← All reports

Changes on 2026-08-26

34 total changes in 4 runs

22:48 EST

🤖 AI Batch Analysis

### 整体摘要 本次更新对应 Claude Code 2.1.247 版本,重点引入了 API 成本优化工具和反馈机制,并修复了多项影响终端体验(特别是 Zed 编辑器和非拉丁键盘布局)的 Bug。文档方面,大幅澄清了云端环境初始化、MCP 连接器认证以及组织级“快速 Web 设置”的具体行为逻辑。 ### 关键主题 * **云端入职流程规范化**:明确了“默认”云端环境在不同计划下的创建逻辑,并详细解释了 Team/Enterprise 计划中“快速 Web 设置”这一组织开关的具体作用(跳过 GitHub App 安装提示、自动创建环境)。 * **终端配置与体验修复**:重点修复了 `/terminal-setup` 在 Zed 编辑器中会覆盖现有 `keymap.json` 的严重问题(现为合并模式),并解决了非拉丁键盘布局下 Ctrl 快捷键失效及终端文本乱码问题。 * **新增功能与 API 工具**:引入了 `/claude-api cost-optimize` 命令用于分析项目成本,新增了反馈草稿工具 (`SendFeedback`),并扩展了 Admin API 的覆盖范围。 * **MCP 与连接器优化**:改进了针对无法支持本地 OAuth 的托管连接器(如 Microsoft 365, Gmail)的错误提示和处理流程,明确指引用户在 claude.ai 进行认证。 ### 影响级别:**中等** **理由**:虽然本次没有破坏性变更,但对 Zed 用户而言,键位配置逻辑的修正从“覆盖”变为“合并”属于重要的安全性/用户体验提升。新增的成本优化工具对高频 API 用户具有高实用价值。文档更新主要解决了用户对环境创建机制的困惑,降低了入门门槛。 ### 开发者行动项 * **检查 API 成本**:如果你关注项目支出,尝试运行 `/claude-api cost-optimize` 以获取缓存和模型选择的优化建议。 * **Zed 用户注意**:如果你之前因担心配置冲突而拒绝 `/terminal-setup`,现在可以安全使用,它只会合并键位绑定并自动备份原文件。 * **解决认证报错**:如果在 MCP 配置中遇到本地 OAuth 失败(特别是 Google/Microsoft 服务),请参考文档移除本地配置并转向 `claude.ai/customize/connectors` 进行连接。 * **回顾环境设置**:Team/Enterprise 管理员应查看“快速 Web 设置”开关,确认是否希望成员通过简化流程连接 GitHub 并自动生成默认环境。
💭 Model Reasoning
1. **Analyze the Request:** * **Role:** World-class programmer with deep expertise in developer tools and documentation. * **Task:** Analyze a batch of documentation changes (5 pages) and summarize changes and their relevance to developers. * **Input:** 5 specific diff pages (`claude-code-on-the-web.md`, `mcp.md`, `slack.md`, `terminal-config.md`, `CHANGELOG.md.md`). * **Output Format:** Concise batch analysis in markdown. * **Structure:** 1. Overall Summary (1-2 sentences) 2. Key Themes (bullets) 3. Impact Level (Low/Medium/High/Breaking) with reasoning 4. Action items for developers (bullets, max 5) * **Language:** Chinese. 2. **Analyze the Source Material (Page by Page):** * **`claude-code-on-the-web.md`**: * *Change 1:* Clarified how the "Default" cloud environment is created during onboarding (creates it OR asks you to create it, depending on the plan). * *Change 2:* Expanded definition of "Quick web setup". Explained it as an org setting (off by default for Team/Enterprise) that allows GitHub connection, skips app install prompts, and auto-creates the Default environment. * *Change 3:* Removed reference to installing the GitHub App during "setup" in the auto-fix note, keeping only the link to the GitHub App page. (Slight simplification). * **`mcp.md`**: * *Change:* Updated wording regarding Anthropic-hosted connectors (Microsoft 365, Gmail, etc.) that don't support local OAuth. Added specific error link `is Anthropic-hosted and doesn't support local OAuth`. Explicitly mentioned the workflow: `claude mcp add`/`login` -> Error -> Remove local entry -> Connect on claude.ai -> Auto-appearance in Claude Code. * **`slack.md`**: * *Change:* Updated error message for "Claude Code is not enabled for your account". Added specific reference to "web onboarding" and creating the default environment. * **`terminal-config.md`**: * *Change:* Added detailed explanation of how `/terminal-setup` handles Zed's `keymap.json` (backups, merging, error handling). It mentions a specific