1. **Overall Summary**
这批文档更新主要增强了安全性与策略控制能力：引入了沙箱网络的“严格允许列表”模式（禁止白名单外访问且不再提示），并明确了 Hooks 机制现在会穿透到子代理内部执行，实现了对 AI 子任务更深层的监控与管理。
2. **Key Themes**
* **沙箱网络零信任 (`strictAllowlist`)**：新增配置项允许管理员强制禁止白名单外的网络访问。不同于之前的“提示用户”，新模式会直接拒绝请求，适用于自动化或高安全环境。
* **Hooks 作用域扩展至子代理**：定义在设置文件、托管策略或插件中的 Hooks（如 `PreToolUse`）现在会在子代理内部触发。这意味着开发者可以通过 Hooks 审控或干预子代理的工具调用。
* **策略执行优先级**：明确了 Hooks 的合并策略是累加而非覆盖，且管理策略中的 Hooks 无法被下层设置禁用。HTTP Hooks 的白名单控制对所有来源生效。
3. **Impact Level**
**Medium**
* **理由**：虽然 `strictAllowlist` 默认关闭，但一旦启用将显著改变沙箱行为（从“询问”变为“拒绝”），若配置不当可能导致运行时错误。Hooks 行为的变化主要是功能增强，通常不会破坏现有逻辑，但改变了开发者对子代理的监控能力。
4. **Action items for developers**
* **评估网络配置**：如果在无人值守或 CI 环境中使用沙箱，考虑在用户或托管设置中启用 `network.strictAllowlist` 以消除阻塞式提示。
* **审查子代理调用**：如果您依赖 Hooks 进行日志记录或权限控制，请验证这些逻辑现在是否能正确捕获子代理内部的工具调用事件。
* **更新安全策略**：注意 `.claude/settings.json` 中无法设置 `strictAllowlist`，必须提升至用户级或 CLI 参数级别，请相应调整配置管理脚本。
💭 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**: Documentation diffs for 4 pages (`hooks.md`, `sandboxing.md`, `settings.md`, `sub-agents.md`).
* **Output Structure**: 1. Overall Summary, 2. Key Themes, 3. Impact Level, 4. Action Items.
2. **Analyze the Content**:
* **Page 1: `hooks.md`**
* *Change*: Added details about hooks running in subagents (carrying `agent_id` and `agent_type`).
* *Change*: Clarified hook entry merging behavior (additive across levels, managed hooks can't be easily disabled).
* *Change*: Clarified HTTP hook allowlists apply to *all* sources (`allowedHttpHookUrls`, `httpHookAllowedEnvVars`).
* *Core concept*: Hooks now have tighter integration with subagents and stricter/defined behavior regarding merging and security (HTTP allowlists).
* **Page 2: `sandboxing.md`**
* *Change*: Added "Strict allowlist" feature (`strictAllowlist`).
* *Details*: If true, denies access outside the allowlist instead of prompting. Applies to sandboxed commands, not in-process tools like `WebFetch`. Only works in user/managed/CLI settings, not repo local settings. Requires v2.1.219+.
* *Core concept*: New security feature for stricter network control in sandboxes.
* **Page 3: `settings.md`**
* *Change*: Added entry for `network.strictAllowlist` in the settings table.
* *Details*: Mirrors the description in `sandboxing.md` (deny instead of prompt, version requirement, scope limitations).
* *Core concept*: Configuration documentation for the new feature.
* **Page 4: `sub-agents.md`**
* *Change*: Updated how hooks work in subagents. Hooks defined in `settings.json` now fire *inside* subagents (e.g., `PreToolUse` before a subagent uses a tool).
* *Change*: Clarified that hooks from settings, managed policies, and plugins all apply inside subagents.
* *Core concept*: Enhanced observability/control over subagent behavior via global/managed hooks.
3. **Synthesize and Structure (Mental Draft -> Chinese)**:
* **Overall Summary**:
* These changes introduce tighter security controls for sandbox networking and extend the scope of hooks to operate consistently within subagents.
* *Chinese*: 这批更新引入了更严格的沙箱网络控制（严格允许列表），并扩展了 Hooks 的作用域，使其在子代理中也能一致地运行。
* **Key Themes**:
* *Enhanced Security*: Strict allowlist mode (`strictAllowlist`) to enforce zero-trust network access without prompts.
* *Hooks Expansion*: Hooks now propagate to subagents (settings/managed/plugin hooks run inside subagents, providing better observability/control).
* *Policy Enforcement*: Clarification that managed hooks and HTTP allowlists are authoritative and additive, preventing users from easily bypassing them.
* *Chinese*:
* **安全性增强**: 引入 `strictAllowlist` 模式，强制执行零信任网络访问，不再提示用户。
* **Hooks 范围扩展**: Hooks 现在会传播到子代理内部，确保设置和策略在子会话中保持一致。
* **策略执行强化**: 明确了管理策略（Managed Settings）下的 Hooks 和 HTTP 允许列表具有优先权且行为是累加的。
* **Impact Level**:
* *Reasoning*: `strictAllowlist` is a potential breaking change if configured incorrectly (denies instead of prompts), but it's opt-in. The hooks changes are additive improvements in functionality. Overall, medium impact for enterprise/admin users, low for casual users.
* *Rating*: Medium.
* *Reasoning (Chinese)*: 新功能 `strictAllowlist` 是可选的，但如果配置不当可能导致拒绝服务（静默拒绝而非提示），对企业级用户影响较大。Hooks 的变化是行为增强，属于功能补全。
* **Action Items**:
* Review network settings if you manage sandboxes (check if `strictAllowlist` is needed).
* Test hooks behavior in subagents if you rely on them for auditing or workflow modification.
* Update documentation for internal teams regarding the additive nature of hooks.
* *Chinese*:
* 如果您管理沙箱环境，请评估是否需要启用 `network.strictAllowlist` 以实现自动化的网络拦截。
* 验证现有的 Hooks 在子代理调用工具时是否按预期触发（特别是 `PreToolUse` 和 `PostToolUse`）。
* 检查并更新组织内部的 Hook 配置，理解它们现在是累加合并且无法被下级设置完全覆盖的。
4. **Final Review against Constraints**:
* Markdown format? Yes.
* Concise? Yes.
* Chinese? Yes.
* Focus on developers? Yes.
5. **Construct Final Output** (Translating thoughts to the final markdown response).