Senparc.Xncf.AgentsManager 0.18.3

Prefix Reserved
.NET 10.0 This package targets .NET 10.0. The package is compatible with this framework or higher. 
dotnet add package Senparc.Xncf.AgentsManager --version 0.18.3
                    


                    

                
NuGet\Install-Package Senparc.Xncf.AgentsManager -Version 0.18.3
This command is intended to be used within the Package Manager Console in Visual Studio, as it uses the NuGet module's version of Install-Package. 
<PackageReference Include="Senparc.Xncf.AgentsManager" Version="0.18.3" />
For projects that support PackageReference, copy this XML node into the project file to reference the package. 
<PackageVersion Include="Senparc.Xncf.AgentsManager" Version="0.18.3" />
                    


                            Directory.Packages.props
                        

                    

                
<PackageReference Include="Senparc.Xncf.AgentsManager" />
                    


                            Project file
For projects that support Central Package Management (CPM), copy this XML node into the solution Directory.Packages.props file to version the package. 
paket add Senparc.Xncf.AgentsManager --version 0.18.3
                    


                    

                
#r "nuget: Senparc.Xncf.AgentsManager, 0.18.3"
#r directive can be used in F# Interactive and Polyglot Notebooks. Copy this into the interactive tool or source code of the script to reference the package. 
#:package Senparc.Xncf.AgentsManager@0.18.3
#:package directive can be used in C# file-based apps starting in .NET 10 preview 4. Copy this into a .cs file before any lines of code to reference the package. 
#addin nuget:?package=Senparc.Xncf.AgentsManager&version=0.18.3
                    


                            Install as a Cake Addin
                        

                    

                
#tool nuget:?package=Senparc.Xncf.AgentsManager&version=0.18.3
                    


                            Install as a Cake Tool

Senparc.Xncf.AgentsManager

Senparc.Xncf.AgentsManager is an NCF module for managing reusable AI-agent templates, chat groups, tasks, collaboration graphs, prompt optimization, and usage records.

Features

  • Stores and manages AgentTemplate, ChatGroup, ChatGroupMember, ChatTask, and chat history data.
  • Supports group execution, task streaming, agent collaboration graphs, and snapshot responses.
  • Provides prompt optimization and prompt-catalyzer application services.
  • Lets each AgentTemplate bind to one published KnowledgeBase without a cross-module database foreign key.
  • Shows per-Agent completed conversation rounds/tasks, prompt/completion/total tokens, average response time, and last activity on the management page.
  • Retrieves bounded KnowledgeBase context before model execution, labels it as untrusted external data, and falls back to the model when retrieval is unavailable.
  • Exposes local application-service requests and DTOs suitable for NCF function rendering.
  • Includes provider-specific EF Core contexts for NCF's multi-database model.
  • Supports A2A remote agents as first-class ChatGroup members while preserving the existing local AgentTemplate and ChatGroupMember data path.
  • Supports per-group context sharing: legacy full history (local-only), instruction plus bounded text conclusion (the secure mixed-group default), or instruction only.
  • Includes an interactive 3D Agent Studio with team desks, an agent lounge, a task board, and a same-screen inspector.

Agent Studio

Open Agents → 3D view (#tab=first&view=three). The existing panel, statistics and full detail pages remain available.

  • Drag an enabled local or remote A2A agent from the roster or directly in the 3D scene to a team desk / group drop target. Membership is appended idempotently; other members, roles, enable state and context-sharing settings are preserved. This adds membership rather than moving the agent out of its other teams.
  • Use + or drop agents into the Team tray, then Form team & run. Choose a local, non-human owner and entry agent, a task title, instructions and an optional chat model. You can create just the team or submit a task immediately. A saved team is retained when task submission fails, so retry does not create another team.
  • Select agents, desks or task cards to inspect and manage them without leaving the studio. The inspector supports editing/enabling agents and groups, removing ordinary members, starting group tasks, reading bounded conversation pages, stopping active tasks and archiving completed tasks. Owner/entry agents must be replaced through the group editor before removal.
  • New agents, groups (including empty groups), membership changes, task statuses and names refresh every three seconds while the studio is visible. Waiting, running, paused, finished, cancelled and failed tasks are filterable; pending human requests link to the existing approval/input detail page.
  • Task submission waits for a persisted task ID, but model execution remains asynchronous. Initialization failures before task creation are returned to the composer; watch task status/output for the actual execution result. Membership edits affect future executions; already running tasks retain their captured participants.
  • Orbit empty space, scroll to zoom, or use Overview to fit the room. Reduced-motion preferences are respected. If WebGL is unavailable, the roster, inspector and team/task operations remain usable.

