@flowingspring/dsh-voco 0.2.9 → 0.3.0

package/README.md

@@ -37,6 +37,7 @@ The plugin uses `XingChenAGI/XingChenASR-V3.2-Ultra` for cloud speech recognitio
 - Speaks a short contextual acknowledgement before starting delegated Agent work.
 - Keeps full task reports in the task UI while speaking a concise, purpose-written result.
 - Supports interruption, browser navigation, reconnects, and restored conversation history.
+- Pure voice sessions get a concise title from the first meaningful voice request, with the transcription as a fallback when the title model is unavailable.
 - Ships the server and browser surfaces as one public npm package.
 
 ## Configuration

package/README.zh.md

@@ -37,6 +37,7 @@ SILICONFLOW_API_KEY=你的密钥
 - 委派任务启动前立即播报一句贴合当前请求的简短确认语。
 - 完整任务报告保留在任务界面，语音只播报专门生成的简洁结果。
 - 支持语音打断、页面切换、断线重连以及历史对话恢复。
+- 纯语音会话会根据第一条有效语音请求生成简短标题；标题模型不可用时使用语音转写作为备用。
 - 服务端和浏览器界面统一通过一个公开 npm 包发行。
 
 ## 配置项

package/lib/client.js

@@ -18,11 +18,11 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var VoiceControl_module_css_default = {
-			"controls": "TFZsua_controls",
+			"interruptButton": "TFZsua_interruptButton",
+			"active": "TFZsua_active",
 			"voice-pulse": "TFZsua_voice-pulse",
 			"button": "TFZsua_button",
-			"active": "TFZsua_active",
-			"interruptButton": "TFZsua_interruptButton"
+			"controls": "TFZsua_controls"
 		};
 		//#endregion
 		//#region ../ui-voice/src/client/VoiceControl.tsx
@@ -141,22 +141,22 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var VoiceNodeViews_module_css_default = {
-			"taskInput": "Kt2smW_taskInput",
-			"miniWave": "Kt2smW_miniWave",
-			"taskSummary": "Kt2smW_taskSummary",
-			"meta": "Kt2smW_meta",
-			"bubble": "Kt2smW_bubble",
-			"taskStatus": "Kt2smW_taskStatus",
-			"taskActions": "Kt2smW_taskActions",
 			"taskLink": "Kt2smW_taskLink",
-			"taskDetails": "Kt2smW_taskDetails",
-			"taskDot": "Kt2smW_taskDot",
+			"taskCard": "Kt2smW_taskCard",
 			"utterance": "Kt2smW_utterance",
+			"taskDot": "Kt2smW_taskDot",
+			"bubble": "Kt2smW_bubble",
 			"voiceBadge": "Kt2smW_voiceBadge",
-			"taskCard": "Kt2smW_taskCard",
-			"taskCancel": "Kt2smW_taskCancel",
+			"taskActions": "Kt2smW_taskActions",
 			"taskChevron": "Kt2smW_taskChevron",
+			"taskSummary": "Kt2smW_taskSummary",
+			"miniWave": "Kt2smW_miniWave",
+			"meta": "Kt2smW_meta",
+			"taskDetails": "Kt2smW_taskDetails",
 			"taskUpdate": "Kt2smW_taskUpdate",
+			"taskCancel": "Kt2smW_taskCancel",
+			"taskInput": "Kt2smW_taskInput",
+			"taskStatus": "Kt2smW_taskStatus",
 			"taskTitle": "Kt2smW_taskTitle"
 		};
 		//#endregion
@@ -285,11 +285,11 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var VoiceOverlay_module_css_default = {
-			"wave": "GPPB2G_wave",
 			"root": "GPPB2G_root",
+			"status": "GPPB2G_status",
+			"wave": "GPPB2G_wave",
 			"voice-wave": "GPPB2G_voice-wave",
-			"stop": "GPPB2G_stop",
-			"status": "GPPB2G_status"
+			"stop": "GPPB2G_stop"
 		};
 		//#endregion
 		//#region ../ui-voice/src/client/VoiceOverlay.tsx