error scenario where it leaves the file unchanged. * **`CHANGELOG.md.md`**: * *Version 2.1.247*: This is the meat of the update. * *New features:* `SendFeedback` tool, cost optimization command `/claude-api cost-optimize`, updated Admin API coverage in `/claude-api`, new spinner tip overrides. * *Fixes:* Arrow-key+Enter navigation fix, sub-agent fallback model fix, GPU text garble fix (mentioned in terminal-config), Ctrl shortcuts in non-Latin layouts, Zed keymap overwriting fix (matches terminal-config), symlink deletion fix in Bash sandbox. * *Improvements:* Plugin marketplace hardening, better error reporting for MCP connection failures on Bedrock/Vertex. * *Changes:* Sonnet 5 auto-compact window change, cross-session message collapsing, terminal hyperlink security hardening. 3. **Synthesize Key Themes:** * **Cloud & Onboarding Clarity:** Heavy focus on how the "Default" cloud environment works and how "Quick web setup" functions across plans (Team/Enterprise vs. others). * **Configuration & Setup Fixes:** specific improvements to how terminal setup works, especially for Zed (merging instead of overwriting) and non-Latin keyboard layouts. * **New Features:** Feedback tools and API cost optimization. * **MCP & Connectors:** Better handling of OAuth flows and connection errors for hosted services. 4. **Determine Impact Level:** * *Reasoning:* Most changes are informational clarifications or bug fixes (Zed keymap, keyboard layouts). The new features (`cost-optimize`, feedback) are additive, not breaking. The documentation updates regarding onboarding are useful but don't change functionality, just explain it better. * *Verdict:* **Medium**. While nothing is breaking, the fixes for Zed (prevent data loss/overwriting) and keyboard shortcuts are significant for affected users. The new cost-optimize tool is high value. 5. **Draft Action Items (for developers):** * Review the new `/claude-api cost-optimize` command if you are worried about API costs. * If using Zed, be aware that `/terminal-setup` now merges keymaps instead of overwriting (safer). * If you encountered the "Claude Code is not enabled" error in Slack, follow the updated onboarding steps. * If using non-Latin keyboards, check out the fixed Ctrl shortcuts. 6. **Translate and Format into Chinese (Target Language):** * *Summary:* 本次更新主要包含 Claude Code 2.1.247 版本的新功能(如成本优化)和多项修复,同时大幅改进了云端环境搭建、MCP 连接器以及终端配置(特别是 Zed)的文档清晰度。 * *Themes:* * 云端环境与入职流程:明确了“默认”环境的创建逻辑及“快速 Web 设置”的组织策略。 * 终端与配置修复:修复了 Zed keymap 覆写问题、非拉丁键盘布局快捷键失效等 Bug。 * 功能增强:新增 API 成本优化工具和反馈草稿功能。 * *Impact:* Medium. (解释:主要是功能增强和 Bug 修复,特别是 Zed 配置的安全性改进对相关用户很重要)。 * *Action Items:* * 尝试使用 `/claude-api cost-optimize` 优化 API 成本。 * 如果你是 Zed 用户,了解 `/terminal-setup` 现在会合并而非覆盖键位映射。 * 查阅更新后的云端环境文档,以了解不同计划下默认环境的创建差异。 7. **Review against constraints:** * Single batch analysis? Yes. * Markdown format? Yes. * Chinese response? Yes. * Concise? Yes. 8. **Final Polish (Self-Correction during drafting):** * Make sure to link the Zed change in terminal-config.md to the CHANGELOG entry for completeness. * Ensure the distinction between "Quick web setup" as an *organization setting* is clear in the summary. 9. **Final Output Generation** (Matches the provided good response).

