myagent-ai 1.18.8 → 1.19.0

package/agents/main_agent.py

@@ -1589,13 +1589,17 @@ class MainAgent(BaseAgent):
                     _fresult = await _fskill.execute(_fpath, _fdesc, stream_callback=stream_callback)
                     result = {"success": True, "output": json.dumps(_fresult, ensure_ascii=False, indent=2), "data": _fresult}
                     # [v1.18.5] 追踪发送的文件，用于持久化到会话记忆
-                    if _fresult.get("success") and _fresult.get("file_id"):
-                        _sent_files.append({
-                            "id": _fresult["file_id"],
-                            "name": _fresult.get("name", ""),
-                            "type": _fresult.get("type", ""),
-                            "size": _fresult.get("size", 0),
-                        })
+                    # [v1.18.9] 修复：_sent_files 可能不在作用域内，安全处理
+                    try:
+                        if _fresult.get("success") and _fresult.get("file_id"):
+                            _sent_files.append({
+                                "id": _fresult["file_id"],
+                                "name": _fresult.get("name", ""),
+                                "type": _fresult.get("type", ""),
+                                "size": _fresult.get("size", 0),
+                            })
+                    except NameError:
+                        pass  # _sent_files 不在当前作用域，跳过文件追踪
                 except Exception as _fse:
                     result = {"success": False, "error": f"文件发送失败: {_fse}"}
                     logger.warning(f"[{task_id}] file_send 工具异常: {_fse}")

package/core/deps_checker.py

@@ -100,10 +100,15 @@ DEPENDENCIES: List[DepInfo] = [
     DepInfo("edge_tts", "edge-tts", "6.1.0", "tts", "all"),
 
     # ── 语音识别 (STT) ──
+    # [v1.18.8] torch/torchaudio 是 SenseVoice (funasr) 的必需依赖
+    DepInfo("torch", "torch", "2.0.0", "stt", "all",
+            note="PyTorch 深度学习框架 (SenseVoice 必需, CPU版约200MB)"),
+    DepInfo("torchaudio", "torchaudio", "2.0.0", "stt", "all",
+            note="PyTorch 音频处理库 (SenseVoice 必需)"),
     DepInfo("funasr", "funasr", "1.1.0", "stt", "all",
-            note="[v1.18.7] SenseVoice 中文语音识别（推荐，需 torch+torchaudio）"),
+            note="[v1.18.8] SenseVoice 中文语音识别（首选，阿里达摩院）"),
     DepInfo("faster_whisper", "faster-whisper", "1.0.0", "stt", "all",
-            note="Whisper 本地语音识别引擎 (需要 C++ 编译)"),
+            note="Whisper 本地语音识别引擎 (备选，需 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/core/logger.py

@@ -219,18 +219,30 @@ def setup_logger(
 
 
 def get_logger(name: str = "myagent") -> logging.Logger:
-    """获取已存在的 Logger，如果不存在则创建默认的"""
+    """获取已存在的 Logger，如果不存在则创建默认的。
+
+    [v1.18.8] 子 logger 不再创建独立文件，而是继承最近有 handlers 的父 logger。
+    例如 get_logger("myagent.api") 会复用 "myagent" 的文件 handler，
+    确保所有日志集中写入同一个 myagent.log 文件。
+    """
     logger = logging.getLogger(name)
     if not logger.handlers:
-        # 子 logger 继承父 logger 的 handlers（如文件 handler）
-        parent_logger = logging.getLogger()
-        if parent_logger.handlers:
-            for handler in parent_logger.handlers:
+        # 向上查找最近有 handlers 的父 logger（如 "myagent"）
+        parts = name.split(".")
+        parent_found = None
+        for i in range(len(parts) - 1, 0, -1):
+            parent_name = ".".join(parts[:i])
+            parent = logging.getLogger(parent_name)
+            if parent.handlers:
+                parent_found = parent
+                break
+        if parent_found:
+            for handler in parent_found.handlers:
                 logger.addHandler(handler)
-            logger.setLevel(parent_logger.level)
-            logger.propagate = parent_logger.propagate
+            logger.setLevel(parent_found.level)
+            logger.propagate = False  # 复制 handlers 后禁止再向上传播（避免重复）
         else:
-            # 没有父 logger，创建默认配置
+            # 没有找到有 handlers 的父 logger，创建默认配置
             return setup_logger(name)
     return logger
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.18.8",
+  "version": "1.19.0",
   "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": {
@@ -43,4 +43,4 @@
     "python": ">=3.10",
     "node": ">=18"
   }
-}
+}

