feat(workflow): add per-agent vision configuration

backend/routes/voice_webrtc.py

@@ -163,8 +163,13 @@ async def _handle_offer_payload(payload, peers):
     else:
         cfg = await _resolve_config(offer)  # 解析放在建连前,配置错就别建连
         # 服务端助手配置是视觉理解的唯一授权来源；客户端 offer 只负责携带媒体轨。
-        vision_enabled = cfg.vision_enabled
-        if vision_enabled:
+        if cfg.type == "workflow":
+            from services.workflow_engine import WorkflowEngine
+
+            vision_enabled = WorkflowEngine(cfg.graph).uses_vision()
+        else:
+            vision_enabled = cfg.vision_enabled
+        if vision_enabled and cfg.type != "workflow":
             has_native_vision = (
                 not cfg.vision_model_resource_id and cfg.llm_support_image_input
             )