📚 Claude Code

📚 Claude Code GitHub

17:34 EST

🤖 AI Batch Analysis

# 文档变更分析 ## 总体摘要 本次更新主要增强了企业级网络配置的文档说明,特别是针对 IP 白名单和代理出口的指导,同时优化了插件系统和平台功能(特别是 Microsoft Foundry)的兼容性描述。 ## 关键主题 - **企业网络配置**:新增 `bridge.claudeusercontent.com` 域名的 IP 白名单和代理路由要求,解决 Chrome 扩展连接问题 - **插件系统改进 (v2.1.246)**:修正技能计数逻辑、插件名称解析规则,并改进 UTF-8 BOM 处理 - **平台功能对齐**:更新 Microsoft Foundry 上 Azure 托管部署的功能支持状态,明确结构化输出、Web 搜索、Web fetch 等功能的可用性 - **数据使用透明化**:新增 Claude 起草反馈机制的工作流程说明 ## 影响级别:中等 - **理由**:虽然主要是文档澄清和错误排查增强,但新增的企业网络配置要求对组织环境下的开发者有实质影响;插件命名规则的变更可能影响现有插件代码。 ## 开发者行动项 1. **检查网络配置**:如果使用组织 IP 白名单,确保 `bridge.claudeusercontent.com` 通过与 `claude.ai` 相同的代理出口路由,或将其专用出口 IP 加入白名单 2. **审查插件 manifest**:确保 `plugin.json` 使用标准 UTF-8 编码(无 BOM),避免 JSON 解析错误 3. **更新插件命名规范**:技能的 `name` 字段若已包含插件前缀,无需重复添加(v2.1.246 已修复双重前缀问题) 4. **了解平台限制**:在 Microsoft Foundry 上使用 Azure 托管部署时,注意仅支持基础版 Web 搜索/获取工具(`web_search_20250305`、`web_fetch_20250910`) 5. **利用反馈新机制**:了解 Claude 起草反馈的本地审核流程,掌握数据传输的控制权
💭 Model Reasoning
用户希望我分析一批文档更改,并用中文提供简洁的见解。我需要以世界级程序员的身份,专注于对开发者重要的内容。 让我先快速浏览所有更改: 1. **chrome.md**: - 更新了错误排查表格 - "Browser extension is not connected" 错误增加了新的可能原因:组织 IP 白名单拒绝连接到 `bridge.claudeusercontent.com` - 增加了对 IP 白名单和代理出口的文档链接 2. **data-usage.md**: - 新增了关于 Claude 起草反馈的信息 - 更新了 /feedback 命令的数据传输说明,包括 Claude 起草反馈的提交路径 3. **desktop.md**: - 新增了关于组织 IP 白名单和代理出口的配置说明 - 解释了 `bridge.claudeusercontent.com` 的路由要求和可能的连接问题 4. **discover-plugins.md**: - 更新了插件重载时技能计数的说明 - 修正了技能计数的描述,现在包括 `commands/` 和 `SKILL.md` 技能 - 添加了版本行为变更说明 (v2.1.246) 5. **network-config.md**: - 新增了"组织 IP 白名单和代理出口"部分 - 解释了 `bridge.claudeusercontent.com` 的路由要求 - 与 desktop.md 的内容类似 6. **plugins.md**: - 移除了关于技能计数只覆盖 `commands/` 目录的说明 7. **plugins-reference.md**: - 新增了关于裸插件名称解析的说明 - 更新了 JSON 语法错误的解释,包括 UTF-8 BOM 问题 插件名称解析和 JSON 语法错误细节是重要的技术更新。v2.1.246 版本引入了更严格的插件名称解析机制,同时改善了错误处理能力。这些变更旨在提高插件系统的健壮性和一致性,帮助开发者更精确地管理和使用插件。 结构化输出现已全面支持 Microsoft Foundry,不再需要特殊部署。代理工具在 Microsoft Foundry 的托管模式下有了新的功能限制,只支持基础版本。技能命名规则也进行了调整,确保插件命名更加规范和一致。 关键主题显示,这些变更主要集中在企业网络配置、插件系统改进和平台功能支持上,旨在提升整体用户体验和系统集成能力。开发者需要特别注意命名规范和平台兼容性。 变更属于中等影响,主要是文档更新和错误排查指南,同时澄清了一些平台限制。开发者应仔细审查网络配置和插件命名细节。
14:53 EST

