@michengai/dsh-agency-agents 0.1.44 → 1.0.1

package/CHANGELOG.md

@@ -2,7 +2,88 @@
 
 [简体中文](CHANGELOG.zh-CN.md)
 
-The five most recent published versions are listed below.
+This file records features and upgrade boundaries. npm and GitHub Releases remain the publication source of truth.
+
+## 1.0.1 - 2026-09-21
+
+This patch is prepared for release; publication is confirmed only by npm and GitHub Releases.
+
+- Replace translated-message comparisons with stable persona error codes. Fall back to the English persona only when the translated file is absent, not when reading it fails.
+- Localize updater errors and request rejections, remove duplicate package file entries, and validate README version consistency before release.
+- Verify language switching through the real DSH tool registry and model-facing prompt assembly, including the locally installed 0.1.6-alpha.2 runtime. This does not claim live model quality or historical-host matrix acceptance.
+
+- Bundle only the referenced Tabler icons through the ESM entry and move the icon package to development dependencies. Add release checks for client size, accidental full-library bundling, and external icon imports.
+- Preserve raw expert enablement records when saving or enabling a team, including temporarily conflicting or unavailable experts. Keep revision checks and atomic writes.
+- Preserve separators while typing tags; validate tags and custom coordinator prompts with actionable field errors and focus. Keep unsaved tag input through language changes and conflict review.
+- Localize team service, delegation guard, and expert prompt errors. Refresh registered team tool descriptions and parameter descriptions with the host language.
+- Remove deleted task IDs from failed dispatch reports while retaining IDs when cleanup fails. Guard team mention registration on older remote services and restore expert conflict tooltips.
+- Remove unused editor section parameters, unused styles, and duplicate team style injection. Strengthen asynchronous, member-limit, locale, and browser regression tests; exercise the real update dialog and derive fixture counts from the roster.
+- Refresh both README screenshot galleries with the expert list, team list, team details, and native Agent Team panel.
+
+## 1.0.0 - 2026-09-21
+
+The first 1.0 release expands individual specialists into configurable expert-team collaboration. It retains the package name, 321 bundled specialists, 22 divisions, and existing individual-expert workflows. The following covers the changes since 0.1.44.
+
+### Retained individual-expert capabilities and tools
+
+- Retain 321 bundled experts across 22 divisions, prompt viewing/copying, category/status/localized keyword filters, enablement, and selection through the composer or @.
+- Retain up to 200 custom experts with creation, copying, avatars and summon emoji, editing, save-and-enable, disabling, and permanent deletion; preserve name-conflict and concurrent-revision checks.
+- Keep list_experts, summon_expert, and summon_experts; add list_expert_teams, get_expert_team, and summon_expert_team for reading team rules before delegation and final synthesis.
+- Preserve the settings version display, update checking/feedback, and trusted external root, provider, divisions, and maxDepth configuration. Do not add an External source tab.
+### Five built-in teams and custom teams
+
+- Add Product Review Team, Technical Review Team, Content Planning Team, Data Analysis Team, and Research Team. Each includes three independent expert roles, a shared goal, delivery requirements, task examples, and a dedicated coordinator template.
+- Create custom teams, copy built-in teams, inspect details, edit, enable, disable, and delete. Each team contains 2–8 distinct experts; up to 100 custom teams are supported.
+- Configure a name, description, up to three tags, member duties and instructions, shared goal, constraints, delivery requirements, and 1–3 examples. Custom coordinator prompts allow up to 12,000 characters.
+- Use the team coordinator template, customize it, or restore it. Rules cover preparation, delegation, evidence checks, disagreements, failures, and final delivery. The existing main conversation coordinates; no extra leader subagent is created.
+- Confirm and enable required experts when enabling a team. Disabling or deleting a team does not disable its members or delete historical conversations.
+
+### Unified expert and team management
+
+- Add an underlined Expert teams tab inside Settings → Experts, sharing the title, version, and links. Preserve individual experts' All/Built-in/Custom source filters, categories, status, and search.
+- Preserve expert filters across tab switches. Place counts at the right of the tabs, with source filters and create/refresh actions on one row. Teams gain equivalent source/status filters, search, empty states, refresh, and more menus.
+- Share cards, editor drawers, save footers, prompt previews, and confirmation dialogs. Confirm deletion and discarding unsaved changes. Retain drafts on revision conflicts, review current settings before saving, and allow deleted experts to continue as new experts.
+- Remove redundant source badges and expand all five team descriptions. Team descriptions grow with their content to avoid truncating English, while card actions remain compact.
+- Align buttons, inputs, dropdowns, and focus states with Archive Manager using compact sizing and official theme backgrounds/borders. Enabled switches are green, disabled switches gray, with text labels; support dark/light themes and narrow layouts.
+
+### Chat selection, examples, and draft protection
+
+- Switch to teams in the Experts composer picker, search, inspect details, enable, and select. The @ picker supports native team references with short localized names.
+- Selecting a team or example updates the draft without sending it. Protect existing text, attachments, and references; confirm team replacement or example insertion in a themed dialog.
+- Recheck the draft revision after asynchronous confirmation. If the draft changed meanwhile, stop insertion and ask the user to select again rather than overwriting new input.
+- Preserve full-roster individual-expert search, enable-and-select, bilingual examples for eight representative specialists, conflict feedback, and retry behavior.
+
+### Ordinary subagents and native Agent Team
+
+- Check host capability separately from service enablement and tool availability in the current Agent scope, without modifying host settings.
+- Use ordinary subagents when unsupported. Recommend enabling Team when supported but disabled, while allowing ordinary execution. Use native Agent Team when services and tools are ready. Explicit maxDepth continues through ordinary mode to enforce depth limits.
+- Ordinary mode performs one parallel analysis with up to four concurrent experts. Freeze team, member, and task settings before execution so later edits cannot change the running assignment.
+- Native mode integrates members, shared tasks, task claiming, result messages, and asynchronous waiting. Reuse idle or offline members for the same team/expert; do not redispatch busy or failed members.
+- Accepted native dispatch means started, not completed. The main conversation waits for actual results for the current tasks before verifying and summarizing. Mid-dispatch failures do not automatically switch to ordinary mode and duplicate work.
+- Localize new child-session expert names in both modes. Adapt native roster display names while retaining stable identities, message routing, and reuse. Do not rewrite old ordinary-session titles or host model labels.
+
+### Collaboration quality and failure handling
+
+- Provide team-specific briefs, role boundaries, structured member reports, coordinator checklists, and handoff rules. Consolidate duplicate findings while preserving evidence, disagreements, and open questions.
+- Ordinary mode reports responsibility coverage and member success/failure, preserving completed results when others fail. Isolate persona-read failures, reject empty results, and do not automatically add model rounds or retries.
+- Cancellation prevents queued work from starting, settles in-flight launches, and interrupts native members started by this invocation. Report cleanup failures explicitly.
+- Preserve child-tool isolation and recursion guards; experts cannot summon more experts or start another team round.
+
+### Complete localization, documentation, and release assets
+
+- Complete Chinese/English UI, details, editors, confirmations, errors, tool descriptions, and runtime feedback, plus all five teams' names, descriptions, tags, duties, examples, and coordinator rules.
+- Language changes preserve custom content and unsaved drafts. Only system presets are localized; both built-in Chinese and English names participate in collision checks, and stable IDs remain unchanged.
+- Regenerate separate Chinese and English 3:1 WebP banners without version numbers. Replace the Node README badge with supported DSH versions, and refresh bilingual feature and upgrade documentation.
+- Set the version baseline to 1.0.0. Generate bilingual release notes from the matching changelog entries to keep feature lists aligned.
+
+- Stop tracking compiled `lib` in Git. The release gate builds and transfers verified artifacts to npm publishing; npm packages retain the runtime `lib` files.
+
+### Upgrade, compatibility, and validation limits
+
+- Retain support for DSH 0.1.0-rc.8, 0.1.1-rc.2, 0.1.2-rc.1, 0.1.5-rc.1, 0.1.5-rc.2, 0.1.6-alpha.1, and 0.1.6-alpha.2. This does not claim coverage of every intervening or future version. Native Team also depends on installation and enablement.
+- Existing expert enablement and custom experts remain in the host agency-agents settings namespace; team settings use that namespace too. No expert-file migration is required. Back up settings before downgrading: old versions lack teams and may not preserve team fields on writes.
+- The engineering regression baseline passes 199 unit/host-integration tests and 40 browser tests, including opt-in visual validation. Fix the old visual fixture that passed despite missing theme variables. Type checks, builds, and package validation pass.
+- Browser tests use real components with mocked Remote services, and model providers are test doubles. These results do not validate real-model summary quality or every supported host end to end. Native model labels remain host-owned and may show creation-time models.
 
