 180px;
+  border-radius: 50%;
+  background: color-mix(in srgb, var(--accent) 9%, transparent);
+  content: "";
+}
+
+.device-card-header {
+  position: relative;
+  z-index: 1;
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 14px;
+}
+
+.device-identity {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+}
+
+.device-icon {
+  display: grid;
+  width: 48px;
+  height: 48px;
+  place-items: center;
+  border-radius: 15px;
+  background: var(--accent);
+  color: #fff;
+  font-size: 22px;
+}
+
+.device-identity h3 {
+  margin: 0 0 4px;
+  font-size: 21px;
+}
+
+.device-identity p {
+  margin: 0;
+  color: var(--muted);
+  font-size: 12px;
+}
+
+.connection-state {
+  display: inline-flex;
+  align-items: center;
+  gap: 7px;
+  font-size: 12px;
+  font-weight: 700;
+  color: var(--accent-ink);
+}
+
+.connection-state i,
+.result-dot {
+  width: 8px;
+  height: 8px;
+  border-radius: 50%;
+  background: var(--accent);
+}
+
+.readout-grid {
+  position: relative;
+  z-index: 1;
+  display: grid;
+  grid-template-columns: repeat(3, 1fr);
+  gap: 10px;
+  margin: 20px 0;
+}
+
+.readout-grid div {
+  border: 1px solid color-mix(in srgb, var(--line) 78%, transparent);
+  border-radius: 12px;
+  padding: 12px;
+  background: color-mix(in srgb, var(--surface) 84%, transparent);
+}
+
+.readout-grid span,
+.card-control span {
+  display: block;
+  margin-bottom: 6px;
+  color: var(--muted);
+  font-size: 11px;
+  font-weight: 700;
+}
+
+.readout-grid strong {
+  font-size: 16px;
+}
+
+.card-controls {
+  position: relative;
+  z-index: 1;
+  display: grid;
+  grid-template-columns: repeat(3, 1fr);
+  gap: 12px;
+}
+
+.card-control input,
+.card-control select {
+  background: color-mix(in srgb, var(--surface) 88%, transparent);
+}
+
+.card-control output {
+  float: right;
+  color: var(--accent);
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.card-control input[type="range"] {
+  min-height: auto;
+  padding: 0;
+  accent-color: var(--accent);
+}
+
+.card-footnote {
+  position: relative;
+  z-index: 1;
+  margin: 15px 0 0;
+  color: var(--muted);
+  font-size: 11px;
+}
+
+.example-row {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px;
+  margin-bottom: 12px;
+}
+
+.example-row button,
+.text-button {
+  border: 0;
+  background: transparent;
+  color: var(--accent);
+  font-size: 12px;
+  font-weight: 700;
+}
+
+.example-row button {
+  border-radius: 999px;
+  padding: 7px 10px;
+  background: var(--accent-soft);
+}
+
+.command-composer {
+  display: grid;
+  grid-template-columns: 1fr auto;
+  gap: 10px;
+  align-items: stretch;
+}
+
+.command-composer textarea {
+  min-height: 64px;
+}
+
+.result-banner {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  margin: 14px 0;
+  border-radius: 13px;
+  padding: 13px 15px;
+  background: var(--surface-soft);
+}
+
+.result-banner[data-status="success"] {
+  background: var(--accent-soft);
+  color: var(--accent-ink);
+}
+
+.result-banner[data-status="warning"] {
+  background: var(--warning-soft);
+  color: var(--warning);
+}
+
+.result-banner[data-status="error"] {
+  background: var(--danger-soft);
+  color: var(--danger);
+}
+
+.result-banner[data-status="warning"] .result-dot {
+  background: var(--warning);
+}
+
+.result-banner[data-status="error"] .result-dot {
+  background: var(--danger);
+}
+
+.result-banner strong {
+  display: block;
+  margin-bottom: 3px;
+  font-size: 14px;
+}
+
+.result-banner p {
+  margin: 0;
+  font-size: 12px;
+  line-height: 1.55;
+}
+
+.result-grid {
+  display: grid;
+  grid-template-columns: minmax(240px, 0.85fr) minmax(280px, 1.15fr);
+  gap: 14px;
+}
+
+.trace-panel,
+.payload-panel {
+  min-width: 0;
+  border: 1px solid var(--line);
+  border-radius: 14px;
+  padding: 15px;
+}
+
+.trace-panel h3,
+.payload-heading h3 {
+  margin: 0;
+  font-size: 14px;
+}
+
+.trace-panel ol {
+  display: grid;
+  gap: 12px;
+  margin: 15px 0 0;
+  padding: 0;
+  list-style: none;
+  counter-reset: trace;
+}
+
+.trace-panel li {
+  position: relative;
+  min-height: 30px;
+  padding-left: 34px;
+  color: var(--muted);
+  font-size: 12px;
+  line-height: 1.5;
+  counter-increment: trace;
+}
+
+.trace-panel li::before {
+  position: absolute;
+  left: 0;
+  top: 0;
+  display: grid;
+  width: 24px;
+  height: 24px;
+  place-items: center;
+  border-radius: 8px;
+  background: var(--surface-strong);
+  color: var(--ink);
+  content: counter(trace);
+  font-size: 11px;
+  font-weight: 800;
+}
+
+.trace-panel li[data-state="success"]::before {
+  background: var(--accent-soft);
+  color: var(--accent-ink);
+}
+
+.trace-panel li[data-state="error"]::before {
+  background: var(--danger-soft);
+  color: var(--danger);
+}
+
+.trace-panel li strong {
+  display: block;
+  color: var(--ink);
+}
+
+.payload-heading {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+}
+
+.payload-panel pre {
+  min-height: 216px;
+  max-height: 360px;
+}
+
+.sr-only {
+  position: absolute;
+  width: 1px;
+  height: 1px;
+  overflow: hidden;
+  clip: rect(0, 0, 0, 0);
+  white-space: nowrap;
+}
+
+@media (prefers-color-scheme: dark) {
+  :root {
+    --page: #111713;
+    --surface: #19211c;
+    --surface-soft: #202923;
+    --surface-strong: #29342d;
+    --ink: #edf4ef;
+    --muted: #a4b1a8;
+    --line: #354139;
+    --accent: #58b98a;
+    --accent-soft: #243f32;
+    --accent-ink: #bdebd3;
+    --warning: #f2c873;
+    --warning-soft: #463b23;
+    --danger: #f0a09a;
+    --danger-soft: #4b2928;
+    --shadow: 0 18px 55px rgba(0, 0, 0, 0.22);
+  }
+
+  .primary-button,
+  .device-icon,
+  .safety-note > span {
+    color: #102019;
+  }
+}
+
+@media (max-width: 980px) {
+  .workspace-grid {
+    grid-template-columns: 1fr;
+  }
+}
+
+@media (max-width: 640px) {
+  .demo-shell {
+    width: min(100% - 24px, 1440px);
+    padding-top: 24px;
+  }
+
+  .page-header {
+    align-items: flex-start;
+    flex-direction: column;
+  }
+
+  .pipeline {
+    justify-content: flex-start;
+    overflow-x: auto;
+  }
+
+  .pipeline span {
+    white-space: nowrap;
+  }
+
+  .field-grid,
+  .card-controls,
+  .result-grid {
+    grid-template-columns: 1fr;
+  }
+
+  .field-wide {
+    grid-column: auto;
+  }
+
+  .readout-grid {
+    grid-template-columns: 1fr 1fr;
+  }
+
+  .command-composer {
+    grid-template-columns: 1fr;
+  }
+}
@@ -15,6 +15,12 @@ _FRACTION = re.compile(
    r"([零〇一二两三四五六七八九十百点\d.]+)"
 )
 _SLASH_FRACTION = re.compile(r"(?<!\d)(\d+)\s*/\s*(\d+)(?!\d)")
+_CHINESE_UNIT_VALUE = re.compile(
+    r"(?P<value>[零〇一二两三四五六七八九十百]+"
+    r"(?:点[零〇一二两三四五六七八九]+)?)"
+    r"(?P<spacing>\s*)"
+    r"(?P<unit>摄氏度|℃|度|档|级)"
+)
 _DIGITS = {
    "零": 0,
    "〇": 0,
@@ -76,6 +82,24 @@ def _number(token: str) -> int | float | None:
    return int(value) if value.is_integer() else value