Studio APIs (authenticated application-service routes):

  • CreateStudioTeam(StudioTeamRequest): validates all participants/roles before transactionally saving the group and members.
  • SetStudioParticipant(groupId, participantKey, remove): incrementally adds/removes a local:{id} or remote:{id} participant under a per-group cache lock.
  • StartStudioTask(ChatGroup_RunGroupRequest, chatGroupId): returns the persisted task reference without waiting for model completion, and propagates initialization failures.
  • GetAgentGraphSnapshot: includes lightweight non-archived Tasks alongside agents, groups, links and active collaborations.

Studio regression checks

Run frontend state tests with node --test src/Extensions/Senparc.Xncf.AgentsManagerTests/Frontend/studio.test.cjs. For browser checks, start node src/Extensions/Senparc.Xncf.AgentsManagerTests/Frontend/studio.fixture.cjs, then run playwright-cli -s=ncf-studio open http://127.0.0.1:51961/ and playwright-cli -s=ncf-studio run-code --filename=src/Extensions/Senparc.Xncf.AgentsManagerTests/Frontend/studio.browser.spec.js. The fixture uses real studio views/scripts and mock APIs; it never writes to the host database or calls an LLM. Run AgentStudioTests.Studio_* through VSTest with EnableMSTestRunner=false, TestingPlatformDotnetTestSupport=false, OutputType=Library, CopyLocalLockFileAssemblies=true and GenerateRuntimeConfigurationFiles=true; the existing executable entry point otherwise runs only its manually listed smoke checks.

The studio reuses the existing Vue, Element UI, Three.js and OrbitControls assets. The containing Razor page must register the MVC Tag Helpers so _Studio, localization scripts and versioned asset URLs are rendered, rather than leaving a literal <partial> tag in the otherwise empty 3D branch. Switching back to the panel or statistics view disposes the renderer; returning to 3D recreates it. Run AgentStudioRenderingTests with the same VSTest properties to verify the compiled page and script/style sections. For a browser regression with the original Vue lifecycle and hash navigation, set NCF_STUDIO_RENDERED_PAGE to the same absolute temporary JSON path when running this test and the fixture, then execute playwright-cli -s=ncf-studio run-code --filename=src/Extensions/Senparc.Xncf.AgentsManagerTests/Frontend/studio.rendering.browser.spec.js. This additional check uses the actual compiled page output, not a manually inserted partial.

Installation

<PackageReference Include="Senparc.Xncf.AgentsManager" Version="0.13.0-preview8" />

Key API

  • AgentTemplateService and AgentTemplateAppService create, find, configure, and inspect agent templates.
  • ChatGroupService, ChatGroupMemberService, and ChatGroupHistoryService manage group membership and conversation records.
  • ChatTaskService, ChatTaskAppService, and ChatTaskStreamController coordinate task execution and streaming output.
  • PromptOptimizationAppService and PromptOptimizationAgentBridge integrate prompt improvement workflows.
  • AgentsTemplateService and the AgentGraph* DTOs support reusable collaboration graphs.

KnowledgeBase-first RAG policy

KnowledgeBase-first is implemented as retrieve-first, conditionally augment, not as a forced answer override. A bound Agent queries its published collection with the user's request, receives at most five chunks and 6,000 characters, and is instructed to use only relevant evidence and identify gaps or conflicts. Empty or failed retrieval is fail-open so an unavailable vector service does not stop the whole multi-Agent task.