 ## 0.1.44 - 2026-09-18
 

package/CHANGELOG.zh-CN.md

@@ -2,7 +2,88 @@
 
 [English](CHANGELOG.md)
 
-以下记录最近发布的五个版本。
+本文件记录版本功能与升级边界。正式发布状态以 npm 和 GitHub Releases 为准。
+
+## 1.0.1 - 2026-09-21
+
+本补丁已准备发行，实际发布状态以 npm 和 GitHub Releases 为准。
+
+- 提示词回退改为稳定错误码，仅译文文件不存在时回退英文；读取失败不再被误判为缺少译文。
+- 补齐更新器错误及请求拒绝的国际化，去重包文件清单，发布检查新增 README 版本一致性校验。
+- 使用真实 DSH 工具注册器和面向模型的提示组装验证语言切换，并通过本机安装的 0.1.6-alpha.2 运行时验证；不代表真实模型质量或历史宿主矩阵验收。
+
+- 图标改用 ESM 入口，仅打包实际引用的图标；图标包移至开发依赖。发布检查新增客户端体积、整库误打包及图标运行时外部引用护栏。
+- 保存并启用或启用专家团时，保留原始专家启用记录，包括暂时冲突或不可用的专家；继续使用修订号校验与原子写入。
+- 标签输入保留分隔符，保存时解析；标签和自定义主理人提示词超限时明确指出字段并聚焦。切换语言及冲突核对保留标签草稿。
+- 补齐团队服务、递归委派门禁及专家提示词错误的国际化；已注册团队工具及参数说明跟随宿主语言变化。
+- 失败委派报告移除已成功删除的任务编号，清理失败时保留编号供排查；旧版远程服务不注册不可用的团队候选源，恢复专家名称冲突悬停提示。
+- 清理编辑分区死参数、无用样式及重复注入；完善异步、成员上限、国际化和浏览器回归测试，使用真实更新弹窗并从名册推导测试数量。
+- 中英文 README 更新专家列表、专家团列表、团队详情及原生 Agent Team 面板截图。
+
+## 1.0.0 - 2026-09-21
+
+从 0.1.44 升级到首个 1.0 正式版本，将单专家助手扩展为可配置的专家团协作。保留原安装包名、321 位基础专家、22 个分区和既有个人专家功能；以下为本轮完整变更。
+
+### 保留的单专家能力与工具接口
+
+- 321 位内置专家、22 个分区；查看与复制专家提示词，按分类、状态与中英文关键词筛选，启停后通过聊天工具栏或 @ 选择。
+- 最多 200 位自定义专家，支持新建、复制、头像与召唤 Emoji、编辑、保存并启用、停用和永久删除；名称冲突与多窗口修订检查继续生效。
+- 保留 list_experts、summon_expert、summon_experts；新增 list_expert_teams、get_expert_team、summon_expert_team，支持先读取团队规则再调用与汇总。
+- 保留设置页版本展示、检查更新与更新反馈，以及可信外部 root、provider、divisions、maxDepth 配置。界面不新增“外部”来源页签。
+### 五个内置专家团与自定义团队
+
+- 新增产品方案评审团、技术方案评审团、内容选题策划团、数据分析诊断团和专题研究专家团。每团提供三位成员的独立职责、共同目标、交付要求、任务示例和专用主理人模板。
+- 支持创建自定义团、复制内置团后修改、查看详情、编辑、启用、停用和删除；每团 2～8 位不重复专家，最多 100 个自定义团。
+- 可配置团队名称、简介、最多三个标签、成员分工与补充指令、共同目标、约束、交付要求和 1～3 条示例；主理人正文最多 12000 字符。
+- 主理人可使用团队模板、自定义提示词或恢复模板；规则涵盖准备、分工、证据核验、分歧处理、异常说明与最终交付。主理人由当前主会话承担，不额外创建团长子代理。
+- 启用团队时确认并同步启用所需专家；停用或删除团队不连带停用成员，也不删除历史会话。
+
+### 专家与专家团统一管理
+
+- 在原「设置 → 专家」内新增下划线「专家团」页签，共用标题、版本及链接；保留专家的全部／内置／自定义来源选项、分类、状态和搜索。
+- 切换页签保留专家筛选状态；数量显示在页签右侧，来源选项与新建、刷新同行。团队支持相同的来源与状态筛选、搜索、空态、刷新和更多菜单。