+def normalize_chinese_unit_values(text: str) -> str:
+    """Convert explicitly unit-qualified Chinese numbers to Arabic digits.
+
+    Keeping the unit in the match is intentional: discourse words such as
+    ``一下`` or ``一点`` must not accidentally become control values.  Both
+    typed text and ASR text can call this helper before their domain grammar,
+    so value recognition does not depend on which input path was used.
+    """
+
+    def replace(match: re.Match[str]) -> str:
+        value = _number(match.group("value"))
+        if value is None:
+            return match.group(0)
+        return f"{value}{match.group('spacing')}{match.group('unit')}"
+
+    return _CHINESE_UNIT_VALUE.sub(replace, text)
+
+
 def parse_percentage_value(text: str) -> int | float | None:
    """Return an explicitly spoken percentage/ratio/extreme position.

@@ -9,7 +9,10 @@ from __future__ import annotations
 import re
 from typing import Any

-from cognitive_agent.application.chinese_values import parse_percentage_value
+from cognitive_agent.application.chinese_values import (
+    normalize_chinese_unit_values,
+    parse_percentage_value,
+)

 WINDOW_TARGETS = {
    "全部": "all",
@@ -177,7 +180,8 @@ _EXPLICIT_TARGET_PATTERNS = {
        re.IGNORECASE,
    ),
    "seat": re.compile(
-        r"主驾|驾驶位|驾驶席|司机位|左前(?:座|座椅|座位)?|"
+        r"主驾|主驾驶(?:座|座椅|席)?|驾驶位|驾驶席|驾驶座(?:椅)?|"
+        r"司机位|司机座(?:椅)?|左前(?:座|座椅|座位)?|"
        r"副驾(?:驶|位)?|乘客(?:位|座|座椅|座位)?|右前(?:座|座椅|座位)?|"
        r"前排|后排|左后|右后|所有座椅|全部座椅|全车座椅|"
        r"seat\s+(?:driver|front-left|passenger|front-right)",
@@ -561,7 +565,11 @@ def _parse_chinese_driver_seat_commands(compact: str) -> list[dict[str, Any]]:
                    clause = f"{clause}{continuation}"
                    continuation_index += 1
                index = continuation_index - 1
-            if re.search(r"主驾|驾驶位|司机位|左前座椅", clause):
+            if re.search(
+                r"主驾|主驾驶(?:座|座椅|席)?|驾驶位|驾驶席|"
+                r"驾驶座(?:椅)?|司机位|司机座(?:椅)?|左前座椅",
+                clause,
+            ):
                inherited_driver_scope = True
            elif re.search(r"副驾|副驾驶|乘客|右前", clause):
                inherited_driver_scope = False
@@ -592,7 +600,13 @@ def _parse_chinese_driver_seat_commands(compact: str) -> list[dict[str, Any]]:
        re.IGNORECASE,
    ):
        return []
-    explicit_driver = bool(re.search(r"主驾|驾驶位|司机位|左前座椅", compact))
+    explicit_driver = bool(
+        re.search(
+            r"主驾|主驾驶(?:座|座椅|席)?|驾驶位|驾驶席|"
+            r"驾驶座(?:椅)?|司机位|司机座(?:椅)?|左前座椅",
+            compact,
+        )
+    )
    if not explicit_driver and not _looks_like_seat_request(compact):
        return []
    default_reason = "用户未指定座椅目标，按驾驶台 CH1 安全默认选择主驾座椅。"
@@ -950,7 +964,7 @@ def _parse_chinese_window_commands(compact: str) -> list[dict[str, Any]]:


 def _parse_chinese_commands(text: str) -> list[dict[str, Any]]:
-    compact = re.sub(r"\s+", "", text)
+    compact = re.sub(r"\s+", "", normalize_chinese_unit_values(text))
    seat_text = _scoped_text(
        compact,
        r"座椅|座位|靠背|椅背|坐盆|坐垫|腿托|脚托|按摩|记忆|预设|"
@@ -981,7 +995,7 @@ def _parse_chinese_commands(text: str) -> list[dict[str, Any]]:

