← All reports

Changes on 2026-09-22

101 total changes in 5 runs

21:10 EST

🤖 AI Batch Analysis

### 文档变更分析 #### 1. Overall Summary 此次更新纯属文档格式修复,移除了代码块标记中冗余重复的 `theme={null}` 属性,确保文档渲染和语法高亮正常,但所有实际安装命令保持不变。 #### 2. Key Themes * **Markdown 语法规范化**:清理了代码块(`bash`, `powershell`, `batch`)开头多余的重复属性定义。 * **文档渲染优化**:修复了因语法错误可能导致代码块显示异常或无法正确高亮的问题。 * **维护性更新**:提升文档源码的可读性与维护性,不涉及产品功能变更。 #### 3. Impact Level **Low** 尽管涉及了“概览”和“快速开始”两个核心页面,但修改仅限于文档元数据格式,未触及任何实际的安装命令、脚本逻辑或产品行为。 #### 4. Action Items for Developers * 无需采取任何行动。
💭 Model Reasoning
1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** Diff for two pages (`overview.md`, `quickstart.md`). * **Output format:** Markdown (concise, insightful). * **Language:** Chinese. * **Structure:** 1. Overall Summary (1-2 sentences). 2. Key Themes (bullets). 3. Impact Level (Low/Medium/High/Breaking) + Reasoning. 4. Action Items for Developers (max 5 bullets). 2. **Analyze the Diffs:** * **Page 1: `overview.md`** * Change: In multiple code blocks (bash, powershell, batch), the attribute `theme={null}` is repeated excessively (7 times) in the "before" state. In the "after" state, it appears only once. * Context: Installation instructions (Native Install, Homebrew, WinGet). * Code content: `curl ...`, `irm ...`, `winget ...` (no changes to the actual commands). * **Page 2: `quickstart.md`** * Change: Identical to `overview.md`. Excessive `theme={null}` attributes removed from code block markers. * Context: Installation instructions. * Code content: Same commands, no changes. 3. **Synthesize the Analysis:** * **What happened?** A formatting cleanup. The Markdown code fence attributes were malformed or duplicated. Specifically, `theme={null}` was repeated 7 times in the opening fence (` ```bash theme={null} theme={null} ...`). This was fixed to just ` ```bash theme={null}`. * **Why does it matter to developers?** It improves the rendering of the documentation. The code blocks likely weren't highlighting correctly or were visually cluttered/parsed incorrectly in some viewers due to the invalid syntax duplication. The *actual* commands to install the software remain unchanged. * **Themes:** Markdown syntax cleanup, formatting standardization, UI/UX improvement for documentation. * **Impact Level:** Low. The functionality described (installation commands) hasn't changed. This is purely a documentation syntax fix. * **Action Items:** None really, since the commands didn't change. Maybe just "re-read the docs if you were confused by the formatting before," but that's weak. It's better to state "No action required." 4. **Draft the Content (in Chinese):** * *Summary:* 修复了 `overview.md` 和 `quickstart.md` 中代码块的 Markdown 语法错误(移除了重复的 `theme={null}` 属性),改善了文档的渲染效果,但安装命令本身未作更改。 * *Key Themes:* * 代码块语法清理 * 文档渲染优化 * 无功能性变更 * *Impact Level:* Low. 只涉及文档源码格式,不涉及产品功能或安装逻辑。 * *Action Items:* * 无需采取任何行动。 5. **Refine the Language (Professional "World-class Programmer" Tone):** * Make it sharper. * "Summary: 此次更新纯属文档格式修复,移除了代码块标记中多余的重复属性,确保语法高亮和渲染正常,但所有安装命令保持不变。" * "Key Themes:" * Markdown 语法标准化 * 代码块渲染修复 * 维护性更新 * "Impact Level: Low。仅修正文档语法,未修改任何安装命令或 CLI 行为。" * "Action Items:" * 无需更新本地环境。 * 无需修改脚本。 6. **Final Review against Constraints:** * Markdown format? Yes. * Chinese response? Yes. * Concise? Yes. * Action items max 5? Yes. 7. **Final Output Generation** (Proceed to generate output based on step 5).