This policy is appropriate because it improves domain grounding without pretending every query belongs to the knowledge base. It still needs production evaluation for similarity thresholds/reranking, document ACLs, prompt-injection filtering, and citation rendering. Token totals are based on provider usage metadata stored with completed chat history; providers that do not return usage cannot be estimated exactly.

A2A remote agent configuration

Use the 远程 A2A 智能体 entry on the Agent management page to register an Agent Card root URL or a /.well-known/agent-card.json URL, test card discovery, and add it to a ChatGroup alongside local agents. The current phase keeps the group owner and entry agent local; the remote agent participates in normal collaboration rounds.

Remote authentication is deliberately reference-only. The database stores AuthSecretKey, never a token value. Configure the real secret in the deployment configuration, for example:

{
  "A2A": {
    "Secrets": {
      "research-agent-token": "replace-with-deployment-secret"
    }
  }
}

Publish a local Agent as an A2A service

An existing local AgentTemplate can also be exposed to another NCF installation, or to any A2A-compatible client, without coupling the two systems. Open the local Agent editor and select 配置 / 发布为 A2A Agent. The publishing configuration is stored in the additive PublishedA2AAgent table; it does not change existing AgentTemplate, ChatGroup, or ChatGroupMember records.

When enabled, the standard discovery URL is:

https://your-public-host/a2a/{public-agent-key}/.well-known/agent-card.json

The Agent Card advertises the standard A2A JSON-RPC interface at https://your-public-host/a2a/{public-agent-key}. A second NCF system can paste the discovery URL into its 远程 A2A 智能体 configuration, test the card, then add that remote Agent to a ChatGroup.

For reverse-proxy deployments, configure the externally visible base address so the Agent Card never advertises an internal host name:

{
  "A2A": {
    "PublicBaseUrl": "https://your-public-host",
    "InboundSecrets": {
      "partner-ncf": "replace-with-deployment-secret"
    }
  }
}

Bearer Token and custom-header incoming authentication read only from A2A:InboundSecrets:{AuthSecretKey}. The secret body is not stored in the database, returned by the admin API, included in logs, or put into the Agent Card. Publishing is disabled by default. Function Calling and MCP tools are also disabled by default for A2A requests; enabling them is an explicit per-published-Agent decision for trusted callers only.

The A2A boundary instructs the local Agent to return shareable conclusions rather than hidden reasoning, system prompt text, secrets, or tool traces. It cannot prevent a model from intentionally placing sensitive content in a normal response, so public Agent descriptions, prompts, connected knowledge bases, and enabled tools still require a deliberate data-access review.

Choose research-agent-token as the remote agent's deployment key name. The default mixed-group policy forwards only the initial instruction and the current round's bounded text conclusion. It removes tool calls, raw provider representations, usage data, and earlier history from the participant broadcast. This reduces exposure but does not transform an agent's voluntarily emitted text into a guaranteed semantic summary; prompts therefore require a concise shared conclusion and prohibit chain-of-thought/private content.

Product Compatible and additional computed target framework versions.
.NET net10.0 is compatible.  net10.0-android was computed.  net10.0-browser was computed.  net10.0-ios was computed.  net10.0-maccatalyst was computed.  net10.0-macos was computed.  net10.0-tvos was computed.  net10.0-windows was computed. 
Compatible target framework(s)
Included target framework(s) (in package)
Learn more about Target Frameworks and .NET Standard.

NuGet packages

This package is not used by any NuGet packages.

GitHub repositories

This package is not used by any popular GitHub repositories.