        temperature = re.search(
            r"(?:空调(?:温度)?|温度)(?:都)?"
-            r"(?:调整到|调整为|调到|调为|设为|设置为|调至|到|为)?"
+            r"(?:调整到|调整为|调到|调成|调为|设到|设为|设置为|调至|到|为)?"
            r"(1[89]|2\d|3[0-2])(?:摄氏)?(?:℃|度)?(?![\d.])",
            climate_text,
        )
@@ -1340,6 +1354,7 @@ def _drop_implicit_passenger_ch0_candidates(
        ambiguities.append(
            {
                "field": "seat.target_scope",
+                "reason": "implicit_driver_default_unresolved",
                "question": (
                    "未指定座椅目标时默认主驾 CH1；当前请求无法安全映射为已验证的主驾能力，"
                    "且不会转入副驾 CH0。"
@@ -133,7 +133,9 @@ CONVERSATION_ID_PATTERN = re.compile(r"^conv-[a-z0-9-]{6,80}$")
 MODE_EDITOR_REQUEST_ID_PATTERN = re.compile(r"^req-[a-z0-9-]{6,100}$")
 SEAT_SCOPE_PATTERN = re.compile(r"\bseats?\b|座椅|坐椅", re.IGNORECASE)
 DRIVER_SEAT_SCOPE_PATTERN = re.compile(
-    r"\bdriver\b|\bfront[-_ ]?left\b|\bleft[-_ ]?front\b|主驾|驾驶位|司机位|左前",
+    r"\bdriver\b|\bfront[-_ ]?left\b|\bleft[-_ ]?front\b|"
+    r"主驾|主驾驶(?:座|座椅|席)?|驾驶位|驾驶席|驾驶座(?:椅)?|"
+    r"司机位|司机座(?:椅)?|左前",
    re.IGNORECASE,
 )
 PASSENGER_SEAT_SCOPE_PATTERN = re.compile(
@@ -2257,13 +2259,9 @@ class SeatDemoService:
                local_preview=local_preview,
                model_parse_source=model_parse_source,
            )
-            if self._is_dynamic_mode_stop_request(utterance):
+            dynamic_mode_stop = self._is_dynamic_mode_stop_request(utterance)
+            if dynamic_mode_stop:
                parsed = self._apply_dynamic_mode_stop_request(parsed, utterance)
-            else:
-                parsed = self._apply_mode_profiles(
-                    parsed, matched_modes, context, vehicle_profile
-                )
-            parsed = self._mark_control_parse_source(parsed, "local_rule")
            relative_resolution = (
                self.native_can_relative_resolver.resolve(utterance, context)
                if self.native_can_relative_resolver.matches(utterance)
@@ -2282,6 +2280,19 @@ class SeatDemoService:
                    relative_resolution.claimed_controls if relative_resolution else None
                ),
            )
+            if not dynamic_mode_stop:
+                # Apply repository-backed modes after the final native-CAN
+                # merge.  Otherwise that merge can discard a valid passenger
+                # mode operation while enforcing its generic unscoped-seat
+                # CH1 default.
+                parsed = self._apply_mode_profiles(
+                    parsed,
+                    matched_modes,
+                    context,
+                    vehicle_profile,
+                    utterance=utterance,
+                )
+            parsed = self._mark_control_parse_source(parsed, "local_rule")
        if direct_mode_profile is not None:
            parsed = self._mark_control_parse_source(
                parsed,
@@ -3327,6 +3338,8 @@ class SeatDemoService:
        matched_modes: list[ModeProfile],
        context: dict[str, Any],
        vehicle_profile: VehicleProfile,
+        *,
+        utterance: str,
    ) -> dict[str, Any]:
        if not matched_modes:
            return parsed
@@ -3354,6 +3367,12 @@ class SeatDemoService:
            ]
            return rejected

+        parsed = self._apply_implicit_mode_target_policy(
+            parsed,
+            matched_modes,
+            utterance,
+        )
+
        safety_ambiguities = [
            item
            for item in parsed.get("ambiguities", [])
@@ -3430,6 +3449,38 @@ class SeatDemoService:
        )
        return merged

+    @staticmethod
+    def _apply_implicit_mode_target_policy(
+        parsed: dict[str, Any],
+        matched_modes: list[ModeProfile],
+        utterance: str,
+    ) -> dict[str, Any]:
+        """Let one matched mode own only an omitted seat target.
+
+        The native CH1 parser can add an implicit-driver ambiguity before or
+        after mode injection.  A unique repository-backed mode already carries
+        authoritative target scopes on every operation, so that one tagged
+        default is redundant.  Explicit scopes and all other safety questions
+        remain untouched.
+        """
+
+        if len(matched_modes) != 1 or (
+            DRIVER_SEAT_SCOPE_PATTERN.search(utterance)
+            or PASSENGER_SEAT_SCOPE_PATTERN.search(utterance)
+        ):
+            return parsed
+        ambiguities = list(parsed.get("ambiguities", []))
+        filtered = [
+            item
+            for item in ambiguities
+            if item.get("reason") != "implicit_driver_default_unresolved"
+        ]
+        if len(filtered) == len(ambiguities):
+            return parsed
+        resolved = deepcopy(parsed)
+        resolved["ambiguities"] = filtered
+        return resolved
+
    @staticmethod
    def _is_dynamic_mode_stop_request(utterance: str) -> bool:
        compact = re.sub(r"\s+", "", utterance)
@@ -165,6 +165,7 @@ class CapabilityResolver:
        *,
        target_scope: str | None,
        target_explicit: bool,
+        action: str = "apply",
    ) -> ModeResolution:
        matches = self.mode_profiles.match_text(text, vehicle_id="denza_n7")
        if not matches:
@@ -172,7 +173,24 @@ class CapabilityResolver:
        if len(matches) > 1:
            return ModeResolution("ambiguous", None, ("multiple_mode_profiles_matched",))
        profile = matches[0]
-        operation_scopes = {operation.target_scope for operation in profile.operations}
+        selected_operations = profile.operations
+        if action == "stop":
+            dynamic_operations = tuple(
+                operation
+                for operation in profile.operations
+                if operation.capability == "seat.start_dynamic_mode"
+            )
+            if not dynamic_operations:
+                return ModeResolution(
+                    "unsupported",
+                    profile,
+                    (f"mode_profile.{profile.id}不支持关闭指令",),
+                )
+            selected_operations = dynamic_operations
+
+        operation_scopes = {
+            operation.target_scope for operation in selected_operations
+        }
        if target_explicit and target_scope not in operation_scopes:
            scopes = ",".join(sorted(operation_scopes))
            return ModeResolution(
@@ -183,22 +201,28 @@ class CapabilityResolver:

        payloads: list[dict[str, Any]] = []
        issues: list[str] = []
-        for operation in profile.operations:
+        for operation in selected_operations:
+            capability = (
+                "seat.stop_dynamic_mode"
+                if action == "stop"
+                else operation.capability
+            )
+            parameters = {} if action == "stop" else operation.parameters
            resolution = self.resolve(
                domain=operation.domain,
                semantic_property="",
                target_scope=operation.target_scope,
                target_device=None,
-                operation=operation.operation,
-                parameters=operation.parameters,
+                operation="set" if action == "stop" else operation.operation,
+                parameters=parameters,
                source_text=text,
-                preferred_capability=operation.capability,
+                preferred_capability=capability,
            )
            if resolution.status != "executable" or resolution.executor_payload is None:
                issues.extend(resolution.issues or ("mode_operation_not_executable",))
                continue
            payloads.append(resolution.executor_payload)
-        if issues or len(payloads) != len(profile.operations):
+        if issues or len(payloads) != len(selected_operations):
            return ModeResolution("unsupported", profile, tuple(dict.fromkeys(issues)))
        return ModeResolution("executable", profile, executor_payloads=tuple(payloads))

@@ -9,7 +9,10 @@ from dataclasses import dataclass, replace
 from pathlib import Path
 from typing import Any