📚 Claude Code

18:38 EST

🤖 AI Batch Analysis

### 1. Overall Summary `output-styles.md` 文档经历了重大重构和扩充,将原本简短的定义扩展为包含详细行为描述、内置风格对比表以及与其他功能(如 Hooks 和 Memory)的清晰界限。此次更新旨在帮助开发者更精准地理解和选择适合会话的 Claude 响应模式。 ### 2. Key Themes * **结构化文档增强**:新增了详细的“内置输出风格”对比表格,将 Default、Proactive、Concise、Explanatory 和 Learning 风格的具体用途和变化点结构化展示。 * **功能边界明确化**:通过新增的 `<Note>` 区块,明确区分了“输出风格”(指令性建议)、`CLAUDE.md`(项目知识)和 `hooks`(强制执行机制)的适用场景,解决了以往功能混用的问题。 * **行为细节澄清**:详细补充了特定模式下的行为逻辑,例如 `Concise` 模式虽然简化回复,但仍保留错误报告和安全警告的完整内容;`Proactive` 模式不改变底层的权限模式。 * **默认定义标准化**:明确了 `Default` 风格的含义(即未应用特定风格,仅使用标准系统提示词),并指出它可以通过命令显式选择。 ### 3. Impact Level **Low** **理由**:此次变更仅涉及文档层面的澄清和结构优化,未涉及 API 变更、命令行参数修改或产品功能的行为变更。它旨在消除歧义,帮助开发者更有效地利用现有功能。 ### 4. Action items for developers * 查阅新增的 **内置输出风格对比表**,快速识别哪种风格最适合你当前的工作流。 * 重新评估 **Concise** 模式的使用预期,注意关键的安全警告和错误信息仍会完整显示。 * 区分 **Output Styles** 和 **Hooks** 的用途:如果你需要“每次必须执行某动作”(如格式化),请考虑使用 Hooks 而非 Output Styles。
💭 Model Reasoning
1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Format:** Markdown. * **Language:** Chinese (as requested: "请用中文回复"). * **Input:** A diff of `output-styles.md`. * **Requirements:** Overall Summary, Key Themes, Impact Level, Action Items. 2. **Analyze the Diff (`output-styles.md`):** * *What changed?* * **Intro/Lead-in:** Moved from a definition-focused text to a more action-oriented summary. Added a "Note" section clarifying that output styles are *instructions*, not guarantees, and differentiating them from `CLAUDE.md` and hooks. Added a comparison table for built-in styles. * **Structure:** Expanded the descriptions of built-in styles (Default, Proactive, Concise, Explanatory, Learning) into dedicated sections with detailed headings. * **Clarifications:** * **Default:** Clarified that "Default" means no specific style is applied, though it's selectable via command. * **Proactive:** Clarified that it doesn't change permission modes (permissions still act as configured). Emphasized that it makes assumptions but checks on destructive actions (which is a separate instruction mechanism). * **Concise:** Detailed what "Concise" means (no preamble/narration) but listed exceptions where full text is preserved (errors, security warnings, destructive action confirmations). Explicitly mentioned version requirement (v2.1.237+). * **Explanatory:** Elaborated on the `Insight` blocks and noted they aren't written as file comments. * **Removed/Changed:** Removed the distinction about "keeping built-in instructions" vs "leaving them out" in the intro (it seems that nuance moved or was replaced by the new "Note" about how it differs from other features). * **Note on Distinction:** A significant addition is the `<Note>` block distinguishing output styles from `CLAUDE.md` (project knowledge), `hooks` (enforcing actions), and other features. 3. **Synthesize the Analysis:** * **Overall Summary:** The documentation for "Output Styles" was significantly expanded and restructured. It now provides clearer definitions, a comparison table for built-in styles, and crucial distinctions between output styles and other configuration features (like hooks or memory). * **Key Themes:** * *Enhanced Clarity:* More precise definitions of what each style (Proactive, Concise, etc.) actually does. * *Differentiation:* Clear boundaries established between Output Styles, `CLAUDE.md`, and Hooks. * *Detailed Behavior:* Specific details added, e.g., Concise mode preserves security warnings, Proactive mode doesn't bypass permissions. * **Impact Level:** Low. This is documentation-only change reflecting existing features (mostly). It clarifies how the tool works but doesn't change the API or behavior itself. * **Action Items:** * Review the new comparison table to pick the right style. * Update understanding of "Concise" mode (it keeps error/security full text). * Check if "Hooks" might be a better fit for rigid enforcement needs compared to Styles. 4. **Drafting the Response (in Chinese):** * *Summary:* `output-styles.md` 文档经过了重大重构和扩充。它提供了更清晰的定义、内置风格的对比表,以及与其他配置功能(如 Hooks 和 Memory)的关键区别说明。 * *Themes:* * **结构化清晰度**: 将内置风格(Proactive, Concise 等)的详细说明拆分为独立章节,并新增了对比表格。 * **功能边界界定**: 明确区分了“输出风格”与 `CLAUDE.md`(项目知识)和 `hooks`(强制执行机制)的适用场景。 * **行为细节补充**: 细化了特定模式下的行为,例如 Concise 模式仍保留完整的安全警告,Proactive 模式不改变权限模式。 * *Impact:* Low. 这是对现有功能文档的澄清和优化,不涉及 API 变更或行为破坏性变更。 * *Action Items:* * 查阅新的对比表格,以快速选择最适合当前任务的内置风格。 * 重新了解 `Concise` 风格的例外情况(错误报告和安全警告不会简化)。 * 区分 `Output Styles` 和 `Hooks` 的用途,以确保配置的正确性。 5. **Refining the Output:** * Ensure the tone is professional and "world-class programmer". * Check markdown formatting. * Verify Chinese translation accuracy (e.g., "permission modes" -> 权限模式, "hooks" -> 钩子/钩子机制). 6. **Final Polish:** * Summary: Concise. * Themes: Bullets. * Impact: Low + Reasoning. * Action Items: Max 5 bullets. (Self-Correction during drafting): The diff mentions "Learning" in the table but the text cut off at "An `I". I should focus on the visible changes. The key takeaway is the structural overhaul and the clarification of *what* output styles are versus what they aren't. Let's write the final response.