Version Downloads Last Updated
0.18.3 49 10/10/2026
0.18.1 79 9/25/2026
0.18.0 78 9/18/2026
0.17.2 73 9/16/2026
0.17.1 80 9/15/2026
0.17.0 95 8/29/2026
0.16.0 157 8/23/2026
0.15.0-preview20 86 8/14/2026
0.15.0-preview11 86 8/12/2026
0.14.1-preview10 106 8/8/2026
0.14.0-preview9 131 8/3/2026
0.13.0-preview8 123 7/31/2026
0.12.1-preview7 106 7/28/2026
0.12.0-preview6 80 7/26/2026
0.11.3-preview5 75 7/15/2026
0.11.2-preview4 81 7/5/2026
0.11.0-preview2 83 7/3/2026
0.10.1-preview1 108 6/17/2026
0.9.20.1-preview 190 5/22/2026
0.9.2-preview 90 6/16/2026
Loading failed

[2026-10-06] Fix Agent Studio Razor partial rendering and validate the compiled page with the existing Three.js components.
       [2026-10-06] Add interactive Agent Studio, incremental team membership, quick team creation and task management.
v0.1.0 创世
[2024-08-06] v0.4.4.0-preview1 update Senparc.AI.Kernel
[2024-08-31] v0.5.0-preview1 PromptRange 支持 Markdown 输出格式展示
[2024-09-03] v0.5.1-preview1 PromptRange 结果提供“复制”功能
[2024-09-04] v0.5.2-preview1 PromptRange 完善结果显示样式，添加“复制”功能
[2024-09-07] v0.5.3-preview1 XncfBuilder 更新默认 Prompt
[2024-09-07] v0.5.3.1-preview1 更新 PromptRange 中代码自动生成的 Prompt
[2024-09-07] v0.5.3.2-preview1 PromptRange 优化输出格式
[2024-09-07] v0.5.3.3-preview1 PromptRange 修复自动初始化 Prompt 调用过程中的一个 bug
[2024-09-07] v0.5.3.4-preview1 更新 XNCF AI 代码生成逻辑
[2024-09-07] v0.5.4-preview1 更新 Senparc.AI 基础库，并在 XncfBuilder 使用 AI 生成代码时，优先使用已选中的模型。
[2024-09-07] v0.5.5-preview1 更新 Senparc.AI 基础库，并在 XncfBuilder 使用 AI 生成代码时，优先使用已选中的模型。
[2024-09-07] v0.5.5.1-preview1 优化 XncfBuilder AI 代码生成过程中的 AI 模型选用
[2024-09-08] v0.5.5.2-preview1 优化 XncfBuilder AI 代码生成过程的模型选用
[2024-09-10] v0.5.6-preview1 升级缓存基础库
[2024-10-09] v0.5.8-preview1 升级基础库
[2024-10-09] v0.5.9-preview1 升级基础库
[2024-10-16] v0.5.10-preview1 添加 Service 对 PagedList 对象快速 Mapping 到 DTO 对象的方法
[2024-10-17] v0.5.11-preview1 完成能够全流程运行的基础版本
[2024-10-18] v0.5.12-preview1 优化功能
[2024-10-20] v0.5.13-preview1 升级 Senparc.Weixin.Work SDK
[2024-10-26] v0.5.15-preview1 发布 MVP 版本
[2024-10-26] v0.5.15.3-preview1 优化 Agents 输出
[2024-11-08] v0.6.2-preview1 升级 Senparc.AI
[2024-11-08] v0.6.3-preview1 升级 Senparc.AI
[2024-11-15] v0.7.0-preview1 使用新版本 AutoGen，升级 AgentsManager；解决 AgentsManager 中状态获取的异常
[2024-11-21] v0.8.2-preview1 更新 CO2NET 及缓存基础库
[2024-11-28] v0.8.3-preview1 update CO2NET, use new SenparcSetting with UseLowerCaseApiName property