🤖 AI Batch Analysis

### 总体摘要 本次更新主要细化了钩子机制,明确区分了“工具使用权限”与“沙箱网络请求权限”的处理逻辑,并新增了 7 个 API Beta 功能头部,扩展了结构化输出和计算机使用等能力。 ### 核心变更点 * **钩子机制精细化 (Hooks & Hooks Guide)** * 明确 `PermissionRequest` 钩子**不会**针对沙箱命令的 [网络请求](/docs/en/sandboxing#network-isolation) 触发。 * 若需监控沙箱网络请求,必须改用 `Notification` 钩子中的 `permission_prompt` 类型。 * 在终端会话中,支持 `permission_prompt` 监控沙箱网络请求需要 Claude Code v2.1.246 或更高版本。 * **API 能力扩展 (Models List & Files Upload)** * 新增 7 个 Beta API headers:`compact-2026-01-12`、`computer-use-2025-11-24`、`mcp-tunnels-2026-06-22`、`structured-outputs-2025-11-13`、`task-budgets-2026-03-13`、`thinking-display-updates-2026-08-18` 和 `ce-user-management-2026-07-13`。 * **网络故障处理描述 (Network Config)** * 细化了看门狗中止流后的处理逻辑,明确区分了重试、报错结束或正常结束的情况,并补充了非交互会话中的提示行为。 * **文档示例修正 (Skills & Slash Commands)** * 将“组件框架迁移”的示例描述和参数更改为“语言迁移”(如从 JavaScript 到 TypeScript),以体现通用性。 ### 影响等级 **中等** **理由**:虽然主要是文档层面的澄清,但 `PermissionRequest` 钩子行为的明确界定意味着依赖此钩子进行全权限自动化控制的开发者需要调整代码逻辑,以覆盖沙箱网络请求场景。同时,新增的 API Beta 头部为开发者提供了新的实验性功能入口。 ### 开发者行动项 1. **检查钩子逻辑**:如果您使用 `PermissionRequest` 钩子来自动批准操作,请确认是否需要补充 `permission_prompt` 监听来处理沙箱内的网络请求。 2. **验证版本兼容性**:确保终端环境升级至 v2.1.246+,以支持最新的沙箱网络请求钩子功能。 3. **评估新 API 功能**:查阅新增的 Beta headers(特别是 `structured-outputs` 和 `computer-use`),评估其能否优化现有的 API 集成。 4. **更新错误处理代码**:参考文档中关于网络超时/中止的详细描述,确保客户端的流式传输错误处理逻辑覆盖所有可能的恢复场景。
💭 Model Reasoning
1. **Understand the Goal:** The user wants a concise, insightful analysis of documentation changes for developers, provided in Chinese. The analysis needs to cover a batch of 7 pages with specific diffs. 2. **Analyze the Diffs Page by Page:** * **Page 1: `hooks.md`** * *Change 1 (PermissionRequest):* Added "to use a tool" to clarify when the hook runs. Added a new paragraph: "Claude Code doesn't run PermissionRequest hooks for a sandboxed command's network request...". It tells developers to use `permission_prompt` notification type instead for that specific case. * *Change 2 (Notification table):* Updated `permission_prompt` description to include "or a sandboxed command's network request". * *Change 3 (Version note):* Added note requiring v2.1.246 or later for `permission_prompt` regarding sandboxed network requests in terminal sessions. * *Change 4 (Note block):* Refined wording about `permission_prompt` vs `PermissionRequest` hook. Clarified that `PermissionRequest` is for "tool" permissions specifically. * *Impact:* Clarifies behavior regarding hooks and sandboxing. Developers need to know that `PermissionRequest` *won't* fire for network requests in sandboxes. * **Page 2: `hooks-guide.md`** * *Change 1 (Notification table):* Same as Page 1 (added network request to description). * *Change 2 (Version note):* Same as Page 1 (v2.1.246 requirement). * *Change 3 (Example context):* Changed "every permission prompt" to "every tool permission prompt" to be precise. * *Impact:* Consistent with Page 1. * **Page 3: `network-config.md`** * *Change 1 (Watchdog behavior):* Refined the description of what happens when a watchdog aborts a stalled stream. Changed "handles it like any other mid-stream failure" to a more detailed explanation: retries, ends turn with error, or ends turn normally. Added mention of non-interactive sessions prompting Claude to continue. * *Impact:* Clarifies error handling and retry logic for network timeouts. * **Page 4 & 5: `skills.md` & `slash-commands.md`** * *Change 1 (Example update):* In the example code, changed "Migrate a component from one framework to another" to "from one language to another". * *Change 2 (Example values):* Changed example arguments from "React", "Vue" to "JavaScript", "TypeScript". * *Impact:* Minor documentation example update to reflect general-purpose usage rather than just