package/lib/plugins/voice-assistant.js

@@ -159,6 +159,13 @@ function optionalWorkspaceMemory(ctx) {
 		return;
 	}
 }
+function optionalSessionTitle(ctx) {
+	try {
+		return ctx.get("sessionTitle");
+	} catch {
+		return;
+	}
+}
 function workspaceMemoryReference(memory) {
 	const sections = [];
 	if (memory.summary.trim() !== "") sections.push(`稳定摘要：\n${memory.summary.trim()}`);
@@ -193,6 +200,55 @@ const REWRITE_EVENT_INSTRUCTIONS = {
 	warning: "这是警告信息。简洁说明风险、影响以及用户需要知道的事项。",
 	error: "这是失败信息。明确说明没有完成、主要原因以及可行的下一步。"
 };
+const VOICE_TITLE_SYSTEM_PROMPT = `你是语音会话标题生成器。根据用户第一条有实际内容的语音请求，生成一个简短、准确、便于在会话列表中识别的中文标题。
+
+只输出标题本身，不要解释，不要引号，不要 Markdown，不要句末标点，不要使用“语音会话”“新会话”等泛化名称，不超过 12 个汉字。不要加入文件大小、行数、路径等执行细节。`;
+function voiceTitleCandidate(value) {
+	const normalized = value.replace(/```[\s\S]*?```/gu, "").replace(/^\s*(?:标题|title)\s*[:：]\s*/iu, "").replace(/["'「」『』]/gu, "").split(/\r?\n/u, 1)[0].replace(/[。！？!?；;，,：:]+$/u, "").replace(/\s+/gu, " ").trim();
+	return Array.from(normalized).slice(0, 12).join("");
+}
+function isMeaningfulVoiceTitleInput(value) {
+	const normalized = value.replace(/\s+/gu, "").trim();
+	if (Array.from(normalized).length < 4) return false;
+	return !/^(?:你好|您好|嗨|哈喽|喂|在吗|有人吗|测试|谢谢|好的)[，。！？!?、]*$/u.test(normalized);
+}
+async function generateVoiceSessionTitle(ctx, session, input, titleService) {
+	let candidate = "";
+	try {
+		const llm = ctx.get("llm");
+		if (llm !== void 0) {
+			const selection = ctx.agentDefaultModel.currentSelection();
+			const message = createUserMessage({
+				content: [{
+					type: "text",
+					text: `用户第一条有效语音请求：${input}`
+				}],
+				source: {
+					kind: "plugin",
+					plugin: "voice-assistant"
+				}
+			});
+			let output = "";
+			for await (const chunk of llm.stream({
+				provider: selection.provider,
+				model: selection.model,
+				...selection.reasoningEffort === void 0 ? {} : { reasoningEffort: selection.reasoningEffort },
+				messages: [message],
+				system: VOICE_TITLE_SYSTEM_PROMPT
+			})) if (chunk.type === "text-delta") output += chunk.text;
+			candidate = voiceTitleCandidate(output);
+		}
+	} catch (error) {
+		ctx.logger.warn(error instanceof Error ? error : new Error(String(error)));
+	}
+	if (candidate === "") candidate = voiceTitleCandidate(input);
+	if (candidate === "" || titleService.get(session) !== void 0) return;
+	try {
+		titleService.rename(session, candidate);
+	} catch (error) {
+		ctx.logger.warn(error instanceof Error ? error : new Error(String(error)));
+	}
+}
 /** Install the driver. @param ctx - composed Agent and voice context. @param config - driver copy and queue bounds. */
 function apply(ctx, config = {}) {
 	const bindings = /* @__PURE__ */ new Map();
@@ -223,6 +279,7 @@ function apply(ctx, config = {}) {
 				interactionMode: void 0,
 				voiceAttached: false,
 				voiceTurnMarked: false,
+				voiceTitleRequested: false,
 				active: void 0,
 				continuousTaskAgent: void 0,
 				lastTerminalTaskId: void 0,
@@ -716,13 +773,22 @@ function apply(ctx, config = {}) {
 	};
 	const endUtterance = (binding, utteranceId, role, state, finalText, responseId) => {
 		const utterance = beginUtterance(binding, utteranceId, role, responseId);
-		requireSourceSession(binding).append("voice/utterance-end", {
+		const sourceSession = requireSourceSession(binding);
+		sourceSession.append("voice/utterance-end", {
 			utteranceId,
 			role,
 			text: finalText ?? utterance.text,
 			state,
 			...responseId === void 0 ? {} : { responseId }
 		});
+		const utteranceText = finalText ?? utterance.text;
+		if (role === "user" && state === "completed" && isMeaningfulVoiceTitleInput(utteranceText)) {
+			const titleService = optionalSessionTitle(ctx);
+			if (titleService !== void 0 && !binding.voiceTitleRequested && titleService.get(sourceSession) === void 0) {
+				binding.voiceTitleRequested = true;
+				generateVoiceSessionTitle(ctx, sourceSession, utteranceText, titleService);
+			}
+		}
 		binding.utterances.delete(utteranceId);
 	};
 	const interruptAssistantUtterances = (binding, responseId) => {

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@flowingspring/dsh-voco",
   "description": "Persistent voice conversations for DSH with cloud speech recognition, Edge TTS, and background Agent delegation",
-  "version": "0.2.9",
+  "version": "0.3.0",
   "keywords": [
     "deepseek-harness",
     "dsh",