+- 复用卡片、编辑抽屉、保存底栏、提示词预览及确认弹窗；删除和未保存离开均需确认。多窗口修订冲突保留草稿，核对最新配置后再保存，已删除专家可作为新专家继续。
+- 卡片去掉重复来源标识，扩充五团用途描述；团队简介按内容展开，避免英文截断，底部操作保持紧凑。
+- 对照归档插件统一按钮、输入、下拉菜单和焦点状态：紧凑控件、统一圆角、官方主题背景与边框。启用开关为绿色，停用为灰色，保留文字状态，适配深浅主题与窄屏。
+
+### 聊天选择、示例与草稿保护
+
+- 聊天「专家」入口支持切换专家团，搜索、查看详情、启用并选择；@ 支持专家团原生引用，显示当前语言的简短名称。
+- 选择团队或示例只更新草稿，不自动发送。原有正文、附件及引用受到保护；替换团队或追加示例通过主题弹窗确认。
+- 异步确认后核对草稿修订；期间输入发生变化则停止插入并提示重新选择，避免覆盖新内容。
+- 保留个人专家全名册搜索、未启用专家的启用选择、八位代表性专家的双语任务示例、冲突提示和失败重试行为。
+
+### 普通子代理与原生 Agent Team
+
+- 分别检测宿主是否支持 Agent Team、功能与当前 Agent 作用域工具是否可用；不修改用户的宿主配置。
+- 不支持时使用普通子代理；支持但未启用时建议开启并继续普通调用；服务和工具就绪后使用原生 Agent Team。显式配置 maxDepth 时继续普通模式以维持深度约束。
+- 普通模式一次并行分析，最多四位专家同时运行；执行前冻结团队、成员与任务配置，防止运行中编辑改变分工。
+- 原生模式接入成员、共享任务板、任务领取、消息回传和异步等待；空闲或离线的同团同专家可复用，忙碌或失败成员不重复派发。
+- 原生委派接受仅表示任务启动；主会话等待本次任务的真实成员结果后核验汇总，不把启动确认当作专家意见。中途失败不自动切换普通模式，避免重复执行。
+- 两种模式的新子会话使用本地化专家名。原生成员卡片适配显示名，同时保留稳定技术标识、消息路由与队友复用；不改写旧普通会话标题或宿主模型标签。
+
+### 协作质量与异常处理
+
+- 五团分别使用任务简报、职责边界、结构化成员回报、主理人核验清单和交接规则，合并重复发现并呈现证据、分歧与待确认事项。
+- 普通模式返回职责覆盖与成员成功／失败信息，部分失败保留已完成结果；身份读取异常单独报告，空白结果不视为成功，不自动追加模型轮次或重试。
+- 取消后不启动排队任务，等待在途启动收敛并中断本次原生成员；清理失败明确反馈。
+- 保留子会话工具隔离和递归门禁，专家不继续召唤专家或组建新一轮团队。
+
+### 完整双语、文档与发布材料
+
+- 补齐团队列表、详情、编辑、确认、错误、工具说明、运行提示，以及五团名称、简介、标签、成员职责、示例和主理人规则的中英文。
+- 切换语言保留自定义内容与未保存草稿；仅系统预设参与本地化，内置团中英文名称均参与冲突校验，稳定 ID 不变。
+- 重制中文与英文 3:1 WebP 横幅，不带版本号；移除 README 的 Node 徽章，改为 DSH 支持版本徽章。重写双语功能说明与升级边界。
+- 版本基线提升至 1.0.0，双语发行说明由对应更新日志自动生成，避免功能清单不同步。
+
+- 编译产物 `lib` 不再提交到 GitHub，由发布门禁构建并作为已验证产物交给 npm 发布任务；npm 包保留运行所需的 `lib`。
+
+### 升级、兼容与验证边界
+
+- 仍支持 DSH 0.1.0-rc.8、0.1.1-rc.2、0.1.2-rc.1、0.1.5-rc.1、0.1.5-rc.2、0.1.6-alpha.1 和 0.1.6-alpha.2；不表示其间所有版本或未来版本均已验证。原生 Team 还取决于实际安装与启用状态。
+- 原有专家启用状态与自定义专家保存在宿主 agency-agents 设置命名空间；新团队配置使用同一命名空间，无需搬迁专家文件。回退旧插件前备份设置，旧版不具备团队功能，不能保证旧版写入保留团队字段。
+- 工程回归基线为 199 项单元／宿主集成测试和 40 项浏览器测试通过，包含按需视觉验收；修正旧视觉夹具缺少主题却仍通过的问题。类型检查、构建与包验证通过。
+- 浏览器使用真实组件与模拟 Remote，模型 provider 使用替身；不等于真实模型汇总质量或全部宿主版本的端到端验收。原生 Team 的模型标签由宿主提供，可能仍显示创建时的模型。
 