📚 Claude Code

15:29 EST

🤖 AI Batch Analysis

# 文档变更分析 ## 总体概览 本次更新主要将 **Claude Opus 5 升级为 Opus 5.5** 作为默认模型,同时改进了多个配置文档的结构和前端元数据规范。 ## 关键主题 - **🚀 模型更新:Opus 5.5 成为新标准** - Opus 5.5 现为所有主要提供商(Anthropic API、Bedrock、Vertex AI)的默认 `opus` 别名解析目标 - 新模型 ID 格式:`claude-opus-5-5`(替代 `claude-opus-5`) - Opus 5.5 强制启用扩展思考,无法关闭 - **💰 定价与性能优化** - Opus 5.5 定价:$4/MTok 输入,$20/MTok 输出(比 Opus 5 便宜 20%) - 缓存命中价格降至 0.05x(Opus 5 为 0.1x) - Fast mode 现支持 Opus 5.5:$8/MTok 输入,$40/MTok 输出 - **📝 文档结构改进** - 为 Rules、Skills、Sub-agents、Output Styles 和 Plugin Agents 添加了统一的 Frontmatter 参考表 - 明确了字段验证规则:未识别的字段会被静默忽略而非报错 - **🔒 功能行为变更** - `PermissionRequest` 事件不再支持 `agent` hooks(仅支持 `command`、`http`、`mcp_tool`、`prompt`) - 子代理现在默认继承会话模型选择(使用 `/model` 切换会影响子代理) - **🌐 网络与插件** - 添加 `github.com` 到网络配置域名列表(用于克隆插件市场) - 插件代理现在支持 `color` 和 `experimental` frontmatter 字段 ## 影响等级:**中等** 虽然存在大量页面变更(62页),但核心是模型别名更新。大多数开发者无需立即采取行动,除非他们显式固定了旧模型 ID。代码示例的更新主要是文档同步。 ## 开发者行动项 1. ⚠️ **检查模型固定设置**:如果你设置了 `ANTHROPIC_DEFAULT_OPUS_MODEL` 或类似的固定模型环境变量,请确认是否需要更新为 `claude-opus-5-5` 2. 💵 **评估成本影响**:Opus 5.5 比旧版本便宜 20%,但注意它会自动使用扩展思考,可能会增加输出 token 消耗 3. 🔧 **审查 Frontmatter 配置**:如果你使用 Skills、Rules 或 Sub-agents,检查 YAML frontmatter 字段拼写,未识别的字段现在会被静默忽略 4. 🪝 **更新 Hooks 配置**:如果你在 `PermissionRequest` 事件上配置了 `agent` hooks,需要迁移到 `command` 或 `http` hooks 5. 🧪 **更新测试和 CI 脚本**:GitHub Actions 示例和其他代码片段中的模型 ID 已更新,建议同步更新以使用最新模型
💭 Model Reasoning
用户想要我分析一批文档变更,共有62个页面。我会专注于开发者相关的关键变更,用中文提供简洁的分析。 让我先浏览一下所有页面的变更: 1. **amazon-bedrock.md**: 主要更新是将Opus 5更新为Opus 5.5,更新了默认模型解析和版本历史 2. **claude-code-on-the-web.md**: 移除了关于effort hold的说明 3. **costs.md**: 更新了Opus 5.5的说明,子代理模型继承说明 4. **desktop.md**: 更新了Opus 5.5的说明 5. **discover-plugins.md**: 更新了插件市场刷新逻辑 6. **github-actions.md**: 更新示例从claude-opus-4-8到claude-opus-5-5 7. **google-vertex-ai.md**: 更新Opus 5到Opus 5.5 8. **hooks.md**: PermissionRequest事件不再支持agent hooks 9. **interactive-mode.md**: 更新了Ctrl+L的描述 10. **memory.md**: 添加了Rule frontmatter参考,更新了AGENTS.md相关说明 11. **model-config.md**: 大量更新Opus 5到Opus 5.5,更新版本历史和默认模型 12. **network-config.md**: 添加了github.com到网络配置 13. **output-styles.md**: 添加了Frontmatter参考部分 14. **plugins-reference.md**: 添加了插件代理frontmatter参考,userConfig选项 15. **settings.md**: 更新示例中的模型名称 16. **skills.md**: 添加了Frontmatter参考 17. **slash-commands.md**: 添加了Frontmatter参考 18. **statusline.md**: 更新示例中的模型ID 19. **sub-agents.md**: 添加了Frontmatter参考,更新模型名称 20. **intro.md**: 更新为Opus 5.5 21. **get-started.md**: 更新所有代码示例中的模型名称 22. **about-claude/models/overview.md**: 更新模型比较表,添加Opus 5.5 我注意到这些变更主要集中在模型更新和配置文档的优化。大多数页面都在反映最新的模型版本Opus 5.5,并增加了更详细的配置参考说明。这些变更旨在提高文档的准确性和开发者体验。 新的Opus 5.5模型引入了重要的定价和功能变化。网络配置、前端内容和插件设置也得到了优化,增强了系统的灵活性和配置选项。 这些变更将显著影响开发者对模型的理解和使用策略。 Opus 5.5的推出标志着AI技术的重要进展,特别是在成本效益和性能平衡方面。 文档更新反映了这些技术演进,为用户提供了更清晰的模型选择和配置指导。 关键变更包括模型名称标准化、定价策略调整以及hook事件的精确管理。这些细微但重要的改进将帮助开发者更精准地配置和使用系统资源。