[2024-12-27] v0.8.4-preview1 Add exception log for UseXncfModules() auto progress
[2025-01-09] v0.9.0-preview1 reconstruction XncfBuilder for generate XNCF
[2025-01-25] v0.9.3-preview1 Update Agents functions, support for function calling
[2025-01-25] v0.9.4-preview1 Update Agents functions, support for function calling
[2025-01-28] v0.9.5-preview1 update Senparc.AI
[2025-01-28] v0.9.6-preview1 update Senparc.AI
[2025-02-06] v0.9.7-preview1 update NeuCharAI model update
[2025-03-24] v0.9.7.12-preview1 update Dm support
[2025-04-29] v0.9.9-preview1 Update CO2NET and AIKernel
[2025-05-15] v0.9.10-preview1 Add MCP to AgentsManager module
[2025-05-18] v0.9.11-preview1 Update Installation
[2025-06-20] v0.9.13-preview Add MCP functions in XncfRegisterBase class
[2025-07-27] v0.9.14-preview.1 update Knowledge Base
[2025-09-10] v0.9.14.5-preview.1 update MCP tools inject progress
[2026-03-24] v0.9.18-preview.1 update Senparc.AI.Agents to v0.7.1
[2025-11-01] update basic libraries, including Senparc.AI
[2026-04-24] Simplify FunctionRender request parameters and keep SelectionList UI metadata compatibility
[2026-07-01] v0.11.0-preview2 Upgrade multi-agent group chat workflow and MAF integration
1、优化 ChatGroupService 的任务管理、并发处理与异常回退
2、改进 AgentsManager 页面历史记录展示与交互体验
[2026-07-02] v0.11.0-preview2 增强多智能体群聊流程、容错与任务调度能力，优化 PromptOptimizationAgentBridge 稳定性与状态恢复能力
1、增强多智能体群聊流程、容错与任务调度能力
2、优化 PromptOptimizationAgentBridge 稳定性与状态恢复能力
3、新增 ChatTask 归档状态（IsArchived）及归档管理能力
4、补充 MySql、SqlServer、Oracle、PostgreSQL、Sqlite、Dm 迁移与模型快照，保障归档字段落库
5、同步应用服务与流式控制器，完善任务归档流程与状态处理
6、Dependency update from Senparc.Xncf.XncfBuilder to 0.36.2-preview1
[2026-07-05] v0.11.1-preview3 重构系统配置初始化与更新流程并统一模型处理，新增批量启动任务接口并优化任务调度
1、重构系统配置初始化与更新流程并统一模型处理
2、新增批量启动任务接口并优化任务调度
[2026-07-05] v0.11.2-preview4 重构系统配置初始化与更新流程并统一模型处理，新增批量启动任务接口并优化任务调度
1、重构系统配置初始化与更新流程并统一模型处理
2、新增批量启动任务接口并优化任务调度
[2026-07-07] v0.11.3-preview5 Dependency updates from upstream modules
1、Dependency update from Senparc.Xncf.PromptRange to 0.16.5-preview4
2、升级 Senparc.AI 至 0.27.3、Senparc.AI.AgentKernel 至 0.1.10
[2026-07-17] v0.12.0-preview6 为 AgentsManager 模块接入统一资源本地化并优化功能文案
1、接入统一资源本地化机制，支持模块功能文案按当前文化动态显示
2、更新 FunctionRender 与请求模型文案的资源化解析
3、统一模块注册信息、菜单名称与功能描述的本地化输出
4、同步管理页面与前端交互文案的多语言显示
5、完善复制操作的兼容性回退与结果反馈，适配内嵌 WebView 场景
[2026-07-29] v0.12.1-preview7 加强 AgentsManager 接口授权与 Markdown/HTML 安全渲染
1、为 Prompt 初始化和优化接口增加 AdminOnly 授权保护