 ## 0.1.44 - 2026-09-18
 

package/README.md

@@ -1,21 +1,21 @@
 <div align="center">
-  <img src="assets/branding/banner.png" alt="DSH Agency Agents" width="100%">
+  <img src="assets/branding/dsh-banner-en.webp" alt="DSH Agency Agents" width="100%">
 </div>
 
 <div align="center">
 
   # DSH Agency Agents
 
-  **321 summonable specialist agents for DeepSeek Harness**
+  **321 specialists · 5 built-in teams · Custom collaboration**
 
-  [简体中文](README.zh-CN.md) · [Expert roster](#expert-roster) · [Installation](#installation) · [Changelog](CHANGELOG.md) · [Apache-2.0](LICENSE)
+  [简体中文](README.zh-CN.md) · [Expert roster](#expert-roster) · [Installation](#installation) · [Changelog](CHANGELOG.md) · [Release notes](RELEASE_NOTES.md) · [Apache-2.0](LICENSE)
 
   [![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](LICENSE)
   [![Bundled agents](https://img.shields.io/badge/Bundled%20agents-321-0f766e.svg)](#expert-roster)
   [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
   [![npm downloads](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=npm%20downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
   [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
-  [![Node.js 22 or later](https://img.shields.io/badge/Node.js-22%20or%20later-339933.svg?logo=node.js&logoColor=white)](https://nodejs.org/)
+  [![DSH supported through 0.1.6-alpha.2](https://img.shields.io/badge/DSH-up%20to%200.1.6--alpha.2-2563eb.svg)](#prerequisites)
 </div>
 
 > DSH Agency Agents is a community-maintained DeepSeek Harness (DSH) plugin, not an official DeepSeek AI product.
@@ -29,19 +29,66 @@ Choose a specialist in DSH for code review, design, operations, or research. All
 - **Describe the task directly**: use `@` or the **Experts** picker, then write your complete request.
 - **Get one final delivery**: experts provide specialist analysis and the parent conversation brings the results together. Experts cannot summon further experts.
 