📚 Anthropic API

📚 Claude Code GitHub

11:21 EST

🤖 AI Batch Analysis

### 总体摘要 本次文档更新属于**文案润色**性质的变更,主要目的是统一术语表达、修正语法细节以及提升可读性。尽管涉及 11 个页面,但没有任何功能性的 API 变更、新增特性或破坏性更改。 ### 关键主题 * **语言风格统一**:全面将拉丁缩写(如 `e.g.`, `E.g.`)替换为全称英文短语 `for example`。这有助于提升机器翻译的准确性及非母语用户的阅读体验。 * **版本措辞规范化**:在描述依赖版本要求时(如 Chrome 扩展、VS Code),将 "or higher" 统一更改为 "or later",语义虽相同,但后者在软件版本语境中更为常用。 * **语法与标点微调**: * 修正了表格表头(`I want to...` -> `What I want to do`)以更符合语法规范。 * 调整了引用内容的标点位置(例如 Skills 页面中将指令内容移出引号范围)。 * **代码块格式异常**:在 `overview.md` 和 `quickstart.md` 中,代码块标记处出现了重复的 `theme={null}` 属性。这看起来像是文档生成工具的渲染产物或特定的格式修复,不影响实际内容展示。 ### 影响等级 **Low** (低) **理由**:所有更改均仅限于文档的文本层面,未涉及 API 接口、配置架构、CLI 命令或软件运行时的行为。开发者无需为了适配此次更新而修改任何代码或环境配置。 ### 开发者行动项 * **无需操作**:本次更新纯粹为了改善文档质量,对 Claude Code 的使用、开发或插件编写没有任何实际影响。
💭 Model Reasoning
1. **Analyze the Request**: * **Role**: World-class programmer/analyst. * **Task**: Analyze documentation changes for Claude Code. * **Input**: A batch of 11 changed pages with diffs. * **Output Format**: Markdown. * **Language**: Chinese. * **Structure**: Overall Summary, Key Themes, Impact Level, Action Items. 2. **Analyze the Diffs**: * *Page: chrome.md*: "version 1.0.36 or higher" -> "version 1.0.36 or later". * *Page: common-workflows.md*: "E.g.," -> "for example". * *Page: overview.md*: * Added multiple `theme={null}` attributes to code blocks (looks like a rendering artifact or specific fix). * "I want to..." -> "What I want to do" (Table header change). * *Page: plugin-marketplaces.md*: `string` (e.g., `"./my-plugin"`) -> `string` (for example, `"./my-plugin"`). * *Page: plugins-reference.md*: "e.g.," -> "for example". * *Page: quickstart.md*: * Added multiple `theme={null}` attributes to code blocks (same as overview). * "how do I..." -> "how do I" question". * *Page: skills.md*: * "Research $ARGUMENTS thoroughly..." -> "Research $ARGUMENTS thoroughly" instructions). * "(e.g., `0.02`)" -> "(for example, `0.02`)". * *Page: slack.md*: "e.g.," -> "for example". * *Page: slash-commands.md*: (Similar to skills.md) * "Research $ARGUMENTS thoroughly..." -> "Research $ARGUMENTS thoroughly" instructions). * "(e.g., `0.02`)" -> "(for example, `0.02`)". * *Page: statusline.md*: "e.g.," -> "for example". * *Page: vs-code.md*: * "1.94.0 or higher" -> "1.94.0 or later". * "1.0.36 or higher" -> "1.0.36 or later". * "e.g.," -> "for example". 