myagent-ai 1.47.22 → 1.47.24

package/agents/main_agent.py

@@ -943,16 +943,77 @@ class MainAgent(BaseAgent):
                 _reasoning_parts.append(delta_text)
                 await self._emit_v2_event("v2_reasoning", {"content": delta_text}, stream_callback)
 
-            if stream_response and self.llm:
-                response = await self._call_llm_stream(
-                    messages,
-                    tools=tools if tools else None,
-                    text_delta_callback=text_delta_callback,
-                    reasoning_delta_callback=_reasoning_delta_cb,
-                    stream_response=stream_response,
-                )
-            else:
-                response = await self._call_llm(messages, tools=tools if tools else None)
+            # 可重试的 LLM 调用（最多 5 次）
+            _max_llm_retries = 5
+            _llm_retry_count = 0
+            response = None
+
+            while _llm_retry_count < _max_llm_retries:
+                try:
+                    if stream_response and self.llm:
+                        response = await self._call_llm_stream(
+                            messages,
+                            tools=tools if tools else None,
+                            text_delta_callback=text_delta_callback,
+                            reasoning_delta_callback=_reasoning_delta_cb,
+                            stream_response=stream_response,
+                        )
+                    else:
+                        response = await self._call_llm(messages, tools=tools if tools else None)
+                except Exception as _llm_exc:
+                    # 异常类错误 → 判断是否可重试
+                    _llm_exc_str = str(_llm_exc).lower()
+                    _is_retryable = any(kw in _llm_exc_str for kw in (
+                        "connection", "timeout", "timed out",
+                        "429", "500", "502", "503", "504",
+                        "rate_limit", "rate limit", "overloaded", "capacity",
+                        "network", "eof",
+                    ))
+                    _llm_retry_count += 1
+                    if _is_retryable and _llm_retry_count < _max_llm_retries:
+                        _delay = 2.0 * (2 ** (_llm_retry_count - 1))  # 2s, 4s, 8s, 16s
+                        logger.warning(
+                            f"[{task_id}] LLM 调用异常 (第 {_llm_retry_count}/{_max_llm_retries} 次)，"
+                            f"{_delay:.0f}s 后重试: {_llm_exc}"
+                        )
+                        await self._emit_v2_event("v2_reasoning", {
+                            "content": f"⏳ 网络不稳定，正在重试 ({_llm_retry_count}/{_max_llm_retries})..."
+                        }, stream_callback)
+                        await asyncio.sleep(_delay)
+                        continue
+                    else:
+                        # 不可重试 或 已达上限 → 包装为失败 response
+                        logger.error(f"[{task_id}] LLM 调用异常 (已重试 {_llm_retry_count} 次): {_llm_exc}")
+                        response = type("LLMResponse", (), {"success": False, "error": str(_llm_exc)})()
+                        break
+
+                # 没有异常 → 检查 response.success
+                if response.success:
+                    break  # 成功 → 跳出重试循环
+
+                # response.success == False 但没抛异常 → 判断错误是否可重试
+                _llm_error = (response.error or "").lower()
+                _is_retryable = any(kw in _llm_error for kw in (
+                    "connection", "timeout", "timed out",
+                    "429", "500", "502", "503", "504",
+                    "rate_limit", "rate limit", "overloaded", "capacity",
+                    "network", "eof",
+                ))
+                _llm_retry_count += 1
+                if _is_retryable and _llm_retry_count < _max_llm_retries:
+                    _delay = 2.0 * (2 ** (_llm_retry_count - 1))
+                    logger.warning(
+                        f"[{task_id}] LLM 返回失败 (第 {_llm_retry_count}/{_max_llm_retries} 次)，"
+                        f"{_delay:.0f}s 后重试: {response.error}"
+                    )
+                    await self._emit_v2_event("v2_reasoning", {
+                        "content": f"⏳ 网络不稳定，正在重试 ({_llm_retry_count}/{_max_llm_retries})..."
+                    }, stream_callback)
+                    await asyncio.sleep(_delay)
+                    continue
+                else:
+                    # 不可重试 或 已达上限 → 保持失败 response，退出循环
+                    break
 
             # LLM 调用失败处理
             if not response.success:
@@ -982,8 +1043,8 @@ class MainAgent(BaseAgent):
                     await self._emit_v2_event("v2_reasoning", {"content": _vision_skip_msg}, stream_callback)
                     break
 
