myagent-ai 1.18.7 → 1.18.8

package/core/deps_checker.py

@@ -100,8 +100,10 @@ DEPENDENCIES: List[DepInfo] = [
     DepInfo("edge_tts", "edge-tts", "6.1.0", "tts", "all"),
 
     # ── 语音识别 (STT) ──
+    DepInfo("funasr", "funasr", "1.1.0", "stt", "all",
+            note="[v1.18.7] SenseVoice 中文语音识别（推荐，需 torch+torchaudio）"),
     DepInfo("faster_whisper", "faster-whisper", "1.0.0", "stt", "all",
-            note="本地语音识别引擎 (需要 C++ 编译)"),
+            note="Whisper 本地语音识别引擎 (需要 C++ 编译)"),
     DepInfo("speech_recognition", "SpeechRecognition", "3.10.0", "stt", "all",
             note="在线语音识别 (Google API，纯 Python 无需编译，Termux 兼容)"),
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.18.7",
+  "version": "1.18.8",
   "description": "本地桌面端执行型AI助手 - Open Interpreter 风格 | Local Desktop Execution-Oriented AI Assistant",
   "main": "main.py",
   "bin": {

package/skills/base.py

@@ -94,10 +94,14 @@ class Skill(ABC):
         pass
 
     def validate_params(self, params: Dict[str, Any]) -> tuple[bool, str]:
-        """校验参数是否合法"""
+        """校验参数是否合法。有默认值的必需参数在缺失时自动填充。"""
         for p in self.parameters:
-            if p.required and p.name not in params:
-                return False, f"缺少必需参数: {p.name}"
+            if p.name not in params:
+                # [v1.18.7] 有默认值的必需参数：自动填充而非报错
+                if p.default is not None:
+                    params[p.name] = p.default
+                elif p.required:
+                    return False, f"缺少必需参数: {p.name}"
             if p.name in params and p.enum and params[p.name] not in p.enum:
                 return False, f"参数 {p.name} 值无效，可选: {p.enum}"
         return True, ""

package/skills/xlsx_skill.py

@@ -8,6 +8,7 @@ from __future__ import annotations
 
 import json
 import os
+import time
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 
@@ -29,18 +30,19 @@ class XLSXCreateSkill(Skill):
     """
     name = "xlsx_create"
     description = (
-        "生成 Excel (XLSX) 电子表格。支持多工作表、表头、数据行、公式、"
+        "生成 Excel (XLSX) 电子表格文件。支持多工作表、表头、数据行、公式、"
         "冻结窗格、自动筛选、列宽设置。"
-        "sheets 为 JSON 对象，key 为工作表名。"
+        "参数 sheets 必须是 JSON 字符串，格式为 {\"工作表名\": {\"headers\":[...], \"rows\":[[...]]}}。"
+        "参数 output_path 指定输出文件路径，默认为工作目录下。"
     )
     category = "doc"
     dangerous = True
     parameters = [
         SkillParameter("sheets", "string",
-            "工作表数据 JSON 对象。key=工作表名, value={headers, rows, col_widths?, formulas?}。"
-            "示例: {\"Sheet1\":{\"headers\":[\"Name\",\"Score\"],\"rows\":[[\"Alice\",95]]}}",
+            "工作表数据，JSON 字符串。key=工作表名, value包含 headers(列名数组) 和 rows(数据行二维数组)。"
+            "示例: '{\"Sheet1\":{\"headers\":[\"姓名\",\"分数\"],\"rows\":[[\"Alice\",95]]}}'",
             required=True),
-        SkillParameter("output_path", "string", "输出 XLSX 文件路径", required=True),
+        SkillParameter("output_path", "string", "输出文件路径(如 /tmp/report.xlsx)", required=False, default=""),
         SkillParameter("title", "string", "文档标题", required=False, default=""),
     ]
 
@@ -62,6 +64,15 @@ class XLSXCreateSkill(Skill):
 
         try:
             out = Path(output_path).expanduser().resolve()
+            if not output_path.strip():
+                # [v1.18.7] 默认输出路径：工作目录/data/workspace下
+                from core.context_manager import get_active_context
+                try:
+                    ctx = get_active_context()
+                    work_dir = Path(ctx.work_dir) if ctx and ctx.work_dir else Path.cwd()
+                except Exception:
+                    work_dir = Path.cwd()
+                out = work_dir / f"report_{int(time.time())}.xlsx"
             out.parent.mkdir(parents=True, exist_ok=True)
 
             wb = openpyxl.Workbook()

package/web/api_server.py

@@ -1545,13 +1545,15 @@ window.toggleFullscreen = function() {{
             return web.json_response({"error": str(e)}, status=500)
 
     async def handle_voice_stt(self, request):
-        """POST /api/voice-stt - 轻量级本地语音转文字
+        """POST /api/voice-stt - 本地语音转文字
 