3. **Synthesize Findings**: * **Pattern 1 (Language Style)**: There is a consistent shift from abbreviations like "e.g." and "E.g.," to the phrase "for example". Also, "I want to..." was changed to "What I want to do". * **Pattern 2 (Version Requirements)**: "or higher" was changed to "or later" in Chrome and VS Code docs. * **Pattern 3 (Code Block Artifacts)**: Repetitive `theme={null}` attributes appearing in `overview.md` and `quickstart.md` code blocks. This looks like a minor bug in the documentation generation or build process, or a specific formatting update. * **Pattern 4 (Grammar/Punctuation)**: Changes to quote usage, e.g., "how do I..." -> "how do I" question", and `("Research...")` -> `(the "Research..." instructions)`. 4. **Drafting the Analysis (in Chinese)**: * **Overall Summary**: This batch primarily consists of stylistic and grammatical refinements to improve documentation clarity and consistency, along with some minor formatting adjustments to code blocks. There are no functional changes to the API or tool behavior. * **Key Themes**: * *Standardization of "Example" phrasing*: Replacing Latin abbreviations ("e.g.", "E.g.,") with plain English ("for example") for better readability. * *Phrasing consistency*: Changing "or higher" to "or later" regarding version requirements. * *Code block formatting*: Addition of redundant `theme={null}` attributes (likely a build artifact or specific styling fix). * *Grammar improvements*: Small tweaks to sentence structure and quote usage (e.g., in the Skills section). * **Impact Level**: **Low**. These are purely documentation updates (cosmetic/linguistic). No breaking changes, new features, or deprecations. * **Action Items**: * None required for developers. These changes affect only how the documentation reads, not how the code runs. 5. **Refining the Output (World-class programmer persona)**: * Keep it concise. * Focus on *why* it matters (readability, localization friendliness). * Note the `theme={null}` weirdness but dismiss it as likely an artifact. * Translating "e.g." to "for example" is often a localization best practice or accessibility improvement. 6. **Final Polish (Chinese)**: * *Summary*: 此次更新主要集中在文档的语言风格规范化上,没有涉及产品功能变更。 * *Themes*: 统一术语、版本号措辞、代码块格式微调。 * *Impact*: 低。 * *Action Items*: 无需操作。 7. **Constructing the Markdown**: ```markdown ### 总体摘要 本次文档更新属于“文案润色”性质的变更,主要目的是统一术语表达和提升可读性。虽然涉及多个页面,但没有任何功能性的 API 变更或版本更新。 ### 关键主题 * **语言风格统一**:将拉丁缩写(如 "e.g.", "E.g.")替换为全称 "for example",这通常有助于提升翻译质量和阅读体验。 * **版本措辞调整**:在 Chrome 扩展和 VS Code 版本要求中,将 "or higher" 改为 "or later"(语义一致,但更符合版本描述习惯)。 * **语法与标点微调**:修复了一些引用语法的细节(如 Skills 页面中的引号位置)和表格表头("I want to..." -> "What I want to do")。 * **代码块格式异常**:`overview.md` 和 `quickstart.md` 中的代码块出现了重复的 `theme={null}` 属性,这看起来像是文档生成系统的渲染产物或轻微的格式漂移。 ### 影响等级 **Low** (低) **理由**:所有变更均仅限于文档文本层面,未涉及 API 端点、配置结构或软件行为。开发者无需修改代码或调整配置。 ### 开发者行动项 * **无需操作**:这些变更仅改善文档阅读体验,不影响 Claude Code 的使用或开发。 ```
01:45 EST