package/requirements.txt

@@ -72,6 +72,12 @@ xlrd>=2.0.0
 # ============================================================
 # 语音识别 (本地 STT，默认启用)
 # ============================================================
+# [v1.18.8] SenseVoice (funasr) 作为首选引擎，中文识别极佳
+# torch/torchaudio 约 200MB (CPU版)，funasr 约 100MB
+# 若仅需 Whisper 备选引擎，可注释下面三行，保留 faster-whisper
+funasr>=1.1.0
+torch>=2.0.0
+torchaudio>=2.0.0
 faster-whisper>=1.0.0
 pydub>=0.25.1
 

package/setup.py

@@ -37,7 +37,10 @@ setup(
         "Pillow>=10.0.0",
         # 语音合成
         "edge-tts>=6.1.0",
-        # 语音识别 (本地 STT)
+        # 语音识别 (本地 STT) - [v1.18.8] SenseVoice 首选
+        "funasr>=1.1.0",
+        "torch>=2.0.0",
+        "torchaudio>=2.0.0",
         "faster-whisper>=1.0.0",
         # 浏览器自动化 (ChromeDev MCP, 无需 Playwright)
         # 桌面 GUI 自动化 (内置技能)
@@ -50,13 +53,16 @@ setup(
         "discord": ["discord.py>=2.3.0"],
         "anthropic": ["anthropic>=0.18.0"],
         "communication": ["cryptography>=41.0.0", "websockets>=12.0"],
-        "voice": ["faster-whisper>=1.0.0"],
+        "voice": ["funasr>=1.1.0", "torch>=2.0.0", "torchaudio>=2.0.0", "faster-whisper>=1.0.0"],
         "all": [
             "python-telegram-bot>=21.0",
             "discord.py>=2.3.0",
             "anthropic>=0.18.0",
             "cryptography>=41.0.0",
             "websockets>=12.0",
+            "funasr>=1.1.0",
+            "torch>=2.0.0",
+            "torchaudio>=2.0.0",
             "faster-whisper>=1.0.0",
         ],
     },

package/skills/chromedev_mcp.py

@@ -109,6 +109,8 @@ class MCPClient:
         self._lock = asyncio.Lock()
         # [v1.17.0] 有头模式 DISPLAY 覆盖
         self._display_override: Optional[str] = None
+        # [v1.19.0] stderr 日志收集（用于调试 MCP Server 启动失败）
+        self._stderr_buffer: List[str] = []
 
     async def start(self) -> bool:
         """启动 MCP Server 子进程"""
@@ -120,7 +122,7 @@ class MCPClient:
             logger.error("npx 不可用，请安装 Node.js (>= 20.19)")
             return False
 
-        args = ["npx", "-y", "chrome-devtools-mcp@latest"]
+        args = ["npx", "-y", "chrome-devtools-mcp@0.21.0"]
         if self._headless:
             args.append("--headless")
         if self._slim:
@@ -160,32 +162,48 @@ class MCPClient:
 
         try:
             # 启动 MCP Server 子进程
-            # 注意: stderr 必须用 DEVNULL 或在后台线程中持续读取
-            # 否则 stderr 管道缓冲区满后子进程会阻塞，导致 JSON-RPC 超时
+            # [v1.19.0] stderr 使用 PIPE 并在后台线程中持续读取，避免管道缓冲区满导致子进程阻塞
+            self._stderr_buffer = []
             self._process = subprocess.Popen(
                 args,
                 stdin=subprocess.PIPE,
                 stdout=subprocess.PIPE,
-                stderr=subprocess.DEVNULL,
+                stderr=subprocess.PIPE,
                 env=env,
             )
 
-            # 启动读取线程
+            # 启动读取线程（stdout 和 stderr 各一个，防止管道阻塞）
             loop = asyncio.get_event_loop()
             loop.run_in_executor(None, self._read_stdout)
+            loop.run_in_executor(None, self._read_stderr)
 
             # 执行 MCP 握手
             success = await self._handshake()
             if success:
                 self._initialized = True
                 logger.info("chrome-devtools-mcp 已连接并初始化")
+            else:
+                # 握手失败时，收集 stderr 输出用于诊断
+                stderr_output = self._get_stderr_tail()
+                if stderr_output:
+                    logger.error(f"MCP 握手失败，stderr 输出:\n{stderr_output}")
+                else:
+                    logger.error(
+                        "MCP 握手失败，无 stderr 输出。可能原因: "
+                        "1) Node.js 版本过低 (需要 >= 20.19); "
+                        "2) chrome-devtools-mcp@0.21.0 下载失败; "
+                        "3) Chrome/Chromium 未找到且自动安装失败"
+                    )
             return success
 
         except FileNotFoundError:
-            logger.error("无法启动 chrome-devtools-mcp，请确认 Node.js 已安装")
+            logger.error("无法启动 chrome-devtools-mcp: npx 命令未找到。请安装 Node.js >= 20.19: https://nodejs.org/")
             return False
         except Exception as e:
+            stderr_output = self._get_stderr_tail()
             logger.error(f"启动 chrome-devtools-mcp 失败: {e}")
+            if stderr_output:
+                logger.error(f"MCP Server stderr:\n{stderr_output}")
             self._cleanup()
             return False
 
