v1.5.3

更新内容

• 新增面向 AI Agent 的可视化工作流能力，支持工作流 DSL 校验、版本发布、默认工作流恢复、节点变量映射和条件分支配置。
• 新增 AI 工作流运行时与节点执行链路，覆盖知识检索、LLM 回复、转人工、会话理解、回复策略、创建工单、人工确认和兜底处理等客服场景。
• 新增工作流运行审计页面和详情图谱，支持查看节点执行、分支决策、错误信息、运行状态和会话消息关联，便于排查自动化接待过程。
• 重构 AI Agent 配置工作台，统一系统提示词、工作流版本、发布记录和编辑入口，并优化节点编辑器、拖拽连线、撤销重做、辅助线和校验反馈。
• 优化会话与消息状态处理，移除 seqNo 读状态依赖，改用消息 ID 语义保持未读、排序和实时同步逻辑一致。
• 修复 DashScope 模型调用参数处理，自动关闭不支持模型的 thinking 参数，避免相关工作流 LLM 调用因参数不兼容超时或失败。

Updates

• Added visual workflow capabilities for AI Agents, including workflow DSL validation, version publishing, default workflow restoration, node variable mapping, and conditional branch configuration.
• Added the AI workflow runtime and node execution path for support scenarios such as knowledge retrieval, LLM replies, human handoff, conversation understanding, reply policy, ticket creation, human confirmation, and fallback handling.
• Added workflow run audit pages and detail graphs so operators can inspect node execution, branch decisions, errors, run status, and linked conversation messages when troubleshooting automated support flows.
• Reworked the AI Agent configuration workbench around system prompts, workflow versions, release records, and editing entry points, while improving node editing, drag connections, undo/redo, helper lines, and validation feedback.
• Improved conversation and message state handling by removing the seqNo read-state dependency and relying on message ID semantics for unread counts, ordering, and realtime synchronization.
• Fixed DashScope model parameter handling by automatically disabling unsupported thinking parameters, preventing related workflow LLM calls from timing out or failing due to incompatible request options.