-from cognitive_agent.application.chinese_values import parse_percentage_value
+from cognitive_agent.application.chinese_values import (
+    normalize_chinese_unit_values,
+    parse_percentage_value,
+)
 from cognitive_agent.application.native_can_intent import merge_native_can_intent
 from cognitive_agent.infrastructure.speech.capability import CapabilityResolver
 from cognitive_agent.infrastructure.speech.contracts import (
@@ -858,11 +861,13 @@ class UnderstandingTranscriptProcessor:

        mode_profile = self.resolver.match_mode(text)
        if mode_profile is not None:
+            mode_operation = "stop" if action == "off" else "apply"
            mode_started = time.perf_counter()
            mode_resolution = self.resolver.resolve_mode(
                text,
                target_scope=scope,
                target_explicit=explicit_scope,
+                action=mode_operation,
            )
            resolution_ms += (time.perf_counter() - mode_started) * 1000
            mode_issues = tuple(mode_resolution.issues)
@@ -873,7 +878,7 @@ class UnderstandingTranscriptProcessor:
                target_scope=command_scope,
                target_device="seat",
                property_name="seat.mode",
-                operation="apply",
+                operation=mode_operation,
                parameters={"mode_profile_id": mode_profile.id},
                value=mode_profile.name,
                unit=None,
@@ -954,6 +959,14 @@ class UnderstandingTranscriptProcessor:
                parameters["state"] = action if action in {"on", "off"} else "on"
                if value is not None:
                    parameters["level"] = int(value)
+                elif action == "on":
+                    # The passenger-seat text parser executes a bare ON at
+                    # its lowest supported level; verified CH1 driver control
+                    # uses level 2 for the same language.  Preserve the
+                    # target-specific executable default in speech semantics,
+                    # but keep clean_text faithful to what the user said.
+                    parameters["level"] = 2 if scope == "driver" else 1
+                    qualifiers["implicit_parameters"] = ["level"]
            else:
                local = ("seat_thermal_action_incomplete",)
            commands.append(
@@ -1317,7 +1330,7 @@ class UnderstandingTranscriptProcessor:

    @staticmethod
    def _numeric_value(text: str) -> int | float | None:
-        text = _NON_NUMERIC_ONE.sub("", text)
+        text = normalize_chinese_unit_values(_NON_NUMERIC_ONE.sub("", text))
        match = _NUMBER.search(text)
        if match:
            value = float(match.group(1))
@@ -1415,6 +1428,8 @@ class UnderstandingTranscriptProcessor:
                return f"将{self._scoped_target(command, feature)}设置为{level}"
            return f"{'打开' if command.operation == 'on' else '关闭'}{self._scoped_target(command, feature)}"
        if prop == "seat.mode":
+            if command.operation == "stop":
+                return f"关闭{command.value}"
            return f"将{self._scoped_target(command, '座椅')}调整为{command.value}"
        if prop == "seat.backrest_recline":
            direction = (
@@ -212,6 +212,9 @@ ACTION_ALIASES: dict[str, tuple[str, ...]] = {
        "调成",
        "调为",
        "变成",
+        "进入",
+        "切换到",
+        "切换为",
        "设为",
        "设成",
        "设到",
@@ -532,6 +532,8 @@ class SeatDemoHandler(BaseHTTPRequestHandler):
            speech_utterance_id = str(
                payload.get("speech_utterance_id", "")
            ).strip()
+            speech_understanding: dict[str, object] | None = None
+            speech_peripheral_route = False
            if speech_utterance_id:
                if path != "/api/requests":
                    raise ValueError(
@@ -540,13 +542,28 @@ class SeatDemoHandler(BaseHTTPRequestHandler):
                speech_understanding = self.speech.cached_result(
                    speech_utterance_id
                )
-                if str(speech_understanding.get("clean_text", "")).strip() != utterance:
+                clean_text = str(
+                    speech_understanding.get("clean_text", "")
+                ).strip()
+                raw_text = str(
+                    speech_understanding.get("raw_text", "")
+                ).strip()
+                peripheral_routing = self._speech_peripheral_routing(raw_text)
+                peripheral_transcript = bool(
+                    peripheral_routing.get("execution_eligible") and raw_text
+                )
+                expected_text = raw_text if peripheral_transcript else clean_text
+                if utterance != expected_text:
                    raise SpeechRecognitionError(
                        "speech_text_changed",
                        "语音文本已被修改，请确认修改后的内容再执行。",
                        status=409,
                    )
-                if speech_understanding.get("execution_eligible") is not True:
+                speech_peripheral_route = peripheral_transcript
+                if (
+                    speech_understanding.get("execution_eligible") is not True
+                    and not speech_peripheral_route
+                ):
                    raise SpeechRecognitionError(
                        "speech_not_executable",
                        "语音意图尚未通过能力或参数校验，已拒绝执行。",
@@ -554,6 +571,7 @@ class SeatDemoHandler(BaseHTTPRequestHandler):
                    )
                provider = "local"
                common_arguments["provider"] = provider
+                if not speech_peripheral_route:
                    common_arguments["_speech_understanding"] = speech_understanding
            if (
                path != "/api/can/commands"
@@ -620,6 +638,14 @@ class SeatDemoHandler(BaseHTTPRequestHandler):
                        )
                    else:
                        result = self.service.process(**common_arguments)
+            if speech_peripheral_route and speech_understanding is not None:
+                result["speech_understanding"] = {
+                    "utterance_id": speech_utterance_id,
+                    "raw_text": speech_understanding.get("raw_text"),
+                    "clean_text": speech_understanding.get("clean_text"),
+                    "semantic_status": speech_understanding.get("semantic_status"),
+                    "routing": "validated_peripheral_text",
+                }
            utterance = str(result.get("utterance") or utterance)
            self.vehicle_display_sync.update_task(
                status=str(result.get("status", "succeeded")),
@@ -815,6 +841,54 @@ class SeatDemoHandler(BaseHTTPRequestHandler):
        )
        return result