@@ -389,6 +407,34 @@ class MCPClient:
             self._initialized = False
             logger.warning("MCP stdout 读取结束（Server 或 Chrome 已断开）")
 
+    def _read_stderr(self):
+        """[v1.19.0] 持续读取 MCP Server 的 stderr 并记录到日志。
+
+        必须在后台线程中持续读取 stderr，否则管道缓冲区满后
+        子进程会阻塞，导致 JSON-RPC 通信超时。
+        stderr 输出会同时收集到 _stderr_buffer 中，用于启动失败时的诊断。
+        """
+        if not self._process or not self._process.stderr:
+            return
+        try:
+            for raw_line in self._process.stderr:
+                line_text = raw_line.decode("utf-8", errors="replace").rstrip("\n")
+                if line_text:
+                    self._stderr_buffer.append(line_text)
+                    # 只保留最近的 200 行，防止内存无限增长
+                    if len(self._stderr_buffer) > 200:
+                        self._stderr_buffer = self._stderr_buffer[-100:]
+                    logger.debug(f"MCP stderr: {line_text}")
+        except Exception:
+            pass
+
+    def _get_stderr_tail(self, max_lines: int = 30) -> str:
+        """[v1.19.0] 获取最近的 stderr 输出（用于错误诊断）"""
+        if not self._stderr_buffer:
+            return ""
+        lines = self._stderr_buffer[-max_lines:]
+        return "\n".join(lines)
+
     async def _handshake(self) -> bool:
         """执行 MCP 初始化握手"""
         try:
@@ -652,6 +698,120 @@ class MCPClient:
         """检查 MCP Server 是否正在运行"""
         return self._process is not None and self._process.poll() is None
 
+    @staticmethod
+    def diagnose() -> Dict[str, Any]:
+        """[v1.19.0] 检查浏览器自动化环境是否就绪，返回诊断信息。
+
+        用于在浏览器启动失败时提供详细的错误原因。
+        检查项:
+          1. Node.js 是否安装及版本
+          2. npx 是否可用
+          3. Chrome/Chromium 是否找到
+          4. chrome-devtools-mcp npm 包是否可访问
+
+        Returns:
+            dict with keys: ok (bool), checks (list), summary (str)
+        """
+        results: Dict[str, Any] = {"ok": True, "checks": [], "summary": ""}
+
+        # 1. 检查 Node.js
+        node_path = shutil.which("node")
+        if not node_path:
+            results["ok"] = False
+            results["checks"].append({
+                "name": "Node.js", "status": "missing",
+                "message": "Node.js 未安装。需要 >= 20.19。请安装: https://nodejs.org/",
+            })
+        else:
+            try:
+                ver_result = subprocess.run(
+                    ["node", "--version"], capture_output=True, text=True, timeout=5,
+                )
+                version = ver_result.stdout.strip()
+                results["checks"].append({
+                    "name": "Node.js", "status": "ok",
+                    "message": f"Node.js {version} ({node_path})",
+                })
+            except Exception as e:
+                results["checks"].append({
+                    "name": "Node.js", "status": "error", "message": str(e),
+                })
+
+        # 2. 检查 npx
+        npx_path = shutil.which("npx")
+        if not npx_path:
+            results["ok"] = False
+            results["checks"].append({
+                "name": "npx", "status": "missing",
+                "message": "npx 未安装（通常随 Node.js 一起安装）",
+            })
+        else:
+            results["checks"].append({
+                "name": "npx", "status": "ok",
+                "message": f"npx 可用 ({npx_path})",
+            })
+
+        # 3. 检查 Chrome/Chromium
+        browser_path = MCPClient._detect_browser()
+        if browser_path:
+            results["checks"].append({
+                "name": "Chrome/Chromium", "status": "ok",
+                "message": f"浏览器已找到: {browser_path}",
+            })
+        else:
+            results["ok"] = False
+            results["checks"].append({
+                "name": "Chrome/Chromium", "status": "missing",
+                "message": (
+                    "未找到 Chrome/Chromium。系统会尝试自动安装，"
+                    "或手动安装: apt install -y chromium-browser 或 apt install -y chromium"
+                ),
+            })
+
+        # 4. 检查 npm / chrome-devtools-mcp 包可访问性
+        npm_path = shutil.which("npm")
+        if not npm_path:
+            results["checks"].append({
+                "name": "chrome-devtools-mcp (npm)", "status": "error",
+                "message": "npm 未安装，无法验证 chrome-devtools-mcp 包",
+            })
+        else:
+            try:
+                pkg_result = subprocess.run(
+                    ["npm", "view", "chrome-devtools-mcp", "version"],
+                    capture_output=True, text=True, timeout=30,
+                )
+                if pkg_result.returncode == 0 and pkg_result.stdout.strip():
+                    ver = pkg_result.stdout.strip()
+                    results["checks"].append({
+                        "name": "chrome-devtools-mcp (npm)", "status": "ok",
+                        "message": f"npm 注册表中可用，最新版本: {ver} (当前固定: 0.21.0)",
+                    })
+                else:
+                    results["checks"].append({
+                        "name": "chrome-devtools-mcp (npm)", "status": "warning",
+                        "message": (
+                            f"无法查询 npm 注册表: "
+                            f"{pkg_result.stderr[:200] if pkg_result.stderr else 'unknown error'}"
+                        ),
+                    })
+            except Exception as e:
+                results["checks"].append({
+                    "name": "chrome-devtools-mcp (npm)", "status": "error",
+                    "message": str(e),
+                })
+
+        # 构建摘要
+        failed = [c for c in results["checks"] if c["status"] in ("missing", "error")]
+        if not failed:
+            results["summary"] = "浏览器自动化环境就绪"
+        else:
+            results["summary"] = "环境检查发现问题: " + "; ".join(
+                f"{c['name']}({c['status']})" for c in failed
+            )
+
+        return results
+
 