-                # 其他错误
-                error_msg = f"LLM 调用失败: {response.error}"
+                # 其他错误（含已耗尽重试次数的连接错误）
+                error_msg = f"LLM 调用失败 (已重试 {_llm_retry_count} 次): {_llm_error}"
                 context.working_memory["final_response"] = error_msg
                 await self._emit_v2_event("v2_reasoning", {"content": error_msg}, stream_callback)
                 break

package/aiskills/browser_stealth.py

@@ -2124,6 +2124,10 @@ class StealthBrowser:
                         message=f"Firefox+VNC 截图已保存",
                         data={"path": save_path},
                     )
+                logger.warning(
+                    f"[_firefox_screenshot] import 截图失败 (rc={result.returncode}): "
+                    f"{result.stderr.decode(errors='ignore')[:200]}"
+                )
 
             # 方法2: 使用 xdotool + scrot
             scrot_cmd = shutil.which("scrot")
@@ -2140,6 +2144,10 @@ class StealthBrowser:
                         message=f"Firefox+VNC 截图已保存 (scrot)",
                         data={"path": save_path},
                     )
+                logger.warning(
+                    f"[_firefox_screenshot] scrot 截图失败 (rc={result.returncode}): "
+                    f"{result.stderr.decode(errors='ignore')[:200]}"
+                )
 
             # 方法3: 使用 xwd + convert
             xwd_cmd = shutil.which("xwd")
@@ -2168,21 +2176,65 @@ class StealthBrowser:
                             message=f"Firefox+VNC 截图已保存",
                             data={"path": save_path},
                         )
+                logger.warning(
+                    f"[_firefox_screenshot] xwd+convert 截图失败 (rc={result.returncode}): "
+                    f"{result.stderr.decode(errors='ignore')[:200]}"
+                )
+
+            # 方法4: 使用 xdotool + xwd + ffmpeg (proot 兼容)
+            xdotool_cmd = shutil.which("xdotool")
+            ffmpeg_cmd = shutil.which("ffmpeg")
+            if xdotool_cmd and ffmpeg_cmd:
+                xwd_tmp = save_path + ".xwd"
+                # 用 xdotool 获取活动窗口并截图
+                result = subprocess.run(
+                    [xwd_cmd or "xwd", "-root", "-display", display, "-out", xwd_tmp],
+                    capture_output=True, timeout=10,
+                    env=env, start_new_session=True,
+                )
+                if result.returncode == 0 and os.path.isfile(xwd_tmp):
+                    # 用 ffmpeg 转换 xwd 到 png
+                    result2 = subprocess.run(
+                        [ffmpeg_cmd, "-y", "-i", xwd_tmp, save_path],
+                        capture_output=True, timeout=10,
+                        env=env, start_new_session=True,
+                    )
+                    try:
+                        os.unlink(xwd_tmp)
+                    except Exception:
+                        pass
+                    if result2.returncode == 0 and os.path.isfile(save_path):
+                        logger.info(f"Firefox 截图已保存 (xwd+ffmpeg): {save_path}")
+                        return SkillResult(
+                            success=True,
+                            message=f"Firefox+VNC 截图已保存",
+                            data={"path": save_path},
+                        )
+
+            _available = []
+            if import_cmd: _available.append("import")
+            if scrot_cmd: _available.append("scrot")
+            if xwd_cmd and convert_cmd: _available.append("xwd+convert")
+            if xdotool_cmd and ffmpeg_cmd: _available.append("xdotool+ffmpeg")
+            _tried = ", ".join(_available) if _available else "无"
 
             return SkillResult(
                 success=False,
-                error="Firefox+VNC 截图失败: 没有可用的截图工具 "
-                      "(需要 ImageMagick import / scrot / xwd+convert)",
+                error=f"Firefox+VNC 截图失败 (尝试过: {_tried}) "
+                      f"— 需要 ImageMagick import / scrot / xwd+convert",
             )
         except Exception as e:
             return SkillResult(success=False, error=f"Firefox+VNC 截图失败: {e}")
 
-    def _firefox_read_sessionstore(self) -> dict:
+    def _firefox_read_sessionstore(self, max_wait: float = 5.0) -> dict:
         """[v1.47.20] 读取 Firefox sessionstore 获取当前标签页 URL/标题。
 
         Firefox 运行时会将当前会话信息写入 sessionstore-backups/recovery.jsonlz4。
         该文件使用 mozLz4 格式（8字节自定义头 + LZ4 压缩数据）。
 