         接受音频文件（WAV/WEBM/OGG），使用本地 STT 引擎转录。
         支持的引擎（按优先级）：
-        1. faster-whisper（推荐，需安装：pip install faster-whisper）
-        2. vosk（备选，需安装：pip install vosk）
-        如果都未安装，返回错误提示。
+        1. [v1.18.7] SenseVoice（推荐，中文识别最佳，需：pip install funasr torch torchaudio）
+        2. faster-whisper（备选，需安装：pip install faster-whisper）
+        3. vosk（备选，需安装：pip install vosk）
+        4. LLM API Whisper 兼容端点
+        5. SpeechRecognition（Google，需外网）
         """
         try:
             reader = await request.multipart()
@@ -1584,6 +1586,59 @@ window.toggleFullscreen = function() {{
 
             import io
 
+            # ── [v1.18.7] 首选: SenseVoice（阿里达摩院，中文识别极佳） ──
+            try:
+                sv_model = getattr(self, '_sensevoice_model', None)
+                if sv_model is None:
+                    os.environ.setdefault("HF_HUB_DISABLE_TELEMETRY", "1")
+                    os.environ.setdefault("HF_HUB_DISABLE_PROGRESS_BARS", "1")
+                    from funasr import AutoModel
+                    model_dir = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), 'models', 'sensevoice')
+                    sv_model = AutoModel(model="iic/SenseVoiceSmall", model_dir=model_dir,
+                                         device="cpu", disable_pbar=True, disable_update=True)
+                    self._sensevoice_model = sv_model
+                    logger.info("SenseVoice 模型已加载 (iic/SenseVoiceSmall, CPU)")
+
+                # SenseVoice 接受 16kHz WAV
+                wav_path = f"/tmp/myagent_stt_{id(audio_data) % 100000}.wav"
+                wav_buf = io.BytesIO()
+                try:
+                    from pydub import AudioSegment
+                    audio_buf = io.BytesIO(audio_data)
+                    seg = AudioSegment.from_file(audio_buf, format=audio_format or "webm")
+                    seg = seg.set_channels(1).set_frame_rate(16000).set_sample_width(2)
+                    seg.export(wav_buf, format="wav")
+                except Exception:
+                    wav_buf = io.BytesIO(audio_data)
+                wav_buf.seek(0)
+                with open(wav_path, 'wb') as f:
+                    f.write(wav_buf.read())
+
+                # SenseVoice 推理
+                res = sv_model.generate(input=wav_path, cache={},
+                                        language="auto",  # 自动检测语言
+                                        use_itn=True,     # 逆文本标准化（数字/日期等）
+                                        batch_size_s=300)
+                if res and len(res) > 0 and len(res[0]) > 0:
+                    text = res[0][0]["text"] if isinstance(res[0][0], dict) else str(res[0][0])
+                    # SenseVoice 可能输出带 <|zh|><|en|><|EMO|> 等特殊 token，清理掉
+                    import re
+                    text = re.sub(r'<\|[^|]+\|>', '', text).strip()
+                    if text:
+                        try:
+                            os.remove(wav_path)
+                        except Exception:
+                            pass
+                        return web.json_response({"text": text, "engine": "sensevoice"})
+                try:
+                    os.remove(wav_path)
+                except Exception:
+                    pass
+            except ImportError:
+                logger.debug("SenseVoice (funasr) 未安装，跳过。安装: pip install funasr torch torchaudio")
+            except Exception as e:
+                logger.warning(f"SenseVoice 转录失败: {e}")
+
             # ── 尝试 faster-whisper ──
             try:
                 whisper_model = self._whisper_model
@@ -1815,10 +1870,11 @@ window.toggleFullscreen = function() {{
             # ── 没有可用的 STT 引擎 ──
             return web.json_response({
                 "error": "未检测到可用的 STT 引擎。请尝试以下方案：\n"
-                         "  1. 配置支持 Whisper 的 LLM API（自动使用，推荐）\n"
-                         "  2. pip install faster-whisper  (离线本地，需 C++ 编译环境)\n"
-                         "  3. pip install vosk             (离线本地，需下载模型)\n"
-                         "  4. pip install SpeechRecognition (需外网，国内不可用)",
+                         "  1. pip install funasr torch torchaudio  (SenseVoice，中文最佳，推荐)\n"
+                         "  2. 配置支持 Whisper 的 LLM API（自动使用，无需安装）\n"
+                         "  3. pip install faster-whisper  (离线本地，需 C++ 编译环境)\n"
+                         "  4. pip install vosk             (离线本地，需下载模型)\n"
+                         "  5. pip install SpeechRecognition (需外网，国内不可用)",
                 "available": False,
             }, status=503)