v1.6.0

更新内容

• 重构 AI Agent 运行时为 Agent First 执行架构，统一自主循环、回复提交、中断恢复、上下文评估和运行轨迹处理，提升复杂客服任务的执行稳定性与可观测性。
• 新增 Agent 修订版本、发布状态和运行记录管理，支持保存当前配置后发布、查看步骤与工具调用详情、质量反馈和运行评估，并确保渠道只能绑定已发布 Agent。
• 升级 MCP 工具管理与执行策略，统一工具发现、风险分级、管理员授权和敏感操作确认，补充工具结果归纳与安全回复处理。
• 全面升级可视化工作流编辑器，基于 FlowGram 提供节点配置、条件分支、变量契约、自动布局、撤销重做、校验反馈、版本发布恢复和使用情况查询。
• 完善 Agent 与已发布工作流版本的绑定机制，发布后保持活动修订版本一致，并优化 Agent 配置工作台、渠道选择和工作流管理交互。
• 构建命令统一迁移到 Taskfile，并补充 LanceDB 原生依赖的下载、构建与开发任务，简化本地开发和多环境构建流程。

Updates

• Rebuilt the AI Agent runtime around an Agent First execution architecture, consolidating autonomous loops, reply commits, interruption recovery, context evaluation, and run tracing for more reliable and observable support automation.
• Added Agent revision, publication, and run management, including save-before-publish behavior, step and tool-call details, quality feedback, run evaluation, and safeguards that allow channels to bind only published Agents.
• Upgraded MCP tool management and execution policies with unified discovery, risk levels, administrator authorization, sensitive-action confirmation, tool-result summarization, and safer reply handling.
• Overhauled the visual workflow editor with FlowGram-based node configuration, conditional branches, variable contracts, auto layout, undo/redo, validation feedback, version publishing and restoration, and usage inspection.
• Completed binding between Agents and published workflow versions, kept active revisions synchronized after publication, and refined the Agent configuration workbench, channel selection, and workflow-management interactions.
• Standardized build commands on Taskfile and added LanceDB native dependency download, build, and development tasks to simplify local development and multi-environment builds.