+    def _speech_peripheral_routing(self, utterance: str) -> dict[str, object]:
+        """Validate whether an ASR transcript may enter peripheral routing.
+
+        Vehicle speech keeps using its cached structured semantic payload.  A
+        learned peripheral cannot be represented by the static vehicle
+        capability catalog, so it may instead reuse the exact server-owned ASR
+        transcript only after the live NFC binding and learned action resolve.
+        """
+
+        text = str(utterance or "").strip()
+        if not text:
+            return {"execution_eligible": False, "action_count": 0}
+        try:
+            catalog = self.extension.list_peripheral_capabilities()
+            nfc_snapshot = self.nfc.snapshot()
+            partition = partition_active_peripheral_intent(
+                text,
+                catalog,
+                nfc_snapshot,
+            )
+            entries = list(partition.get("peripheral_actions") or [])
+            unresolved = list(
+                partition.get("unresolved_peripheral_clauses") or []
+            )
+            denied = list(partition.get("denied_peripheral_actions") or [])
+            plan_catalog = active_peripheral_plan_catalog(
+                text,
+                catalog,
+                nfc_snapshot,
+            )
+            direct = None
+            if not entries and not plan_catalog and not unresolved and not denied:
+                direct = resolve_active_peripheral_command(
+                    text,
+                    catalog,
+                    nfc_snapshot,
+                )
+        except (ExtensionIntegrationError, GatewayError, ValueError):
+            return {"execution_eligible": False, "action_count": 0}
+        action_count = len(entries) or (1 if direct is not None else 0)
+        if plan_catalog:
+            action_count = max(action_count, 1)
+        return {
+            "execution_eligible": bool(action_count and not unresolved and not denied),
+            "action_count": action_count,
+            "has_ordered_plan": bool(plan_catalog),
+        }
+
    @staticmethod
    def _residual_has_vehicle_language(utterance: str) -> bool:
        """Recognize only obvious vehicle-domain text in a claimed mixed request."""
@@ -2206,6 +2280,14 @@ class SeatDemoHandler(BaseHTTPRequestHandler):
                self.rfile.read(content_length),
                utterance_id=self.headers.get("X-Speech-Utterance-Id"),
            )
+            raw_text = str(result.get("raw_text") or "").strip()
+            peripheral_routing = self._speech_peripheral_routing(raw_text)
+            if peripheral_routing.get("execution_eligible"):
+                result = {
+                    **result,
+                    "routing_text": raw_text,
+                    "peripheral_routing": peripheral_routing,
+                }
            self._json(result)
        except SpeechRecognitionError as exc:
            self._json(
@@ -624,7 +624,7 @@ operations:
        ) as response:
            gesture_script = response.read().decode("utf-8")
        with self.opener.open(
-            f"{self.base_url}/console-v2.js?v=delivery-state-54", timeout=3
+            f"{self.base_url}/console-v2.js?v=delivery-state-55", timeout=3
        ) as response:
            console_script = response.read().decode("utf-8")
        with self.opener.open(
@@ -632,7 +632,7 @@ operations:
        ) as response:
            console_styles = response.read().decode("utf-8")
        with self.opener.open(
-            f"{self.base_url}/speech-input.js?v=sensevoice-6", timeout=3
+            f"{self.base_url}/speech-input.js?v=sensevoice-7", timeout=3
        ) as response:
            speech_script = response.read().decode("utf-8")
        with self.opener.open(
@@ -685,9 +685,9 @@ operations:
        self.assertIn('href="/styles.css?v=dual-seat-21"', html)
        self.assertIn('src="/gesture-control.js?v=android-pad-9"', html)
        self.assertNotIn('class="vehicle-system-hud"', html)
-        self.assertIn('src="/console-v2.js?v=delivery-state-54"', html)
+        self.assertIn('src="/console-v2.js?v=delivery-state-55"', html)
        self.assertIn('href="/console-v2.css?v=delivery-state-50"', html)
-        self.assertIn('src="/speech-input.js?v=sensevoice-6"', html)
+        self.assertIn('src="/speech-input.js?v=sensevoice-7"', html)
        self.assertIn('href="/speech-input.css?v=sensevoice-5"', html)
        self.assertIn('id="speechDictationButton"', html)
        self.assertIn('class="speech-waveform"', html)
@@ -959,9 +959,9 @@ operations:
        self.assertIn("CAN 指令已提交，等待车辆状态确认", console_script)
        self.assertIn('canvas.addEventListener("webglcontextlost"', app_script)
        self.assertIn('"cognitive-agent:module-change"', app_script)
-        self.assertIn('const CACHE_VERSION = "cognitive-agent-pad-v227"', service_worker)
+        self.assertIn('const CACHE_VERSION = "cognitive-agent-pad-v228"', service_worker)
        self.assertIn('"/app.js?v=dual-seat-114"', service_worker)
-        self.assertIn('"/speech-input.js?v=sensevoice-6"', service_worker)
+        self.assertIn('"/speech-input.js?v=sensevoice-7"', service_worker)
        self.assertIn('"/speech-input.css?v=sensevoice-5"', service_worker)
        self.assertIn('"/speech-audio-worklet.js?v=sensevoice-4"', service_worker)
        self.assertIn('fetch("/api/speech/transcribe"', speech_script)
@@ -979,6 +979,9 @@ operations:
        self.assertIn("speechExecutionEligible", app_script)
        self.assertIn("speech_utterance_id", app_script)
        self.assertIn("speechUtteranceId", speech_script)
+        self.assertIn("result.peripheral_routing?.execution_eligible", speech_script)
+        self.assertIn("result.routing_text", speech_script)
+        self.assertIn('speech_semantic: "语音语义"', console_script)
        self.assertIn("语音意图已识别，但当前车辆能力不支持执行", app_script)
        self.assertIn("语音控制参数不完整，请补充后再执行", app_script)
        self.assertIn(".speech-dictation.is-listening", speech_styles)
@@ -1741,6 +1744,143 @@ operations:
        self.assertEqual(operation["domain"], "self_extension")
        self.assertTrue(operation["telemetry"]["gateway_acknowledged"])