+## Expert teams
+
+Manage teams in **Settings → Experts → Expert teams**, sharing the header, source/status filters, search, and card interactions with individual experts.
+
+| Built-in team | Focus | Delivery |
+| --- | --- | --- |
+| Product Review Team | User value, experience, technical feasibility | Priorities and actions |
+| Technical Review Team | Architecture, security, acceptance boundaries | Risks and acceptance checks |
+| Content Planning Team | Angles, distribution strategy, fact checking | Topic options and content structure |
+| Data Analysis Team | Metrics, data quality, business interpretation | Findings and validation steps |
+| Research Team | Research scope, evidence, synthesis | Source-backed conclusions |
+
+### Configure and use
+
+1. Use a built-in team, copy it to customize, or create a team of 2–8 distinct experts. Up to 100 custom teams can be saved.
+2. Set the name, description, tags, member duties, shared goal, constraints, delivery requirements, and 1–3 examples. Use, customize, or restore the coordinator template.
+3. Confirm enabling required experts when enabling a team. Switch to Expert teams in the composer picker or select through `@`, then write the task. Selecting or adding an example never sends the message.
+4. The current main conversation delegates, verifies evidence, handles disagreements, and delivers one result; no additional leader subagent is created. Ordinary mode runs up to four experts concurrently and preserves completed results if others fail.
+
+### Automatic collaboration mode
+
+| Host state | Behavior |
+| --- | --- |
+| Agent Team unsupported | Use ordinary subagents without recommending an unavailable feature. |
+| Supported but disabled, or current-session tools unavailable | Recommend enabling Team while continuing to allow ordinary execution. |
+| Services and current-session tools available | Use native Agent Team, shared tasks, and teammate messages; wait for actual results before summarizing. |
+
+Explicit `maxDepth` uses ordinary mode to preserve depth limits. Native dispatch acceptance is not expert completion; partial startup failures do not automatically switch modes and duplicate work. Members inherit the main conversation's model selection; there is no per-expert model setting. Native model labels are provided by DSH and may display creation-time values.
+
+### Tool interfaces
+
+Retain `list_experts`, `summon_expert`, and `summon_experts`. Add `list_expert_teams`, `get_expert_team`, and `summon_expert_team`. The main conversation reads the enabled team's rules and duties before delegating the complete task, then verifies and summarizes actual member results.
+
+### Drafts, language, and data
+
+- System presets, member names, coordinator templates, UI, and runtime feedback support Chinese and English. Custom content remains unchanged, and language switches preserve unsaved drafts.
+- Confirm team replacement or example insertion. If the draft changes during confirmation, stop insertion and retain the new text, attachments, and references.
+- Team configuration lives in the host `agency-agents` settings namespace. Keep drafts on edit conflicts and review the latest configuration. Disabling or deleting teams does not disable experts or delete historical conversations.
+- Experts cannot summon further experts or automatically add analysis rounds. Cancellation prevents queued starts; completed results and failures are reported separately.
+
+> This README describes version 1.0.1; npm and GitHub Releases determine publication status. See the [changelog](CHANGELOG.md) for the full changes and [release notes](RELEASE_NOTES.md) for the bilingual release description. Upgrades retain expert settings. Back up settings before downgrading because older versions do not support teams.
 ## Screenshots
 
+The screenshots below show the 1.0.0 interface in Chinese. Experts and expert teams share the same settings page.
+
 Filter by category or search in **Settings → Experts**, then enable the experts you need:
 
 ![DSH Experts panel](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/agent-roster.png)
 
-Use `@` or the composer's **Experts** picker to choose an enabled expert:
+Switch to **Expert teams** to browse the five built-in teams, filter by status, or create a custom team:
 
-![Experts picker](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/expert-picker.png)
+![Expert teams panel](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/expert-teams.png)
 
-The localized expert name is inserted as a short tag; write the complete task next:
+Open team details to review its purpose, example tasks, and member responsibilities, then summon it or copy it for customization:
 
-![Summoning an expert from the composer](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/summon-prompt.png)
+![Expert team details](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/expert-team-details.png)
+
+When the host supports Agent Team and the feature is enabled, view team members and shared tasks in the native panel:
+
+![Native Agent Team panel](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/agent-team-panel.png)
 
 ## DSH product ecosystem
 
@@ -65,7 +112,7 @@ For a desktop workbench, download [DSH Codex Desktop](https://github.com/Micheng
 
 - A working DeepSeek Harness Web installation with `dsh` available in PowerShell.
 - Examples use the `web` profile; replace it with the target profile.
-- Source installation and development require Node.js 22+ and pnpm. npm installation does not require running `pnpm install` separately.
+- Source installation and development require Node.js ^22.19.0 or ≥24.0.0 and pnpm. npm installation does not require running `pnpm install` separately.
 