 # ── 全局 MCP 客户端单例 ──────────────────────────────────────
 
@@ -902,18 +1062,49 @@ class BrowserOpenSkill(Skill):
         # 检查依赖
         dep_err = await asyncio.get_event_loop().run_in_executor(None, _ensure_node_deps)
         if dep_err:
-            return SkillResult(success=False, error=f"ChromeDev MCP 依赖缺失: {dep_err}")
+            # [v1.19.0] 依赖缺失时也运行诊断，给出完整信息
+            diag = await asyncio.get_event_loop().run_in_executor(None, MCPClient.diagnose)
+            diag_lines = [f"  - {c['name']}: {c['message']}" for c in diag.get("checks", [])]
+            return SkillResult(
+                success=False,
+                error=f"ChromeDev MCP 依赖缺失: {dep_err}\n环境诊断:\n" + "\n".join(diag_lines),
+            )
 
         global _mcp_client
         max_attempts = 2  # 最多重试 1 次
         for attempt in range(max_attempts):
             try:
-                client = await get_mcp_client()
+                # [v1.19.0] 添加超时保护，防止 get_mcp_client() 永久阻塞
+                # 首次启动可能需要下载 chrome-devtools-mcp 和 Chrome，给 90 秒
+                mcp_timeout = 90 if attempt == 0 else 60
+                try:
+                    client = await asyncio.wait_for(
+                        get_mcp_client(), timeout=mcp_timeout,
+                    )
+                except asyncio.TimeoutError:
+                    logger.error(f"获取 MCP 客户端超时 ({mcp_timeout}s)")
+                    diag = await asyncio.get_event_loop().run_in_executor(None, MCPClient.diagnose)
+                    diag_lines = [f"  - {c['name']}: {c['message']}" for c in diag.get("checks", [])]
+                    return SkillResult(
+                        success=False,
+                        error=(
+                            f"浏览器启动超时 ({mcp_timeout}s)。"
+                            f"首次启动可能需要下载 chrome-devtools-mcp 和 Chrome，请稍后重试。\n"
+                            f"环境诊断:\n" + "\n".join(diag_lines)
+                        ),
+                    )
+
                 if not client or not client.is_running() or not client._initialized:
+                    # [v1.19.0] 启动失败时运行完整诊断
+                    diag = await asyncio.get_event_loop().run_in_executor(None, MCPClient.diagnose)
+                    diag_lines = [f"  - {c['name']}: {c['message']}" for c in diag.get("checks", [])]
                     return SkillResult(
                         success=False,
-                        error="浏览器启动失败。系统未安装 Chrome/Chromium，自动安装也未成功。"
-                              "请手动执行: apt install -y chromium-browser 或 apt install -y chromium"
+                        error=(
+                            "浏览器启动失败。系统未安装 Chrome/Chromium 或 MCP Server 初始化失败。\n"
+                            "环境诊断:\n" + "\n".join(diag_lines) + "\n\n"
+                            "建议: apt install -y chromium-browser 或 apt install -y chromium"
+                        ),
                     )
 
                 # 1. 导航到目标 URL