+    def test_speech_routes_exact_asr_text_to_learned_peripheral_and_ui_config(
+        self,
+    ) -> None:
+        original_extension = SeatDemoHandler.extension
+        original_nfc = SeatDemoHandler.nfc
+        original_speech = SeatDemoHandler.speech
+        extension = SelfExtensionService(
+            SimulationGateway(),
+            catalog_db_path=Path(
+                self.extension_temporary_directory.name,
+                "speech-nfc-route.sqlite3",
+            ),
+        )
+        extension.sync_peripheral_catalog(
+            [
+                {
+                    "id": "cap-third-party-light-on",
+                    "status": "learned",
+                    "enabled": True,
+                    "device_id": "third-party-light",
+                    "device_name": "第三方氛围灯",
+                    "object_names": ["第三方氛围灯"],
+                    "card_uid": "1A53DD95",
+                    "action": "开启",
+                    "action_names": ["打开", "开启"],
+                    "slot": 9,
+                }
+            ]
+        )
+        SeatDemoHandler.extension = extension
+        SeatDemoHandler.nfc = SimpleNamespace(
+            snapshot=lambda: {
+                "revision": 1,
+                "reader": {"online": True},
+                "card": {"action": "present", "uid": "1A53DD95"},
+            }
+        )
+        understanding = UnderstandingTranscriptProcessor(REPOSITORY_ROOT).process(
+            "打开第三方氛围灯"
+        ).to_dict()
+        understanding["utterance_id"] = "e2e-speech-peripheral"
+
+        class SpeechStub:
+            @staticmethod
+            def transcribe_wav(
+                content: bytes,
+                *,
+                utterance_id: str | None = None,
+            ) -> dict[str, object]:
+                if content != b"RIFF-speech-peripheral":
+                    raise AssertionError("unexpected audio payload")
+                return {**understanding, "utterance_id": utterance_id}
+
+            @staticmethod
+            def cached_result(utterance_id: str) -> dict[str, object]:
+                if utterance_id != "e2e-speech-peripheral":
+                    raise AssertionError("unexpected speech utterance id")
+                return dict(understanding)
+
+        SeatDemoHandler.speech = SpeechStub()
+        try:
+            transcription_request = urllib.request.Request(
+                f"{self.base_url}/api/speech/transcribe",
+                method="POST",
+                headers={
+                    "Content-Type": "audio/wav",
+                    "X-Speech-Utterance-Id": "e2e-speech-peripheral",
+                },
+                data=b"RIFF-speech-peripheral",
+            )
+            with self.opener.open(transcription_request, timeout=3) as response:
+                transcription = json.loads(response.read().decode("utf-8"))
+
+            wrong_target_request = urllib.request.Request(
+                f"{self.base_url}/api/requests",
+                method="POST",
+                headers={"Content-Type": "application/json"},
+                data=json.dumps(
+                    {
+                        "utterance": understanding["clean_text"],
+                        "speech_utterance_id": "e2e-speech-peripheral",
+                        "provider": "local",
+                        "conversation_id": "conv-e2e-speech-wrong-target",
+                        "execution_mode": "dry_run",
+                    },
+                    ensure_ascii=False,
+                ).encode("utf-8"),
+            )
+            with self.assertRaises(urllib.error.HTTPError) as rejected:
+                self.opener.open(wrong_target_request, timeout=3)
+            self.assertEqual(rejected.exception.code, 409)
+            rejected_payload = json.loads(
+                rejected.exception.read().decode("utf-8")
+            )
+            self.assertEqual(rejected_payload["error"], "speech_text_changed")
+
+            control_request = urllib.request.Request(
+                f"{self.base_url}/api/requests",
+                method="POST",
+                headers={"Content-Type": "application/json"},
+                data=json.dumps(
+                    {
+                        "utterance": transcription["routing_text"],
+                        "speech_utterance_id": "e2e-speech-peripheral",
+                        "provider": "openai",
+                        "conversation_id": "conv-e2e-speech-peripheral",
+                        "execution_mode": "dry_run",
+                        "client_type": "web_console",
+                    },
+                    ensure_ascii=False,
+                ).encode("utf-8"),
+            )
+            with self.opener.open(control_request, timeout=3) as response:
+                result = json.loads(response.read().decode("utf-8"))
+        finally:
+            SeatDemoHandler.extension = original_extension
+            SeatDemoHandler.nfc = original_nfc
+            SeatDemoHandler.speech = original_speech
+
+        self.assertEqual(transcription["routing_text"], "打开第三方氛围灯")
+        self.assertTrue(
+            transcription["peripheral_routing"]["execution_eligible"]
+        )
+        self.assertEqual(result["status"], "succeeded")
+        self.assertEqual(result["orchestration_route"], "nfc_peripheral")
+        self.assertEqual(set(result["function_configs"]), {"self_extension"})
+        operation = result["function_configs"]["self_extension"]["spec"][
+            "operations"
+        ][0]
+        self.assertEqual(operation["capability"], "infrared.emit")
+        self.assertEqual(operation["parameters"]["slot"], 9)
+        self.assertEqual(operation["parse_source"], "local_rule")
+        self.assertEqual(
+            result["speech_understanding"]["routing"],
+            "validated_peripheral_text",
+        )
+
    def test_main_request_combines_explicit_ir_clause_with_existing_can_parser(self) -> None:
        original_extension = SeatDemoHandler.extension
        original_nfc = SeatDemoHandler.nfc
@@ -58,5 +58,8 @@
  {"name":"elliptical_temperature","raw":"空调二十四度","clean":"将空调设置为24度。","capabilities":["climate.set_temperature"],"cancelled":0,"uncertain":false,"execution_eligible":true},
  {"name":"elliptical_value_continuation","raw":"打开空调，然后二十四度","clean":"打开空调，然后将空调设置为24度。","capabilities":["climate.set_power","climate.set_temperature"],"cancelled":0,"uncertain":false,"execution_eligible":true},
  {"name":"punctuated_ambient_slots","raw":"氛围灯打开，调成蓝色，亮度百分之五十","clean":"打开氛围灯，然后将氛围灯颜色调整为蓝色并将亮度设置为50%。","capabilities":["ambient_light.set_power","ambient_light.set_illumination"],"cancelled":0,"uncertain":false,"execution_eligible":true},
-  {"name":"non_control_continuation_fails_closed","raw":"打开空调，然后开玩笑。","clean":"打开空调。","capabilities":["climate.set_power"],"cancelled":0,"uncertain":true,"execution_eligible":false}
+  {"name":"non_control_continuation_fails_closed","raw":"打开空调，然后开玩笑。","clean":"打开空调。","capabilities":["climate.set_power"],"cancelled":0,"uncertain":true,"execution_eligible":false},
+  {"name":"rocking_horse_mode_unscoped","raw":"将座椅调整为木马模式。","clean":"将座椅调整为木马模式。","capabilities":[null],"cancelled":0,"uncertain":false,"execution_eligible":true},
+  {"name":"rocking_horse_mode_enter","raw":"进入木马模式","clean":"将座椅调整为木马模式。","capabilities":[null],"cancelled":0,"uncertain":false,"execution_eligible":true},
+  {"name":"rocking_horse_mode_stop","raw":"关闭木马模式","clean":"关闭木马模式。","capabilities":[null],"cancelled":0,"uncertain":false,"execution_eligible":true}
 ]
@@ -47,7 +47,7 @@ def test_theme_asset_versions_are_consistent_with_the_pwa_shell() -> None:
    ):
        assert asset in html
        assert f'"{asset}"' in service_worker
-    assert 'const CACHE_VERSION = "cognitive-agent-pad-v227"' in service_worker
+    assert 'const CACHE_VERSION = "cognitive-agent-pad-v228"' in service_worker


 def test_vehicle_twin_icons_and_nested_overlays_use_semantic_tokens() -> None:
@@ -139,4 +139,5 @@ def test_main_parsed_control_panel_renders_validated_infrared_operations() -> No
    assert 'frameLabel: "GSCU1B TCP"' in console
    assert "已解析控制指令" in console
    assert 'li[data-can-channel="IR"]' in styles
-    assert "console-v2.js?v=delivery-state-54" in html
+    assert "console-v2.js?v=delivery-state-55" in html
+    assert 'speech_semantic: "语音语义"' in console
@@ -568,6 +568,73 @@ def test_mode_editor_prepares_one_to_many_mqtt_message_sequences() -> None:
    )