+        Args:
+            max_wait: 最长等待时间（秒），给 Firefox 时间写入 sessionstore
+
         Returns:
             dict: {"url": str, "title": str, "tabs": [{"url": str, "title": str}]}
         """
@@ -2193,30 +2245,56 @@ class StealthBrowser:
         if self._firefox_profile_dir:
             search_dirs.append(self._firefox_profile_dir)
         # 也搜索 Firefox 默认 profile 和 vnc_manager 启动的 profile
+        ff_base = os.path.expanduser("~/.mozilla/firefox")
         for extra in [
-            os.path.expanduser("~/.mozilla/firefox/default"),
-            os.path.expanduser("~/.mozilla/firefox"),
+            os.path.join(ff_base, "default"),
+            ff_base,
         ]:
             if os.path.isdir(extra) and extra not in search_dirs:
                 search_dirs.append(extra)
+        # [v1.47.24] 搜索所有 Firefox profile 子目录（如 xxx.default-release）
+        try:
+            if os.path.isdir(ff_base):
+                for entry in os.listdir(ff_base):
+                    entry_path = os.path.join(ff_base, entry)
+                    if os.path.isdir(entry_path) and entry_path not in search_dirs:
+                        # Firefox profile 目录通常包含 user.js 或 prefs.js
+                        if os.path.isfile(os.path.join(entry_path, "prefs.js")) or \
+                           os.path.isfile(os.path.join(entry_path, "user.js")):
+                            search_dirs.append(entry_path)
+        except Exception:
+            pass
 
-        recovery_files = []
-        for base_dir in search_dirs:
-            ss_dir = os.path.join(base_dir, "sessionstore-backups")
-            if os.path.isdir(ss_dir):
-                for fname in ("recovery.jsonlz4", "recovery.baklz4",
-                              "previous.jsonlz4"):
-                    fpath = os.path.join(ss_dir, fname)
+        # [v1.47.24] 等待 Firefox 写入 sessionstore（刚导航完可能还没写入）
+        _start_time = time.time()
+        while True:
+            recovery_files = []
+            for base_dir in search_dirs:
+                ss_dir = os.path.join(base_dir, "sessionstore-backups")
+                if os.path.isdir(ss_dir):
+                    for fname in ("recovery.jsonlz4", "recovery.baklz4",
+                                  "previous.jsonlz4"):
+                        fpath = os.path.join(ss_dir, fname)
+                        if os.path.isfile(fpath):
+                            recovery_files.append(fpath)
+                # 也检查 base_dir 本身（有些 Firefox 版本）
+                for fname in ("sessionstore.jsonlz4", "sessionstore-backups/recovery.jsonlz4"):
+                    fpath = os.path.join(base_dir, fname)
                     if os.path.isfile(fpath):
                         recovery_files.append(fpath)
-            # 也检查 base_dir 本身（有些 Firefox 版本）
-            for fname in ("sessionstore.jsonlz4", "sessionstore-backups/recovery.jsonlz4"):
-                fpath = os.path.join(base_dir, fname)
-                if os.path.isfile(fpath):
-                    recovery_files.append(fpath)
+
+            # 如果找到了文件，或者超时了，就跳出
+            if recovery_files or (time.time() - _start_time) >= max_wait:
+                break
+
+            # 等 0.5 秒再试
+            time.sleep(0.5)
 
         if not recovery_files:
-            logger.debug("[_firefox_read_sessionstore] 未找到 sessionstore 文件")
+            logger.warning(
+                f"[_firefox_read_sessionstore] 未找到 sessionstore 文件 "
+                f"(搜索了 {len(search_dirs)} 个目录: {[d.replace(os.path.expanduser('~'), '~') for d in search_dirs]})"
+            )
             return result_info
 
         # 读取 mozLz4 格式
@@ -2290,9 +2368,19 @@ class StealthBrowser:
         screenshot_path = ""
         if screenshot_result.success and screenshot_result.data:
             screenshot_path = screenshot_result.data.get("path", "")
+            logger.info(f"[_firefox_get_content] 截图成功: {screenshot_path}")
+        else:
+            logger.warning(
+                f"[_firefox_get_content] 截图失败: "
+                f"{screenshot_result.error or '无数据'}"
+            )
 
         # 2. 读取 sessionstore
         session_info = self._firefox_read_sessionstore()
+        logger.info(
+            f"[_firefox_get_content] sessionstore: url={session_info.get('url', '')}, "
+            f"title={session_info.get('title', '')}, tabs={len(session_info.get('tabs', []))}"
+        )
 
         # 3. 组合返回信息
         url = session_info.get("url", "")
@@ -2315,7 +2403,17 @@ class StealthBrowser:
         if screenshot_path:
             content_parts.append(f"截图: {screenshot_path}")
 
-        content_text = "\n".join(content_parts) if content_parts else "无法获取页面内容"
+        if not content_parts:
+            # 全部失败 → 至少给截图失败原因
+            fallback_msg = "Firefox+VNC 无法获取页面内容"
+            if not screenshot_result.success:
+                fallback_msg += f" (截图失败: {screenshot_result.error})"
+            if not session_info.get("url"):
+                fallback_msg += " (sessionstore 未读取到 URL)"
+            logger.error(f"[_firefox_get_content] {fallback_msg}")
+            content_text = fallback_msg
+        else:
+            content_text = "\n".join(content_parts)
 
         return SkillResult(
             success=True,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.47.22",
+  "version": "1.47.24",
   "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": {