Changes on 2026-02-21

用户希望我分析 8 个文档页面的变更并提供摘要。变更主要涉及： 1. `cli-reference.md`: 新增 `claude agents` 命令，`--worktree` 标志位置调整 2. `common-workflows.md`: 新增 Subagent worktrees 部分，非 git 版本控制系统支持 3. `hooks.md` 和 `hooks-guide.md`: 新增 WorktreeCreate 和 WorktreeRemove 事件，生命周期图更新 4. `model-config.md`: 新增禁用 1M 上下文的环境变量 5. `overview.md`: 代码块主题属性修复（重复的 theme={null}） 6. `settings.md`: 环境变量排序调整，新增 CLAUDE_CODE_DISABLE_1M_CONTEXT，移除 CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS 7. `sub-agents.md`: 新增 `claude agents` 命令说明 主要主题： - Worktree 功能扩展：支持 subagent worktrees 和非 git VCS - Hooks 系统增强：新增 WorktreeCreate/WorktreeRemove 事件 - CLI 命令增强：新增 `claude agents` 命令 - 配置选项：新增禁用 1M 上下文的选项 - 文档修复：代码块格式问题 影响级别：Medium - 新增功能但非破坏性变更 行动项： - 了解新的 worktree 隔离功能用于 subagents - 如使用非 git VCS，考虑配置 WorktreeCreate/WorktreeRemove hooks - 企业环境可能需要禁用 1M 上下文功能以适应特定安全要求 - 新的 `claude agents` 命令提供了更便捷的 subagent 列表查看方式

1. **Analyze the Request:** * **Role:** World-class programmer/analyst. * **Task:** Analyze documentation changes for `model-config.md` as a single batch. * **Format:** Markdown (Concise, insightful, Chinese response). * **Inputs:** Diff of `model-config.md` (+2/-6 lines). 2. **Analyze the Diff:** * **Section 1 (Lines 78-84 in old):** Removed a specific table mapping user types (Max/Team/Pro vs. Pay-as-you-go) to specific model versions (Opus 4.6 vs. Sonnet 4.5). * **Section 2 (Lines 78-79 in new):** Replaced the table with a link to a different section (`#default-model-setting`). This suggests the documentation is being centralized or decoupled to avoid hardcoding specific model version strings in multiple places, likely because these versions change frequently. * **Section 3 (Lines 94 in old -> 89 in new):** Changed an example description. * Old: "ensures all users run Sonnet 4.5..." * New: "ensures all users run Sonnet 4.6..." * **Context:** The specific model versions (4.5 vs 4.6) are changing. The documentation is updating references to reflect the current default or preferred model (Sonnet 4.6 seems to be the new standard/example), while removing a hardcoded table of default models in favor of a link. 3. **Synthesize the Analysis:** * **Summary:** The documentation has been updated to reflect the latest model version (Sonnet 4.6) in examples and removed a hardcoded table mapping subscription tiers to specific models, replacing it with a link. * **Key Themes:** * Model version updates (4.5 -> 4.6). * Documentation maintainability (removing hardcoded data). * **Impact:** Low. This is purely documentation, reflecting the current state of the API/tool. No code changes are implied for the developer unless they were hardcoding logic based on the old docs (which is bad practice anyway). * **Action Items:** * Update internal docs if they reference the old specific versions. * Be aware that "Default" models change over time. 4. **Draft the Response (in Chinese):** * *Overall Summary:* 文档更新了默认模型设置的描述，移除了硬编码的订阅层级对应模型表格，并将示例中的模型版本从 Sonnet 4.5 更新为 Sonnet 4.6。 * *Key Themes:* * **模型版本迭代**: 文档反映了最新的模型版本（Sonnet 4.6），替换了旧版本引用。 * **文档维护优化**: 移除了具体的版本映射表，转而使用链接指向动态设置部分，避免因版本更新导致文档多处需要修改。 * *Impact Level:* **Low**. 这仅是文档层面的文字修正与模型版本同步，未涉及 API 结构或代码逻辑的重大变更。 * *Action Items:* * 查阅链接部分 (`#default-model-setting`) 以获取最新的默认模型映射逻辑。 * 如果依赖文档中的具体版本号做硬编码判断，请更新为动态获取或关注新的版本号。 5. **Review and Refine:** * The request asks for "insightful analysis". * The change from 4.5 to 4.6 in the example is a signal that Sonnet 4.6 is the new baseline. * The removal of the table suggests the "Default" model is dynamic and shouldn't be assumed to be static. * The translation looks natural. 6. **Final Output Generation** (Matches the drafted Chinese response).