 ## Host dependencies and compatibility
 
@@ -96,7 +143,7 @@ dsh plugin --profile web add @michengai/dsh-agency-agents@latest --registry=http
 dsh --profile web --dump-config
 ```
 
-To pin a release, replace `@latest` with a version such as `@0.1.17`.
+To pin a release, replace `@latest` with a version such as `@1.0.1`.
 
 The configuration output should contain `agency-agents` and `agency-agents-remote`. Restart DSH Web and hard-refresh the browser. Do not copy client files manually: the Settings page needs the mounted Remote service.
 
@@ -705,6 +752,9 @@ Set `AGENCY_AGENTS_ROOT` to use an external expert directory. Persona bodies fro
 
 ## Secondary development
 
+`lib` is not tracked in Git. Run `pnpm build` before using a source checkout. The release gate builds and verifies `lib`; npm packages still include the compiled runtime.
+
+
 ### Install from source
 
 Use this for debugging or unpublished changes. The cloned directory becomes the plugin source path:
@@ -762,7 +812,7 @@ pnpm exec playwright install chromium
 pnpm test:browser
 ```
 
-Coverage includes asynchronous loading, focus restoration after Escape, close-button and backdrop dismissal, StrictMode, and successive triggers. These component tests do not replace real-host or model-delegation acceptance. Browser tests bundle source rather than loading the published `lib/client.js`; artifacts still require build, package checks, and real-host acceptance. Override the default port 18769 with the `AGENCY_BROWSER_TEST_PORT` environment variable if needed.
+Coverage includes expert/team filtering, creation, copying, editing, enablement, deletion, draft protection, revision conflicts, localization, dark/light themes, narrow layouts, and focus restoration under normal rendering and StrictMode. These component tests do not replace real-host or model-delegation acceptance. Browser tests bundle source rather than loading the published `lib/client.js`; artifacts still require build, package checks, and real-host acceptance. Override the default port 18769 with the `AGENCY_BROWSER_TEST_PORT` environment variable if needed.
 
 ## Custom experts
 

package/README.zh-CN.md

@@ -1,21 +1,21 @@
 <div align="center">
-  <img src="assets/branding/banner.png" alt="DSH Agency Agents" width="100%">
+  <img src="assets/branding/dsh-banner.webp" alt="DSH Agency Agents" width="100%">
 </div>
 
 <div align="center">
 
   # DSH Agency Agents
 
-  **为 DeepSeek Harness 提供 321 名可召唤的专业智能体**
+  **321 位专业智能体 · 5 个内置专家团 · 自定义协作**
 
-  [English](README.md) · [专家列表](#专家列表) · [安装](#安装) · [更新日志](CHANGELOG.zh-CN.md) · [Apache-2.0](LICENSE)
+  [English](README.md) · [专家列表](#专家列表) · [安装](#安装) · [更新日志](CHANGELOG.zh-CN.md) · [发行说明](RELEASE_NOTES.md) · [Apache-2.0](LICENSE)
 
   [![许可证：Apache-2.0](https://img.shields.io/badge/许可证-Apache--2.0-blue.svg)](LICENSE)
   [![内置智能体](https://img.shields.io/badge/内置智能体-321-0f766e.svg)](#专家列表)
   [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
   [![npm 下载量](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=npm%20%E4%B8%8B%E8%BD%BD%E9%87%8F)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
   [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
-  [![Node.js 22 or later](https://img.shields.io/badge/Node.js-22%20or%20later-339933.svg?logo=node.js&logoColor=white)](https://nodejs.org/)
+  [![DSH 支持至 0.1.6-alpha.2](https://img.shields.io/badge/DSH-up%20to%200.1.6--alpha.2-2563eb.svg)](#前置条件)
 </div>
 
 > DSH Agency Agents 是社区维护的 DeepSeek Harness（DSH）插件，并非 DeepSeek AI 官方产品。
@@ -29,19 +29,66 @@
 - **直接交代任务**：在输入框用 `@` 或「专家」选择角色，再写下完整需求。
 - **由主会话统一交付**：专家提供专业分析，主会话汇总结果；专家不会继续召唤其他专家。
 
