# 文档变更分析
## 1. 总体摘要
本次更新核心在于引入了全新的 **"Auto Mode"（自动模式）** 权限管理功能，旨在通过后台安全检查减少交互式提示，并对 **LiteLLM** 依赖发布了紧急**安全警报**。此外，文档还对权限模式相关的链接和配置结构进行了标准化整理。
## 2. 关键主题
* **Auto Mode (自动模式) 全面发布**：新增了一种权限模式，允许 Claude 在后台进行安全检查后自动执行操作，从而减少权限弹窗。该功能目前处于研究预览阶段，需 Team 计划（企业版即将推出）及 Claude Sonnet 4.6 / Opus 4.6 模型支持。
* **关键安全警告**：针对 LiteLLM PyPI 版本 1.82.7 和 1.82.8 发布了严重安全警报，确认存在窃取凭据的恶意软件。
* **管理设置与配置更新**：新增了 `disableAutoMode` 和 `permissions.disableBypassPermissionsMode` 等管理选项，以便企业管控可用模式。
* **文档链接重组**：所有关于权限模式的内部链接统一指向 `/en/permission-modes`，并细化了不同子模式的锚点（如 `#skip-all-checks-with-bypasspermissions-mode`）。
* **子代理 继承逻辑明确化**：明确了当父会话处于 "Auto Mode" 时，子代理将继承该模式并忽略其自身的 `permissionMode` 配置。
## 3. 影响程度：**高**
* **理由**：
1. **安全性**：LiteLLM 的安全警告直接涉及凭据安全，对使用该集成的开发者极为关键。
2. **功能性**：Auto Mode 的引入显著改变了 Claude 的交互方式和自动化能力，对于企业/团队用户的工作流有重大影响。
3. **兼容性**：Hooks 和 Sub-agents 的行为变更可能需要现有脚本进行适配更新。
## 4. 开发者行动项
* **立即检查安全依赖**：如果你使用了 LiteLLM，请立即检查版本是否为 1.82.7 或 1.82.8，如有安装需立即卸载、轮换密钥并按照官方指引进行补救。
* **更新 Hooks 代码**：检查你的自定义 Hooks 是否解析 `permission_mode` 字段，确保代码能处理新增的 `"auto"` 状态值。
* **审查模型版本要求**：如果想启用 Auto Mode，请确认你的环境已配置 Claude Sonnet 4.6 或 Opus 4.6。
* **调整管理配置**：企业管理员应更新管理设置，根据公司安全策略决定是否通过 `disableAutoMode` 禁用自动模式。
* **更新子代理配置**：检查现有的 Sub-agent 定义，特别是当主会话可能切换到 Auto Mode 时，确认子代理的行为是否符合预期。
💭 Model Reasoning
1. **Understand the Goal**: The user wants a concise, insightful analysis of a batch of documentation changes (9 pages) provided as diffs. I need to summarize what changed, why it matters, key themes, impact level, and action items. The response must be in Chinese.
2. **Analyze the Diffs Page by Page**:
* **Page: `cli-reference.md`**
* *Addition*: `claude auto-mode defaults` command to print built-in classifier rules.
* *Change*: `--dangerously-skip-permissions` link changed from `/en/permissions#permission-modes` to `/en/permission-modes#skip-all-checks-with-bypasspermissions-mode`.
* *Addition*: `--enable-auto-mode` flag added.
* *Change*: `--permission-mode` link changed to `/en/permission-modes`.
* *Theme*: Introduction of "Auto Mode" feature, reorganizing permission mode documentation links.
* **Page: `desktop.md`**
* *Table Change*: Permission modes table updated.
* *Addition*: "Auto" mode added to the table (Settings key `auto`). Description mentions background safety checks, reducing prompts, research preview, Team plan requirement, Claude Sonnet 4.6 or Opus 4.6 requirement.
* *Change*: "Bypass permissions" description updated link and text (e.g., "runs without any permission prompts" vs "runs without permission prompts").
* *Change*: Mention of `dontAsk` mode link updated.
* *Managed Settings*:
* Key `disableBypassPermissionsMode` moved/renamed to `permissions.disableBypassPermissionsMode`.
* New key `disableAutoMode` added to prevent users from enabling Auto mode.
* Truncated diff at `autoMode`, likely more config options there.
* *Theme*: "Auto Mode" is a major new feature for desktop, specifically for Team/Enterprise. Managed settings structure updated.
* **Page: `hooks.md`**
* *Change*: `permission_mode` field description updated. Added `"auto"` to the list of possible modes.
* *Theme*: Hooks now recognize/propagate the "Auto" permission mode.
