myagent-ai 1.16.19 → 1.17.1

package/agents/main_agent.py

@@ -701,9 +701,15 @@ class MainAgent(BaseAgent):
                         )
                     break
 
-                # [v1.16.13] 特殊处理模型不支持图片输入 — 去掉图片用纯文本重试
-                _vision_keywords = ["doesn't support image", "does not support image", "model_incompatible", "image input", "not support vision", "unsupported multimodal", "image capability"]
-                if any(kw in _llm_error.lower() for kw in _vision_keywords) and context.metadata.get("user_images"):
+                # [v1.16.13→17.0] 特殊处理模型不支持图片输入 — 去掉图片用纯文本重试
+                # 支持中英文错误关键词匹配（ModelScope 等国产 API 可能返回中文错误）
+                _vision_keywords = [
+                    "doesn't support image", "does not support image", "model_incompatible",
+                    "image input", "not support vision", "unsupported multimodal", "image capability",
+                    "不支持图片", "不支持图像", "图片输入", "图像输入", "不支持多模态",
+                    "视觉", "image_url", "multimodal", "vision",
+                ]
+                if any(kw.lower() in _llm_error.lower() for kw in _vision_keywords) and context.metadata.get("user_images"):
                     logger.warning(f"[{task_id}] 模型不支持图片输入，去掉图片用纯文本重试")
                     context.metadata["user_images"] = []
                     # 用纯文本消息替换最后一条多模态消息

package/config.py

@@ -316,8 +316,21 @@ class ConfigManager:
             elif isinstance(value, list) and key == "models_library":
                 models = []
                 for item in value:
+                    # [v1.17.0] 兼容 "input" → "input_modes" 字段映射
+                    # 用户配置中常使用 "input": ["text", "image"]，内部字段名是 "input_modes"
+                    normalized = dict(item)
+                    if "input" in normalized and "input_modes" not in normalized:
+                        normalized["input_modes"] = normalized.pop("input")
+                    if "baseUrl" in normalized and "base_url" not in normalized:
+                        normalized["base_url"] = normalized.pop("baseUrl")
+                    if "apiKey" in normalized and "api_key" not in normalized:
+                        normalized["api_key"] = normalized.pop("apiKey")
+                    if "reasoning" in normalized:
+                        # 确保 reasoning 是 bool 类型
+                        v = normalized["reasoning"]
+                        normalized["reasoning"] = bool(v) if not isinstance(v, bool) else v
                     me = ModelEntry(**{
-                        k: v for k, v in item.items() if k in ModelEntry.__dataclass_fields__
+                        k: v for k, v in normalized.items() if k in ModelEntry.__dataclass_fields__
                     })
                     models.append(me)
                 setattr(target, key, models)

package/core/deps_checker.py

@@ -110,6 +110,10 @@ DEPENDENCIES: List[DepInfo] = [
             note="鼠标/键盘控制 (~500KB)"),
     DepInfo("pygetwindow", "pygetwindow", "0.0.9", "gui", "windows_macos",
             note="窗口管理 (仅 Windows/macOS)"),
+
+    # ── 远程桌面 (VNC) [v1.17.0] ──
+    # VNC 依赖通过 apt 安装 (xvfb, x11vnc)，websockify 通过 pip 安装
+    # 实际检测和安装在 VNCManager.ensure_dependencies() 中完成
 ]