frameworks. Not a functional change to the tool. * **Page 6 & 7: `api/models/list.md` & `api/beta/files/upload.md`** * *Change 1 (API Beta headers):* Expanded the list of supported Beta headers. Count went from "31 more" to "38 more". * *New Headers Added:* `"compact-2026-01-12"`, `"computer-use-2025-11-24"`, `"mcp-tunnels-2026-06-22"`, `"structured-outputs-2025-11-13"`, `"task-budgets-2026-03-13"`, `"thinking-display-updates-2026-08-18"`, `"ce-user-management-2026-07-13"`. * *Impact:* New capabilities/betas are available via API headers. This is significant for API users. 3. **Synthesize and Structure the Analysis:** * **Overall Summary:** The update primarily refines hook behavior regarding sandbox network requests and adds new API beta capabilities. * **Key Themes:** * **Hook Granularity:** Distinguishing between "tool" permissions and "sandbox network" permissions in hooks. * **Network Reliability:** Clarifying timeout/watchdog behavior. * **New API Features:** Introduction of several new beta headers (computer use, structured outputs, etc.). * **Impact Level:** **Medium**. While mostly documentation refinements, the new API beta headers represent new capabilities, and the hook changes might break existing logic if developers assumed `PermissionRequest` covered sandbox network calls. * **Action Items:** * Update hook logic if you rely on `PermissionRequest` for sandbox network access (switch to `Notification`). * Check version compatibility (v2.1.246+) for the specific hook behavior. * Explore new API beta headers for relevant features (structured outputs, computer use, etc.). * Review error handling logic if building custom network integrations. 4. **Drafting the Content (in Chinese):** * *Summary:* 本次更新主要细化了钩子机制中关于“工具权限”与“沙箱网络请求”的区别,并引入了多项新的 API Beta 功能。 * *Themes:* * **钩子机制精细化:** 明确 `PermissionRequest` 钩子不处理沙箱命令的网络请求,需改用 `Notification` 钩子。 * **网络超时处理:** 详细阐述了看门狗机制中止流时的行为,包括重试逻辑。 * **API 能力扩展:** 新增了 7 个 Beta API 头部,涉及结构化输出、计算机使用和任务预算等功能。 * *Impact:* **Medium**. 主要涉及行为澄清和功能扩展。依赖旧钩子行为监控沙箱网络的开发者可能需要调整代码。 * *Action Items:* * 检查并更新依赖 `PermissionRequest` 钩子的代码,确保沙箱网络请求监控逻辑正确。 * 确认 Claude Code 版本 >= v2.1.246 以支持新的钩子行为。 * 调研新增的 API Beta headers(如 `structured-outputs`, `computer-use`)是否适用于当前项目。 * 更新技能/命令示例文档中的描述(无需代码变更)。 5. **Refining and Polishing:** Ensure the tone is professional ("world-class programmer"). The "skills.md" and "slash-commands.md" changes are trivial examples, so I'll downplay them in the summary but acknowledge them as minor updates. The API headers are the most "feature-rich" change. * *Self-Correction during drafting:* Make sure to emphasize *why* the hook change matters—it prevents developers from wondering why their auto-approve logic isn't working for sandboxed network calls. 6. **Final Output Generation** (matching the markdown format requested). * *Overall Summary:* 本次文档更新重点阐明了 `PermissionRequest` 钩子在沙箱网络请求上的行为差异,并引入了多项新的 API Beta 能力。 * *Themes:* 1. **钩子行为修正**: 明确 `PermissionRequest` 不再触发沙箱命令的网络请求权限,需改用 `permission_prompt` 通知类型。 2. **网络故障处理细化**: 重新描述了看门狗中止流后的具体处理流程(重试、错误或正常结束)。 3. **API 功能扩展**: 新增 7 个 API Beta headers,支持结构化输出、任务预算等新特性。 4. **示例更新**: 将代码示例中的“框架迁移”描述调整为更通用的“语言迁移”。 * *Impact:* **Medium**. 新增 API 功能带来了扩展机会,但钩子机制的澄清意味着部分现有的自动化权限逻辑可能需要调整以覆盖沙箱网络请求。 * *Action Items:* * **适配钩子逻辑**: 如果您的代码依赖 `PermissionRequest` 自动批准权限,请注意它不再涵盖沙箱网络请求,需补充 `permission_prompt` 监听。 * **版本检查**: 确保环境升级至 Claude Code v2.1.246+ 以支持最新的沙箱网络请求钩子行为。 * **探索新特性**: 审查新增的 API Beta headers(如 `structured-outputs-2025-11-13` 和 `computer-use-2025-11-24`),评估是否可用于增强功能。 * **网络错误处理**: 更新网络超时错误处理逻辑,以匹配文档中关于重试和提示的详细描述。 7. **Final Review against constraints:** * Markdown format? Yes. * Chinese reply? Yes. * Concise? Yes. * Developer focus? Yes. (Proceed to output)