* **Page: `interactive-mode.md`**
* *Change*: "Shift+Tab" shortcut description updated. Previously: "Switch between Auto-Accept Mode, Plan Mode, and normal mode". Updated to mention "Auto" explicitly? The diff cuts off but it's clearly related to mode switching. *Correction*: The diff provided cuts off the content of the table row, but the previous pages confirm Auto mode is being integrated into shortcuts/shifters.
* **Page: `llm-gateway.md`**
* *Change*: A `<Note>` warning about LiteLLM was upgraded to a `<Warning>`.
* *Addition*: Specific security advisory added for LiteLLM PyPI versions 1.82.7 and 1.82.8 (malware). Instructions to remove, rotate credentials, and check issue #24518.
* *Theme*: Critical security update regarding a third-party dependency (LiteLLM).
* **Page: `overview.md`**
* *Change*: In the "I want to..." table, "iMessage" added to the list of sources for Channels.
* *Theme*: Minor feature expansion (iMessage support via Channels).
* **Page: `settings.md`**
* *Change*: The table header for permissions changed.
* *Diff cuts off*: The diff is cut off right after `model`, but the context implies settings updates.
* *Theme*: Configuration management updates (likely related to Auto mode, though the specific lines aren't fully visible in the snippet, the context from desktop.md suggests managed settings updates are happening).
* **Page: `sub-agents.md`**
* *Change*: Clarification on permission modes inheritance.
* *Addition*: If parent uses `auto mode`, subagent inherits it and ignores its own `permissionMode` frontmatter (classifier evaluates calls).
* *Change*: Link updates for permission modes documentation.
* *Theme*: Sub-agent behavior refined for the new Auto mode.
* **Page: `vs-code.md`**
* *Change*: `initialPermissionMode` setting description updated to include `auto`.
* *Change*: `allowDangerouslySkipPermissions` description updated. It now adds "Auto" to the mode selector. Clarifies Auto requires Team plan + Sonnet 4.6/Opus 4.6.
* *Theme*: VS Code extension updated to support Auto mode with plan restrictions.
3. **Synthesize the Themes**:
* **"Auto Mode" (Auto) Launch**: This is the big change across CLI, Desktop, VS Code, Hooks, and Sub-agents. It's a new permission mode that reduces prompts via background safety checks. It requires specific model versions (Sonnet 4.6/Opus 4.6) and a Team/Enterprise plan.
* **Documentation Reorganization**: Links to permission mode docs are being consolidated (`/en/permission-modes`).
* **Security Alert**: Critical warning about LiteLLM versions.
* **Minor Additions**: iMessage support in Channels.
4. **Determine Impact**:
* **High (Security)**: The LiteLLM warning is critical for anyone using that integration.
* **Medium/High (Feature)**: Auto Mode is a significant workflow enhancement for Team/Enterprise users. It changes how permissions work.
* **Medium (Configuration)**: Managed settings keys are changing/being added (`disableAutoMode`, `permissions.disableBypassPermissionsMode`).
* *Overall*: High impact due to the security advisory combined with a major feature release.
5. **Draft Action Items**:
* Check LiteLLM versions if used.
* Review "Auto Mode" availability and requirements (Team plan + specific models).
* Update managed settings if enforcing strict permission controls (disable Auto mode if needed).
* Update hooks/scripts that parse `permission_mode` to handle the new `"auto"` value.
* Review sub-agent configuration for potential inheritance changes with Auto mode.
6. **Draft the Response (in Chinese)**:
* *Summary*: 此次更新主要引入了新的 **"Auto Mode"（自动模式）**，该模式通过后台安全检查减少权限提示，同时针对集成 **LiteLLM** 的用户发布了关键的**安全警报**。文档结构也进行了微调。
* *Key Themes*:
* Auto Mode 的全面推出。
* LiteLLM 安全漏洞预警。
* 权限相关文档链接的重组与统一。
* Sub-agent 继承逻辑的明确化。
* Channels 支持 iMessage。
* *Impact*: High. (Reason: Security advisory is critical; new permission mode significantly alters workflow for Enterprise/Team users).
* *Action Items*:
* (Security) 如果使用 LiteLLM，立即检查并升级。
* (Dev) 更新解析 `permission_mode` 的 Hooks 代码以适配 `"auto"` 值。
* (Admin) 如果是管理员，审查 `disableAutoMode` 等管理设置。
* (User) 确认是否符合 Auto Mode 要求。
7. **Refine Language**: Ensure terms like "permission mode", "hooks", "managed settings" are handled naturally in the context of a developer-focused response.
* *Self-Correction during drafting*: Make sure to mention the model requirements (Sonnet 4.6/Opus 4.6) for Auto mode, as that's a technical constraint developers need to know.
8. **Final Polish**: Check against markdown formatting requirements.
* Summary: Concise.
* Themes: Bulleted.
* Impact: High (Security + Feature).
* Action Items: Max 5 bullets.
9. **Final Output Generation** (proceeding to generate output based on these steps).