2、限制 Markdown 链接、图片和任务 HTML 的协议与危险标签，降低脚本注入风险
[2026-07-31] v0.13.0-preview8 扩展 AgentsManager 管理界面的多语言能力
1、新增管理页与前端交互资源，支持界面文案按当前文化显示
2、升级 Senparc.AI 组件与 CO2NET WebApi 兼容依赖
[2026-08-04] v0.14.0-preview9 新增 Agent 模板知识库关联与管理统计
1、为 AgentTemplate 增加 KnowledgeBaseId 并补齐六类数据库迁移与快照
2、扩展应用服务、DTO 与管理页面的知识库选择和统计展示
3、优化群聊服务的知识库集成与状态处理
[2026-08-08] v0.14.1-preview10 Dependency updates from upstream projects
1、Dependency update from Senparc.Xncf.PromptRange to 0.19.1-preview9
2、Dependency update from Senparc.Xncf.SenMapic to 0.4.2-preview6
3、Dependency update from Senparc.Xncf.XncfBuilder to 0.39.1-preview9
4、Dependency update from Senparc.Xncf.MCP to 0.4.4-preview7
5、Dependency update from Senparc.Xncf.KnowledgeBase to 0.5.1-preview7
6、Dependency update from Senparc.Xncf.AgentsManager.Abstractions to 0.2.4-preview4
7、Dependency update from Senparc.Xncf.PromptRange.Abstractions to 0.2.4-preview4
8、Dependency update from Senparc.Ncf.AreaBase to 0.23.2-preview6
[2026-08-13] v0.15.0-preview11 增强 A2A 智能体、ChatGroup 执行能力与管理界面
1、新增远程 A2A 智能体接入、发布管理及多数据库迁移
2、统一本地与 A2A AgentTemplate 执行、支持群聊启停和取消回退
3、优化 AgentsManager 统计、操作按钮、历史记录与响应式布局
[2026-08-13] v0.15.0-preview12 修复发布型 A2A 与本地 Agent 的模型执行一致性
1、发布型 A2A 复用 IWantToRun 严格流式入口，保留真实上游异常并支持取消请求
2、Azure / NeuCharAI 在发生真实 403 时仅回退一次 DeploymentName=ModelId，不切换 Endpoint、Key 或系统默认模型
3、依赖 Senparc.AI.AgentKernel 0.1.12，模型 SDK 统一使用 AgentKernel 配置的 HttpClient transport
[2026-08-13] v0.15.0-preview13 修复发布型 A2A 与本地 Agent 的请求模式差异
1、发布型 A2A 改用 IWantToRun 严格非流式响应入口，A2A 协议流与模型服务流解耦
2、保持原始上游异常、会话回退和 403 的受控 DeploymentName 兼容回退
3、执行诊断新增 modelRequest=strict-non-streaming，便于与本地 Agent 调用比对
[2026-08-13] v0.15.0-preview14 增强发布型 A2A 的 Azure-compatible 403 兼容与诊断
1、真实 403 后可按 AIModel 配置的 API Version 受控重试，不改变模型、Endpoint、Key 或系统默认模型
2、记录脱敏模型请求路由、API Version、鉴权形态和服务端 request-id，便于定位上游拒绝原因
[2026-08-13] v0.15.0-preview15 统一本地与发布型 Agent 的模型错误语义
1、本地工作流与发布型 A2A 共用严格响应入口，避免把上游 401/403 伪装为 Agent 正常文本
2、补充 AIModel API Version 受控替换回归测试，保证仅修改 api-version 查询参数
[2026-08-13] v0.15.0-preview16 修复 NeuCharAI 的 API Version 兼容回退边界
1、保留 AIModel 配置的有效 API Version，并在为空或等于 SDK 固定版本时尝试同 Endpoint 的 NeuChar 兼容默认版本
2、执行日志显示配置版本和回退来源，便于区分 AIModel 数据与 SDK 默认请求
[2026-08-13] v0.15.0-preview17 对齐发布 A2A 与本地 Agent 的模型传输边界
1、诊断传输遇到真实 403 后，使用完全相同模型配置通过 AgentKernel 标准本地传输重试，不切换系统默认模型
2、安全记录请求结构与上游标准错误字段，保留响应给 SDK 处理且不记录 Prompt、输入或凭证