+def test_left_sidebar_extracts_speech_and_peripheral_control_commands() -> None:
+    node = shutil.which("node")
+    if node is None:
+        return
+    console = source("console-v2.js")
+    helper_start = console.index("  const normalizedCanTarget")
+    helper_end = console.index("\n\n  function renderParsedCanCommands", helper_start)
+    helpers = console[helper_start:helper_end]
+    sample = {
+        "execution_mode": "dry_run",
+        "function_configs": {
+            "native_climate": {
+                "spec": {
+                    "operations": [
+                        {
+                            "id": "speech-temperature",
+                            "capability": "climate.set_temperature",
+                            "target": {"device": "hvac", "zone": "driver"},
+                            "parameters": {"temperature_c": 24},
+                            "parse_source": "speech_semantic",
+                        }
+                    ]
+                }
+            },
+            "self_extension": {
+                "spec": {
+                    "operations": [
+                        {
+                            "id": "speech-peripheral",
+                            "capability": "infrared.emit",
+                            "target": {
+                                "device": "third-party-light",
+                                "name": "第三方氛围灯",
+                            },
+                            "parameters": {"action": "开启", "slot": 9},
+                            "parse_source": "local_rule",
+                        }
+                    ]
+                }
+            },
+        },
+    }
+    script = (
+        f"{helpers}\n"
+        f"const result = {json.dumps(sample, ensure_ascii=False)};\n"
+        "process.stdout.write(JSON.stringify({"
+        "commands:parsedCanCommands(result),"
+        'speechLabel:parsedCommandSourceLabel("speech_semantic")'
+        "}));"
+    )
+    completed = subprocess.run(
+        [node, "-e", script],
+        check=True,
+        capture_output=True,
+        text=True,
+    )
+    output = json.loads(completed.stdout)
+
+    assert output["speechLabel"] == "语音语义"
+    assert len(output["commands"]) == 2
+    assert {item["channel"] for item in output["commands"]} == {"CAN1", "IR"}
+    assert {item["command"] for item in output["commands"]} == {
+        "ac temp driver 24",
+        "infrared emit 9",
+    }
+
+
 def test_interactive_console_keeps_actions_inside_composer_and_collapsed_rails_aligned() -> None:
    console = source("console-v2.js")
    styles = source("console-v2.css")
@@ -148,7 +148,7 @@ class SpeechCapabilityAlignmentTests(unittest.TestCase):
                        "passenger",
                        "seat.ventilation",
                        "on",
-                        {"state": "on"},
+                        {"state": "on", "level": 1},
                    ),
                ),
            },
@@ -432,18 +432,38 @@ class SpeechCapabilityAlignmentTests(unittest.TestCase):

    def test_colloquial_seat_aliases_do_not_turn_pragmatic_one_into_level(self) -> None:
        cases = (
-            ("把驾驶座通风打开", "driver", "seat.set_driver_ventilation_native"),
-            ("把乘客座通风打开", "passenger", "seat.set_ventilation"),
-            ("给副驾驶开一下座椅通风", "passenger", "seat.set_ventilation"),
+            (
+                "把驾驶座通风打开",
+                "driver",
+                "seat.set_driver_ventilation_native",
+                {"state": "on", "level": 2},
+            ),
+            (
+                "把乘客座通风打开",
+                "passenger",
+                "seat.set_ventilation",
+                {"state": "on", "level": 1},
+            ),
+            (
+                "给副驾驶开一下座椅通风",
+                "passenger",
+                "seat.set_ventilation",
+                {"state": "on", "level": 1},
+            ),
        )
-        for text, target, capability in cases:
+        for text, target, capability, parameters in cases:
            with self.subTest(text=text):
                result = self.processor.process(text)
                self.assertTrue(result.execution_eligible)
                self.assertEqual(len(result.commands), 1)
                self.assertEqual(result.commands[0].target_scope, target)
                self.assertEqual(result.commands[0].capability, capability)
-                self.assertEqual(result.commands[0].parameters, {"state": "on"})
+                self.assertEqual(result.commands[0].parameters, parameters)
+                self.assertIn(
+                    "level",
+                    result.commands[0].qualifiers.get("implicit_parameters", ()),
+                )
+                self.assertNotIn("设置为", result.clean_text)

        explicit_level = self.processor.process("主驾座椅通风开到一档")
        self.assertEqual(explicit_level.commands[0].parameters["level"], 1)
@@ -455,7 +475,7 @@ class SpeechCapabilityAlignmentTests(unittest.TestCase):
        )
        expected = (
            ("seat.set_driver_ventilation_native", "driver", {"state": "on", "level": 2}),
-            ("seat.set_ventilation", "passenger", {"state": "on"}),
+            ("seat.set_ventilation", "passenger", {"state": "on", "level": 1}),
            ("climate.set_temperature", "driver", {"temperature_c": 24}),
        )
        for text in cases:
@@ -15,6 +15,44 @@ from cognitive_agent.infrastructure.speech.understanding import (
 REPOSITORY_ROOT = Path(__file__).resolve().parents[2]