📚 Claude Code

📚 Anthropic API

12:13 EST

🤖 AI Batch Analysis

### 文档变更分析 #### 1. 总体概述 本次文档更新主要提升了 Google Cloud (Vertex AI) 平台对 Claude 高级工具的支持级别,正式启用了 `Computer use` 和 `Browser use` 工具。此外,文档中包含少量的文本大小写规范化修正。 #### 2. 核心变更主题 * **Google Cloud (Vertex AI) 功能增强**:Google Cloud 平台现已正式支持最新的 `Computer use` 工具集 (`computer_toolset_20260801`) 和 `Browser use` 工具,不再仅限于旧版 Beta 工具。 * **平台兼容性更新**:在多个文档中更新了工具的可用性矩阵,明确将 Vertex AI 列为支持 Computer use 和 Browser use 的平台。 * **限制解除**:移除了此前关于 Google Cloud 暂不支持 `computer_toolset_20260801` 和 `browser_toolset_20260801` 的说明。 #### 3. 影响级别 **Medium (中等)** **理由**:虽然这不是破坏性 API 变更,但对于在 Google Cloud (Vertex AI) 上构建应用的开发者来说,这是一个显著的解锁功能。开发者现在可以访问标准的 Computer use 和 Browser use 工具集,而无需依赖即将淘汰的 Beta 版本。 #### 4. 开发者行动项 * **迁移工具集**:如果你正在 Vertex AI 上使用旧版 Beta 的 Computer use 工具 (`computer_20251124`),请计划迁移至新的 `computer_toolset_20260801`。 * **启用新功能**:Vertex AI 用户现在可以集成 `Browser use` 工具来处理网页浏览任务。 * **检查代码**:如果你在代码中硬编码了文档链接或文本匹配,请注意 "Playground" 在部分文档中已改为小写 "playground"。
💭 Model Reasoning
1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** 6 pages of diffs (Markdown format). * **Output Format:** Markdown, concise, insightful, Chinese language. * **Structure:** 1. Overall Summary. 2. Key Themes. 3. Impact Level with reasoning. 4. Action Items for developers (max 5). 2. **Analyze the Diffs:** * **`intro.md`:** * Change: "Playground" -> "playground" (lowercase). * Type: Minor text styling/grammar correction. * **`api/overview.md`:** * Change: "Playground" -> "playground" (lowercase). * Type: Minor text styling/grammar correction. * **`build-with-claude/overview.md` & `build-with-claude/text-generation.md`:** * Change: Table formatting adjustment (column width/layout implied by diff whitespace, though content is mostly identical). * *Wait, look closer at `build-with-claude/overview.md`*: * `Browser use` availability: Changed from `<PlatformAvailability claudeApi />` to `<PlatformAvailability claudeApi vertexAi />`. * `Computer use` availability: Changed from `<PlatformAvailability claudeApi claudePlatformAwsBeta bedrockBeta vertexAiBeta azureAiBeta />` to `<PlatformAvailability claudeApi claudePlatformAwsBeta bedrockBeta vertexAi azureAiBeta />`. Note: `vertexAiBeta` -> `vertexAi`. * *Wait, look closer at `build-with-claude/text-generation.md`*: Same changes as above regarding availability tags for Browser use and Computer use on Vertex AI. * **`build-with-claude/claude-on-vertex-ai.md`:** * Change 1: Added `[Browser use tool]` to the list of supported tools in the "Supported features" section. * Change 2: Removed a bullet point under "Not currently supported" section that specifically listed Computer use and Browser use toolsets (`computer_toolset_20260801` and `browser_toolset_20260801`) as not currently available on Google Cloud. * *Implication:* Google Cloud (Vertex AI) now fully supports Computer use (new toolset) and Browser use. It's no longer in the "not supported" list (or the specific restriction about the new toolsets is gone). * **`agents-and-tools/tool-use/computer-use-tool.md`:** * Change 