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Format:** Markdown, concise, insightful, focusing on developer impact. * **Language:** Chinese (requested at the end). * **Input:** Diff of 4 pages (`overview.md`, `prefill-claudes-response.md`, `define-success.md`, `develop-tests.md`). 2. **Analyze the Diffs:** * *Page 1 & 2 (`overview.md`, `prefill-claudes-response.md`):* * **Change:** Updated an external link reference. * **Old:** Check out [Define your success criteria](...) and [Create strong empirical evaluations](...). * **New:** Check out [Define success criteria and build evaluations](/docs/en/test-and-evaluate/develop-tests). * **Implication:** The documentation structure changed. Two distinct topics ("Define success criteria" and "Create strong empirical evaluations") were likely merged or the content was consolidated into a single page (`develop-tests.md`). The link now points to `develop-tests.md` instead of `define-success.md`. * *Page 3 (`define-success.md`):* * **Title Change:** "Define your success criteria" -> "Define success criteria and build evaluations". * **Content Shift:** The page starts adding content related to *building evaluations* (e.g., "Eval design principles", "Example evals" with Python code snippets for sentiment analysis and cosine similarity). * **Observation:** This page now covers both defining criteria *and* building the actual evaluations (test cases/code), which was previously split. * *Page 4 (`develop-tests.md`):* * **Title Change:** "Create strong empirical evaluations" -> "Define success criteria and build evaluations". * **Content Shift:** This page *also* starts with the "Define your success criteria" content (which was originally in `define-success.md`). It contains the definitions of Specific, Measurable, Achievable, Relevant criteria and common success criteria lists. * **Observation:** This looks like a merge or massive cross-referencing scenario. * **Critical Check:** Wait, looking at the diffs for `define-success.md` and `develop-tests.md`, both seem to have changed titles to "Define success criteria and build evaluations". * *Re-evaluating Page 3 vs Page 4:* * `define-success.md` (old) + `develop-tests.md` (old) -> merged into `develop-tests.md` (new)? * Let's look at the "Next steps" in `develop-tests.md` (Page 4 diff). It links to "Brainstorm criteria". * Let's look at the links in Page 1 & 2. They link to `develop-tests.md`. * Let's look at the content of `define-success.md` (Page 3 diff). It adds `Build evaluations` section with code. * Let's look at the content of `develop-tests.md` (Page 4 diff). It adds `Define your success criteria` section at the top. * *Conclusion on Structure:* It appears the content from the old `define-success.md` page was *moved* or *copied* into `develop-tests.md`. The old `define-success.md` page (Page 3) *also* changed its title and added eval content, but wait... * *Closer look at Page 3 (`define-success.md`)*: The diff shows +267/-9. It adds a huge chunk of Python code for evals. * *Closer look at Page 4 (`develop-tests.md`)*: The diff shows +57/-7. It adds the "Define your success criteria" content at the beginning. * *Hypothesis:* The two pages (`define-success.md` and `develop-tests.md`) are becoming mirrors of each other or merging into one comprehensive guide titled "Define success criteria and build evaluations". The "Next steps" in Page 4 suggests users might go to a tool, but doesn't link back to `define-success.md`. * *Wait, look at the links in Page 1 & 2 again:* They point to `/docs/en/test-and-evaluate/develop-tests`. * *This suggests:* `develop-tests.md` is the canonical page now. `define-success.md` might be redirected or deprecated, or perhaps both pages now serve as the same entry point (consolidation). * *However*, looking at the raw changes: * Page 4 (`develop-tests.md`) imported the "Define success criteria" content from Page 3. * Page 3 (`define-success.md`) imported the "Build evaluations" code examples from Page 4? * Actually, looking at the diffs, it looks like a **Consolidation**. The two concepts are now treated as one single workflow in a single document (or two identical documents?). * Let's assume the goal is: "Define criteria and build evaluations" is a single unit. The documentation has been consolidated so developers see the