+## 专家团
+
+在「设置 → 专家 → 专家团」管理团队，与专家页共用头部、来源筛选、状态筛选、搜索和卡片交互。五个内置团覆盖以下场景：
+
+| 专家团 | 分析重点 | 交付内容 |
+| --- | --- | --- |
+| 产品方案评审团 | 需求价值、用户体验、技术可行性 | 优先级与行动清单 |
+| 技术方案评审团 | 架构、安全、验收边界 | 风险与验收清单 |
+| 内容选题策划团 | 选题角度、传播策略、事实核验 | 选题方案与内容结构 |
+| 数据分析诊断团 | 指标、数据质量、业务解释 | 分析结论与验证步骤 |
+| 专题研究专家团 | 研究范围、证据、综合判断 | 有来源的研究结论 |
+
+### 配置与使用
+
+1. 直接使用内置团，或复制后修改；也可新建 2～8 位成员的自定义团，最多保存 100 个。
+2. 设置名称、简介、标签、成员职责、共同目标、约束、交付要求和 1～3 条任务示例。主理人提示词可使用团队模板、自定义或恢复模板。
+3. 启用团队时确认同步启用所需专家。聊天工具栏切到「专家团」，或通过 `@` 选择团队，再填写任务；选择与示例均不会自动发送。
+4. 当前主会话分工、核验证据、处理分歧并统一交付，不额外创建团长。普通模式最多四位专家并发；部分失败会保留已完成结果。
+
+### 自动选择协作模式
+
+| 宿主状态 | 行为 |
+| --- | --- |
+| 不支持 Agent Team | 使用普通子代理，不提示开启不存在的能力。 |
+| 支持但未启用，或当前会话工具未就绪 | 建议开启，仍可继续普通调用。 |
+| 服务和当前会话工具均可用 | 使用原生 Agent Team、共享任务板和队友消息，等待实际结果后汇总。 |
+
+显式配置 `maxDepth` 时使用普通模式以保留深度约束。原生模式任务启动确认不等于专家已经完成；创建中途失败不会自动切换模式重复执行。队友默认沿用主会话的模型选择，不提供独立专家模型配置。原生面板的模型标签由 DSH 提供，可能显示创建时的值。
+
+### 工具接口
+
+保留个人专家工具 `list_experts`、`summon_expert`、`summon_experts`；新增 `list_expert_teams`、`get_expert_team`、`summon_expert_team`。主会话应先读取已启用团队的规则与分工，再提交完整任务，最后核验并汇总成员结果。
+
+### 草稿、语言与数据
+
+- 系统预设、成员名称、主理人模板、界面和运行反馈支持中英文；自定义内容保持原文，切换语言不覆盖未保存草稿。
+- 替换团队或追加示例需确认；确认期间草稿变化会停止插入，保留用户新输入、附件和引用。
+- 团队配置保存在宿主 `agency-agents` 命名空间。编辑冲突保留草稿并提示核对最新配置；停用或删除团队不会停用专家，也不会删除历史会话。
+- 专家不得继续召唤专家或自动追加分析轮次。取消会停止未启动任务；已有结果和失败原因分别报告。
+
+> 本 README 描述 1.0.1，正式发行以 npm 和 GitHub Releases 为准。完整改动见[更新日志](CHANGELOG.zh-CN.md)，双语版本说明见[发行说明](RELEASE_NOTES.md)。升级保留已有专家配置；回退旧版前先备份设置，旧版不具备团队功能。
 ## 界面预览
 
+以下为 1.0.0 中文界面截图，专家与专家团共用同一个设置页面。
+
 在「设置 → 专家」中按分类筛选或搜索，再启用需要的专家：
 
 ![DSH 专家面板](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/agent-roster.png)
 
-在输入框用 `@` 或「专家」选择已启用的专家：
+切换到「专家团」，浏览五个内置专家团、按状态筛选，或新建自定义专家团：
 
-![专家选择器](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/expert-picker.png)
+![专家团面板](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/expert-teams.png)
 
-回填当前语言的专家名称标签后，写出完整任务：
+打开团队详情，查看适用场景、任务示例和成员分工，再召唤团队或复制并自定义：
 
-![召唤专家的输入方式](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/summon-prompt.png)
+![专家团详情](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/expert-team-details.png)
+
+宿主支持且已启用 Agent Team 时，可在原生面板查看团队成员和共享任务：
+
+![原生 Agent Team 面板](https://raw.githubusercontent.com/MichengAI/dsh-agency-agents/main/assets/screenshots/agent-team-panel.png)
 
 ## DSH 产品生态
 
@@ -65,7 +112,7 @@
 
 - 已可正常运行 DeepSeek Harness Web，且可在 PowerShell 中使用 `dsh`。
 - 以下示例使用 `web` profile；请替换为实际目标 profile。
-- 从源码安装或二次开发需要 Node.js 22+ 与 pnpm；仅从 npm 安装无需单独执行 `pnpm install`。
+- 从源码安装或二次开发需要 Node.js ^22.19.0 或 ≥24.0.0 与 pnpm；仅从 npm 安装无需单独执行 `pnpm install`。
 
 ## 宿主依赖与兼容性
 
@@ -96,7 +143,7 @@ dsh plugin --profile web add @michengai/dsh-agency-agents@latest --registry=http
 dsh --profile web --dump-config
 ```
 
-需要钉死某一版时，把 `@latest` 换成具体版本，例如 `@0.1.17`。
+需要钉死某一版时，把 `@latest` 换成具体版本，例如 `@1.0.1`。
 
 配置输出中应包含 `agency-agents` 与 `agency-agents-remote`。安装后重启 DSH Web 并在浏览器硬刷新；请勿手工复制客户端文件，否则设置页所需的 Remote 服务不会被挂载。
 
@@ -705,6 +752,9 @@ dsh --profile web --dump-config
 
 ## 二次开发
 
+`lib` 不提交到 GitHub；从源码使用前运行 `pnpm build`。发布门禁生成并验证 `lib`，npm 包仍包含可直接运行的编译产物。
+
+
 ### 从源码安装
 
 适用于调试或使用未发布改动。克隆后的本地路径就是插件安装路径：
@@ -762,7 +812,7 @@ pnpm exec playwright install chromium
 pnpm test:browser
 ```
 