[2026-08-14] v0.15.0-preview18 明确处理发布 A2A 的上游应用停用错误
1、识别 NeuChar AI 应用不可用或停用响应并停止无意义的传输、Deployment 与 API Version 重试
2、向 Workflow/A2A 返回脱敏且可操作的模型配置错误，并保留 DiagnosticId
[2026-08-14] v0.15.0-preview19 发布型 A2A 完整复用本地 Agent 执行配置
1、A2A 正常路径复用本地 RunAsync 与 AgentKernel 标准传输，不再替换 HttpClient、Deployment 或 API Version
2、沿用 Prompt 版本的 MaxToken、Temperature、TopP、Penalty 与 StopSequences，避免固定参数覆盖 DeepSeek 等模型配置
3、执行诊断补充参数来源和脱敏参数摘要，便于核对本地与 A2A 的实际执行配置
[2026-08-14] v0.15.0-preview20 修正 ChatGroup 远程 A2A 超时边界
1、防止 Aspire 默认 Polly 管道的 30 秒 AttemptTimeout 覆盖 RemoteAgent.TimeoutSeconds
2、远程 A2A 请求统一使用可配置的 5-600 秒整体超时，并避免对可能已开始执行的 POST 进行隐式重试
3、AgentTemplate 与 ChatGroup 执行入口传递取消令牌，完善手动 Prompt 与远程请求的可取消边界
4、补充知识库向量发布状态与管理界面反馈，并保持本地 Agent 与发布型 A2A 的执行配置一致
5、完善手动 Prompt 判定回归测试，避免自然语言被错误当作 PromptRange 版本查询
[2026-08-17] v0.16.0 新增 ChatTask Human-in-the-Loop 人工审批与人类参与者能力
1、为 ChatTask 增加 HIL 等级、最大对话轮次与工具权限策略，并补齐多数据库迁移
2、支持人工审批工具调用、Human 参与者发言，以及 ChatGroup 按配置自动纳入人类参与者
3、完善 ChatGroup/ChatTask 请求校验、显式 GroupId 传参与管理端交互反馈
4、AgentTemplate 支持绑定 AI 模型（AiModelId）并补齐多数据库迁移
5、独立 Agent 空输出时按最低 Token 预算自动重试一次，避免过低 MaxToken 导致误失败
6、优化管理端人类参与者配置、任务执行策略展示与 Agent 编辑入口
7、增加工作流引用校验和函数绑定序列化能力，优化 Agent 编辑、任务筛选及响应式管理界面
8、补充 Agent 群组会话查看、执行引用和任务状态处理，提升管理端交互一致性
[2026-08-29] v0.17.0 增强 Agent 请求诊断、ChatGroup 状态处理与工作流对象支持
1、新增模型请求结果与异常诊断摘要，脱敏记录敏感配置并保留可追踪信息
2、改进 ChatGroup Token、审批日志、状态更新和服务作用域隔离，减少无效流式通知
3、完善工作流对象解析及消息历史处理，提升多智能体执行兼容性
[2026-09-15] v0.17.1 适配 Agent Framework Harness 依赖并保持代理执行兼容
1、升级 AgentKernel 依赖并接入 Microsoft Agents AI Harness 包
2、保持代理工作流执行、工具交互与历史状态处理兼容
[2026-09-16] v0.17.2 升级 Senparc.Weixin.Work 依赖以保持企业微信兼容
1、升级 Senparc.Weixin.Work 至 3.32.4，保持代理管理模块的企业微信能力兼容
[2026-09-17] v0.18.0 增强 Agent FunctionRender 工具的 AI schema 元数据能力
1、为 AgentTemplate 注册的 FunctionRender 工具注入参数元数据与选择项 schema
2、升级共享 Senparc.AI.AgentKernel 至 0.1.14-preview5 并保持代理执行兼容
[2026-09-24] v0.18.1 支持从 MCP 列表选择 Endpoint（EventBus 跨模块查询）
1、新增 GetMcpEndpointOptions 接口，通过 EventBus 向 Senparc.Xncf.MCP 模块查询已登记端点
2、Agent 与群组配置提供“从 MCP 列表选择”抽屉，复用 Function 绑定选择样式，支持搜索与勾选
3、MCP 模块不可用或未响应时自动降级为手动输入并给出提示
[2026-10-05] v0.18.1 feat: add MCP endpoint selection feature in AgentsManager
1、Updated Senparc.Xncf.AgentsManager project behavior and integration for the latest feature and reliability improvements.
       [2026-10-09] v0.18.3 统一 Agent Studio 群组与提示配置并完善任务管理交互
       1、支持群组与 Prompt 配置联动、快速建组、增量成员管理和任务操作
       2、完善 Studio 页面渲染、Three.js 交互及前端状态管理