1: Platforms list: "Google Cloud (beta)" -> "Google Cloud". (Moved out of beta?). * Change 2: Clarification: "Platforms other than the Claude API currently offer only the earlier beta tool versions" -> "Platforms other than the Claude API and Google Cloud currently offer only the earlier beta tool versions". * Change 3: Note section: Added "[Google Cloud](...)" as a platform where the `computer_toolset_20260801` toolset is available (no beta header required). 3. **Synthesize Findings:** * **Theme 1: Google Cloud (Vertex AI) Updates.** This is the biggest technical change. * `Browser use` is now supported on Vertex AI. * `Computer use` (specifically the `computer_toolset_20260801` toolset) is now officially supported on Vertex AI (seemingly out of beta or at least fully integrated). * The limitation note in the Vertex AI specific doc regarding these toolsets has been removed. * **Theme 2: Minor Text Standardization.** * "Playground" is consistently lowercase in these changes. 4. **Draft the Response (Internal Monologue/Drafting):** * *Overall Summary:* The primary update is the formalization of support for the new `Computer use` and `Browser use` tools on Google Cloud's Vertex AI platform, moving them out of the beta-only toolset restriction. There are also minor typo corrections ("Playground" -> "playground"). * *Key Themes:* * Google Cloud (Vertex AI) integration enhancement. * Feature availability expansion (Browser use & Computer use). * Minor documentation styling consistency. * *Impact Level:* Medium. While the code didn't break, the availability status changed significantly for Vertex AI developers. They can now use the standard `computer_toolset_20260801` without specific beta headers (or restrictions apply differently). * *Action Items:* * Check if you are integrating with Vertex AI: You can now use the standard `computer_toolset_20260801` toolset. * Review `Browser use` availability if you are on Vertex AI. * (Optional) Ignore the "Playground" casing changes unless you are doing strict string matching on docs. 5. **Refine into Chinese:** * *Summary:* 本次更新主要提升了 Google Cloud (Vertex AI) 平台对 Claude 工具的支持,特别是正式启用了 `Computer use` 和 `Browser use` 工具。同时包含少量的文本格式修正。 * *Themes:* * Google Cloud (Vertex AI) 功能扩展:正式支持最新的 Computer use 和 Browser use 工具集。 * Beta 状态调整:Google Cloud 对 Computer use 的支持不再仅限于旧版 Beta 工具。 * 文本规范化:统一 "Playground" 的首字母大小写。 * *Impact:* Medium. 因为这直接影响 Google Cloud 用户可以使用哪些工具版本,移除了之前的限制。 * *Action Items:* * Vertex AI 用户:现在可以使用标准的 `computer_toolset_20260801` 工具集,而无需使用旧版 Beta 工具。 * Vertex AI 用户:确认 `Browser use` 现已在平台上可用。 * 如果你在文档中进行字符串匹配,请注意 "playground" 现为小写。 6. **Final Polish:** Ensure Markdown format is clean and professional. *(Self-Correction during drafting)*: Make sure to emphasize the specific toolset name `computer_toolset_20260801` because that is technically specific. The removal of the limitation note in `claude-on-vertex-ai.md` confirms the removal of the blocker. 7. **Final Output Generation** (matches the structure requested).