-覆盖异步加载后通过 Escape、关闭按钮和遮罩退出时的焦点恢复，以及 StrictMode 和连续打开不同专家。该测试验证组件交互，不替代真实宿主和模型委派验收。 浏览器测试从源码构建，不加载发布的 `lib/client.js`；产物仍需构建、包校验及真实宿主验收。端口默认 18769，占用时可通过 `AGENCY_BROWSER_TEST_PORT` 环境变量覆盖。
+覆盖专家与团队的筛选、创建、复制、编辑、启停、删除、草稿保护、配置冲突、双语切换、深浅主题、窄屏以及 Escape、关闭按钮和遮罩的焦点恢复。该测试验证组件交互，不替代真实宿主和模型委派验收。 浏览器测试从源码构建，不加载发布的 `lib/client.js`；产物仍需构建、包校验及真实宿主验收。端口默认 18769，占用时可通过 `AGENCY_BROWSER_TEST_PORT` 环境变量覆盖。
 
 ## 自定义专家
 

package/RELEASE_NOTES.md

@@ -0,0 +1,33 @@
+## 中文说明
+
+本补丁已准备发行，实际发布状态以 npm 和 GitHub Releases 为准。
+
+- 提示词回退改为稳定错误码，仅译文文件不存在时回退英文；读取失败不再被误判为缺少译文。
+- 补齐更新器错误及请求拒绝的国际化，去重包文件清单，发布检查新增 README 版本一致性校验。
+- 使用真实 DSH 工具注册器和面向模型的提示组装验证语言切换，并通过本机安装的 0.1.6-alpha.2 运行时验证；不代表真实模型质量或历史宿主矩阵验收。
+
+- 图标改用 ESM 入口，仅打包实际引用的图标；图标包移至开发依赖。发布检查新增客户端体积、整库误打包及图标运行时外部引用护栏。
+- 保存并启用或启用专家团时，保留原始专家启用记录，包括暂时冲突或不可用的专家；继续使用修订号校验与原子写入。
+- 标签输入保留分隔符，保存时解析；标签和自定义主理人提示词超限时明确指出字段并聚焦。切换语言及冲突核对保留标签草稿。
+- 补齐团队服务、递归委派门禁及专家提示词错误的国际化；已注册团队工具及参数说明跟随宿主语言变化。
+- 失败委派报告移除已成功删除的任务编号，清理失败时保留编号供排查；旧版远程服务不注册不可用的团队候选源，恢复专家名称冲突悬停提示。
+- 清理编辑分区死参数、无用样式及重复注入；完善异步、成员上限、国际化和浏览器回归测试，使用真实更新弹窗并从名册推导测试数量。
+- 中英文 README 更新专家列表、专家团列表、团队详情及原生 Agent Team 面板截图。
+
+---
+
+## English
+
+This patch is prepared for release; publication is confirmed only by npm and GitHub Releases.
+
+- Replace translated-message comparisons with stable persona error codes. Fall back to the English persona only when the translated file is absent, not when reading it fails.
+- Localize updater errors and request rejections, remove duplicate package file entries, and validate README version consistency before release.
+- Verify language switching through the real DSH tool registry and model-facing prompt assembly, including the locally installed 0.1.6-alpha.2 runtime. This does not claim live model quality or historical-host matrix acceptance.
+
+- Bundle only the referenced Tabler icons through the ESM entry and move the icon package to development dependencies. Add release checks for client size, accidental full-library bundling, and external icon imports.
+- Preserve raw expert enablement records when saving or enabling a team, including temporarily conflicting or unavailable experts. Keep revision checks and atomic writes.
+- Preserve separators while typing tags; validate tags and custom coordinator prompts with actionable field errors and focus. Keep unsaved tag input through language changes and conflict review.
+- Localize team service, delegation guard, and expert prompt errors. Refresh registered team tool descriptions and parameter descriptions with the host language.
+- Remove deleted task IDs from failed dispatch reports while retaining IDs when cleanup fails. Guard team mention registration on older remote services and restore expert conflict tooltips.
+- Remove unused editor section parameters, unused styles, and duplicate team style injection. Strengthen asynchronous, member-limit, locale, and browser regression tests; exercise the real update dialog and derive fixture counts from the roster.
+- Refresh both README screenshot galleries with the expert list, team list, team details, and native Agent Team panel.

package/assets/branding/banner.txt

@@ -1,4 +1,5 @@
-README 顶部使用 assets/branding/banner.png
-尺寸：1983x793
-说明：完整产品横幅，含标识、卖点、功能入口与专家列表界面
-旧版 dsh-logo.png 与 ASCII 字标已不再作为 README 顶部展示
+中文 README 使用 dsh-banner.webp，英文 README 使用 dsh-banner-en.webp，全宽展示。
+两张横幅均不包含版本号，尺寸 2172×724，比例 3:1；通过 custom API 参考归档插件横幅生成，实际返回尺寸沿用参考图。
+采用 WebP quality=90、method=6，保留生成图原始分辨率。
+原图及提示词位于本地 docs 的设计资源目录，不随 npm 包发布。
+旧 banner.png、banner.svg 和 dsh-logo.png 作为历史资产保留；README 不再引用旧版横幅，发布清单使用新的双语 WebP。