🤖 AI Batch Analysis

# 文档变更批量分析 ## 1. 总体摘要 本次更新涵盖 25 个页面,主要围绕**云会话中插件/技能加载机制的变更**(从 repo 声明转向 claude.ai 同步插件)、**安全增强**(不可见字符过滤)、**VS Code 交互体验改进**(逐项 diff 审查、内存/指令管理入口),以及多项行为澄清和错误修复说明。 ## 2. 关键主题 - **云会话插件机制重构**:云会话不再从 `.claude/settings.json` 的 `enabledPlugins` 安装插件,必须通过 claude.ai 账户启用同步插件;多仓库云会话仅读取 `enabledPlugins` 和 `extraKnownMarketplaces` 两个键 - **安全增强**:提交 prompt 前自动移除不可见 Unicode 字符(零宽空格、双向控制符等),二次确认后才发送;粘贴内容在 hooks 中展开传递 - **VS Code 体验升级**:diff 视图支持逐项 Accept/Reject(>100 处变更时回退为整文件审查);Customize 菜单新增 Memory 开关和 Instructions 编辑入口;文件选择自动遵守 `files.exclude`/`search.exclude` 和 gitignore - **Subagent 模型解析澄清**:当主会话模型属于某家族时,家族别名(如 `opus`)解析为主会话的精确模型(含扩展上下文);`CLAUDE_CODE_SUBAGENT_MODEL` 中的别名始终解析为别名指向的版本 - **Claude for Teams 功能提升**:现在包含 SSO 和 server-managed settings,缩小了与 Enterprise 的差距 - **内存与运维**:堆内存超过 2.5GB 时弹出警告,可用 `--continue` 重启或 `/compact` 释放;OTel helper 失败时停止导出并显示警告通知 - **Skills/Agents 增强**:worktree 中自动加载主 checkout 的 skills;同步技能为只下载不上传;插件 agent 支持子文件夹及作用域命名(`plugin:folder:name`) - **费用与计时修正**:数据驻留 1.1× 倍率现在也计入 `--max-budget-usd`;`total_duration_ms` 跨 resume 累计且不含空闲时间 ## 3. 影响等级:**Medium** **理由**:云会话插件加载机制是**破坏性变更**——依赖 `enabledPlugins` 在云会话中自动安装插件的工作流会失效,需迁移到同步插件方式。其余多为新增功能和行为澄清,不破坏现有本地工作流,但需要开发者注意若干行为变化(如 `excludedCommands` 对管道调用的局限性、粘贴内容的 hook 处理等)。 ## 4. 开发者行动项 - **⚠️ 迁移云会话插件配置**:若你在 `.claude/settings.json` 中声明了 `enabledPlugins` 期望云会话自动安装,需改为在 claude.ai 账户中启用同步插件,否则云会话将不再加载这些插件 - **更新 hooks 中 prompt 解析逻辑**:`UserPromptSubmit` hook 现在收到展开后的粘贴内容(含 `<pasted_content>` 标签),解析 prompt 的 hook 需适配此格式 - **注意 `excludedCommands` 对管道调用的局限**:将 `pbcopy *` 等加入 `excludedCommands` **不能**让 Claude 管道调用的该命令逃出沙箱;改用 `/copy` 命令或调整工作流 - **利用 VS Code 新功能提效**:升级至 v2.1.275+ 后可使用逐项 diff 审查、Memory 浏览/编辑、Instructions 快速编辑等新入口 - **关注内存使用**:若遇到会话变慢,留意 2.5GB 堆内存警告,及时用 `--continue` 重启或 `/compact` 释放
💭 Model Reasoning
Let me analyze the documentation changes across all 25 pages and provide a comprehensive summary. Key changes I'm seeing: 1. **CLI reference** - Table reformatting (mostly whitespace/formatting changes) 2. **Costs** - Clarification that 1.1× data residency multiplier also counts toward `--max-budget-usd` 3. **Desktop** - Cloud sessions no longer install plugins from repo's `.claude/settings.json`; must use synced plugins 4. **Discover plugins** - Cloud sessions now require synced plugins instead of `enabledPlugins` in settings.json; marketplace add wording improvements; marketplace confirmation behavior change 5. **Headless** - Cost tracking now reports conversation's whole total when using `--continue`/`--resume`; subagent forwarding now includes forked skills 6. **Hooks** - Cloud session hooks come from synced plugins not repo-declared plugins; UserPromptSubmit now receives expanded pasted content; SubagentStop fires for internal agents