+def _canonical_actions(result: dict[str, object]) -> list[dict[str, object]]:
+    intent_ir = result.get("intent_ir")
+    assert isinstance(intent_ir, dict)
+    spec = intent_ir.get("spec")
+    assert isinstance(spec, dict)
+    actions = spec.get("actions")
+    assert isinstance(actions, list)
+    return [
+        {
+            "domain": action.get("domain"),
+            "capability": action.get("capability"),
+            "target": action.get("target"),
+            "operation": action.get("operation"),
+            "parameters": action.get("parameters"),
+        }
+        for action in actions
+        if isinstance(action, dict)
+    ]
+
+
+def _sorted_canonical_actions(result: dict[str, object]) -> list[dict[str, object]]:
+    return sorted(
+        _canonical_actions(result),
+        key=lambda action: (
+            str(action["domain"]),
+            str(action["capability"]),
+            repr(action["target"]),
+            repr(action["parameters"]),
+        ),
+    )
+
+
+def _mode_service(directory: str) -> SeatDemoService:
+    service = SeatDemoService(REPOSITORY_ROOT)
+    service.artifact_root = Path(directory) / "artifacts"
+    return service
+
+
@pytest.mark.parametrize(
    ("utterance", "capability", "target", "parameters"),
    (
@@ -194,3 +232,157 @@ def test_every_registered_native_runtime_capability_keeps_native_bridge_binding(
            assert [
                action["capability"] for action in parsed["native_can_actions"]
            ] == [capability]
+
+
+def test_text_and_speech_share_implicit_mode_target_and_start_command() -> None:
+    processor = UnderstandingTranscriptProcessor(REPOSITORY_ROOT)
+    understanding = processor.process("将座椅调整为木马模式。").to_dict()
+    understanding["utterance_id"] = "mode-start-equivalence"
+
+    with tempfile.TemporaryDirectory() as directory, patch.dict(
+        os.environ,
+        {
+            "COGNITIVE_AGENT_HARDWARE_EXECUTION_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_CAN_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_MQTT_ENABLED": "false",
+            "COGNITIVE_AGENT_MODE_PROFILE_ROOT": str(Path(directory) / "modes"),
+        },
+        clear=False,
+    ):
+        service = _mode_service(directory)
+        text_result = service.process(
+            "将座椅调整为木马模式。",
+            provider="local",
+            conversation_id="conv-text-mode-start",
+        )
+        speech_result = service.process(
+            understanding["clean_text"],
+            provider="local",
+            conversation_id="conv-speech-mode-start",
+            _speech_understanding=understanding,
+        )
+
+    assert understanding["execution_eligible"] is True
+    assert text_result["status"] == speech_result["status"] == "succeeded"
+    assert _canonical_actions(text_result) == _canonical_actions(speech_result)
+    assert _canonical_actions(text_result) == [
+        {
+            "domain": "seat",
+            "capability": "seat.start_dynamic_mode",
+            "target": {
+                "device": "seat",
+                "zone": "front_right",
+                "instance": "passenger",
+            },
+            "operation": "activate",
+            "parameters": {"mode_id": 1},
+        }
+    ]
+
+
+def test_text_and_speech_share_dynamic_mode_stop_command() -> None:
+    processor = UnderstandingTranscriptProcessor(REPOSITORY_ROOT)
+    understanding = processor.process("关闭木马模式").to_dict()
+    understanding["utterance_id"] = "mode-stop-equivalence"
+
+    with tempfile.TemporaryDirectory() as directory, patch.dict(
+        os.environ,
+        {
+            "COGNITIVE_AGENT_HARDWARE_EXECUTION_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_CAN_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_MQTT_ENABLED": "false",
+            "COGNITIVE_AGENT_MODE_PROFILE_ROOT": str(Path(directory) / "modes"),
+        },
+        clear=False,
+    ):
+        service = _mode_service(directory)
+        text_result = service.process(
+            "关闭木马模式",
+            provider="local",
+            conversation_id="conv-text-mode-stop",
+        )
+        speech_result = service.process(
+            understanding["clean_text"],
+            provider="local",
+            conversation_id="conv-speech-mode-stop",
+            _speech_understanding=understanding,
+        )
+
+    assert understanding["clean_text"] == "关闭木马模式。"
+    assert understanding["execution_eligible"] is True
+    assert text_result["status"] == speech_result["status"] == "succeeded"
+    assert _canonical_actions(text_result) == _canonical_actions(speech_result)
+    assert _canonical_actions(text_result)[0]["capability"] == "seat.stop_dynamic_mode"
+    assert _canonical_actions(text_result)[0]["parameters"] == {}
+
+
+def test_explicit_incompatible_mode_target_fails_closed_for_both_inputs() -> None:
+    processor = UnderstandingTranscriptProcessor(REPOSITORY_ROOT)
+    understanding = processor.process("将主驾座椅调整为木马模式。").to_dict()
+
+    with tempfile.TemporaryDirectory() as directory, patch.dict(
+        os.environ,
+        {
+            "COGNITIVE_AGENT_HARDWARE_EXECUTION_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_CAN_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_MQTT_ENABLED": "false",
+            "COGNITIVE_AGENT_MODE_PROFILE_ROOT": str(Path(directory) / "modes"),
+        },
+        clear=False,
+    ):
+        service = _mode_service(directory)
+        text_result = service.process(
+            "将主驾座椅调整为木马模式。",
+            provider="local",
+            conversation_id="conv-text-mode-rejected",
+        )
+
+    assert understanding["semantic_status"] == "recognized_unsupported"
+    assert understanding["execution_eligible"] is False
+    assert text_result["status"] == "needs_confirmation"
+    assert _canonical_actions(text_result) == []
+
+
+def test_text_and_speech_share_colloquial_multidomain_commands() -> None:
+    utterance = "先打开主驾座椅通风，再打开副驾座椅通风，最后把空调调成二十四度"
+    processor = UnderstandingTranscriptProcessor(REPOSITORY_ROOT)
+    understanding = processor.process(utterance).to_dict()
+    understanding["utterance_id"] = "multidomain-equivalence"
+
+    with tempfile.TemporaryDirectory() as directory, patch.dict(
+        os.environ,
+        {
+            "COGNITIVE_AGENT_HARDWARE_EXECUTION_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_CAN_ENABLED": "false",
+            "COGNITIVE_AGENT_VEHICLE_MQTT_ENABLED": "false",
+        },
+        clear=False,
+    ):
+        service = _mode_service(directory)
+        text_result = service.process(
+            utterance,
+            provider="local",
+            conversation_id="conv-text-multidomain",
+        )
+        speech_result = service.process(
+            understanding["clean_text"],
+            provider="local",
+            conversation_id="conv-speech-multidomain",
+            _speech_understanding=understanding,
+        )
+
+    assert understanding["clean_text"] == (
+        "打开主驾座椅通风，然后打开副驾座椅通风，然后将空调设置为24度。"
+    )
+    assert understanding["execution_eligible"] is True
+    assert text_result["status"] == speech_result["status"] == "succeeded"
+    assert _sorted_canonical_actions(text_result) == _sorted_canonical_actions(
+        speech_result
+    )
+    assert {
+        action["capability"] for action in _canonical_actions(text_result)
+    } == {
+        "seat.set_driver_ventilation_native",
+        "seat.set_ventilation",
+        "climate.set_temperature",
+    }
@@ -186,9 +186,10 @@ def test_frontend_two_stage_contract_and_source_badges() -> None:
    assert '"cognitive-agent:validated-can-commands"' in app
    assert "本地规则命令已执行；云端补充失败" in app
    assert 'class="v2-can-command-source"' in console
-    assert 'item.parseSource === "direct_control" ? "设备控件"' in console
-    assert 'item.parseSource === "saved_mode_command_flow" ? "模式卡片"' in console
-    assert 'item.parseSource === "local_rule" ? "本地规则" : "云端模型"' in console
+    assert 'direct_control: "设备控件"' in console
+    assert 'speech_semantic: "语音语义"' in console
+    assert 'saved_mode_command_flow: "模式卡片"' in console
+    assert 'local_rule: "本地规则"' in console
    assert 'item.deliveryLabel' in console
    assert 'data-delivery-state="${escapeHtml(item.deliveryState)}"' in console
    assert '"cognitive-agent:validated-can-commands"' in console