too 7. **IAM** - Claude for Teams now includes SSO and server-managed settings; Enterprise simplified 8. **Interactive mode** - New section on invisible character removal from prompts 9. **Memory** - `/init` interactive flow env var setup clarified; additional directories env var persistence; user vs project rule priority clarification (neither overrides the other) 10. **Model config** - Fable model picker behavior changed - now shows unless restricted, grays out when unavailable 11. **Monitoring usage** - OTel headers helper failure now stops telemetry exports and shows warning notification 12. **Overview** - Added Claude Academy courses link 13. **Plugins reference** - Plugin agents can now be in subfolders with scoped naming; manifest `agents` field can override naming 14. **Quickstart** - Added Claude Academy link 15. **Sandboxing** - `$TMPDIR` handling improved when unset; `excludedCommands` links added; clipboard tools in `excludedCommands` don't take calls out of sandbox on their own 16. **Settings** - Multi-repo cloud sessions now only read `enabledPlugins` and `extraKnownMarketplaces` from repo settings; declared plugins still don't load 17. **Skills** - Git worktree skill loading; cloud sessions don't load repo-declared plugins; synced skills are download-only (never uploaded) 18. **Slash commands** - Same skill changes as skills.md 19. **Statusline** - `total_duration_ms` now accumulates across resumes, excludes idle time 20. **Sub-agents** - Explore/Plan skip git status snapshot (not parent session's); tools/disallowedTools now accept YAML list format; subagent model family alias resolution clarified 21. **Terminal config** - Paste behavior restructured; invisible character handling; how Claude treats pasted text; file-based workflow recommendation moved up 22. **Third-party integrations** - Same IAM changes (Teams gets SSO + server-managed settings) 23. **Troubleshooting** - Memory warning at 2.5GB with recovery steps; clipboard excludedCommands clarification 24. **VS Code** - Per-change review in diff view; Memory and Instructions in Customize section; file selection filtering by gitignore/exclude settings; image/file attachment clarified 25. **Effort** - Clarification that in tool-use loops, follow-up requests processing tool results can skip thinking Key themes: - Cloud session plugin/skill behavior changes (significant) - Security improvements (invisible character stripping) - VS Code UX improvements (per-change diff review, memory/instructions management) - Subagent and skill improvements (worktree support, model resolution, forked skills) - Teams plan now includes SSO and server-managed settings - Memory management guidance (2.5GB warning) - Various behavior clarifications and corrections

📚 Anthropic API