whole picture (Criteria -> Tests) in one go. 3. **Synthesize the Analysis:** * **Summary:** The documentation consolidated "Defining Success Criteria" and "Building Evaluations" into a unified workflow, likely merging content into `develop-tests.md` and updating cross-references. * **Key Themes:** * *Content Consolidation:* Merging distinct steps of the prompt engineering workflow (defining metrics + building tests) into a single, comprehensive guide. * *Updated Navigation:* External pointers now direct to the unified page (`develop-tests.md`). * *Enhanced Code Examples:* Added concrete Python code snippets (sentiment analysis, cosine similarity) for evaluation. * **Impact:** Medium. This is a structural improvement. Developers won't lose information, but the link structure changed. It improves discoverability by keeping related steps together. * **Action Items:** * Update internal bookmarks if pointing to the old specific URLs. * Check for 301 redirects if `define-success.md` is being retired (though the diff shows it still exists, just changed). * Review the new "Eval design principles" section for best practices. 4. **Drafting the Response (in Chinese):** * *Headings:* 总体概述, 关键主题, 影响等级, 开发者行动. * *Drafting Content:* * Summary: 文档将“定义成功标准”和“构建评估”合并为一个统一的工作流页面。 * Themes: 内容整合, 导航更新, 代码增强. * Impact: Medium. Improves flow, but links changed. * Actions: Update bookmarks, review new code examples. 5. **Refining the Analysis (Self-Correction during drafting):** * *Observation:* The diff for `define-success.md` *also* shows it is receiving eval content. It's possible both pages are being kept in sync or merged into the same file path in the final build, but the file changes suggest `develop-tests.md` is now the primary destination (based on the links in Page 1 & 2). * *Key takeaway:* It's no longer "Step 1: Define Criteria (Link A), Step 2: Build Tests (Link B)". It's "Step 1 & 2: Define Criteria and Build Tests (Link A)". 6. **Final Polish (Markdown & Chinese):** * Use professional tone. * Ensure specific file names are mentioned. *(Self-Correction on the `define-success.md` diff)*: It seems the `define-success.md` page itself changed title and added the "Build evaluations" section. It looks like *both* pages might be converging, OR `define-success.md` is being renamed/replaced by `develop-tests.md`, but the diff provided shows changes to both files as if they are being synced. However, the most actionable change is that *other pages* now link specifically to `develop-tests.md`. Let's focus on the *developer experience*: 1. **Unified Workflow:** Instead of reading two separate pages to set up evals, the information is now combined. 2. **Link rot:** Links to `define-success.md` in the prompt engineering pages were removed/consolidated. 3. **Practical Value:** The new Python examples are very useful. *Final Structure:* 1. **Summary:** Consolidated evaluation docs. 2. **Themes:** Consolidation, Better Examples, Logical Flow. 3. **Impact:** Medium. 4. **Actions:** Update links, read new code. *Translation to Chinese:* - Consolidated -> 整合/合并 - Success criteria -> 成功标准 - Empirical evaluations -> 经验评估/实证评估 - Link rot -> 链接失效/更新 Let's write it out. *Draft:* **总体概述**: 文档对“测试与评估”部分进行了重大重组，将“定义成功标准”与“构建测试”这两个步骤合并到一个统一的页面中，并提供了更多可执行的代码示例。 **关键主题**: - **文档结构整合**: `define-success.md` 和 `develop-tests.md` 的内容被整合为一个名为“Define success criteria and build evaluations”的完整工作流。这表明文档编写方认为这两者不可分割。 - **导航链接更新**: Prompt Engineering 概览页面中的引用链接已更新，不再指向两个独立的页面，而是直接指向整合后的 `develop-tests.md`。 - **增强的实操内容**: 新增了大量具体的 Python 代码示例（如情感分析的精确匹配评估、FAQ 机器人的余弦相似度评估），使评估方法更具可操作性。 **影响等级**: Medium **原因**: 这次变更主要是信息架构的优化。它通过合并相关概念改善了开发者的阅读体验，没有破坏现有的 API 或代码逻辑，但需要开发者关注新的文档路径。 **开发者行动**: - **更新书签**: 如果您保存了旧的 `define-success.md` 链接，请更新为新的 `develop-tests.md`。 - **查阅新示例**: 利用新增的 Python 代码片段（精确匹配、余弦相似度）来参考如何构建自动化的评估脚本。 - **遵循新工作流**: 采用新的“定义标准 -> 构建评估”一体化流程，确保在编写 Prompt 前建立完整的评估基准。 This looks solid. Proceed to generate output.

1. **Analyze the Request:** * **Role:** World-class programmer/analyst. * **Task:** Analyze documentation changes as a single batch. * **Input:** Two diffs (`common-workflows.md` and `vs-code.md`). * **Output Format:** Markdown (Concise, insightful, in Chinese). * **Structure:** 1. Overall Summary. 2. Key Themes. 3. Impact Level. 4. Action Items. 2. **Analyze the Diffs:** * **Page 1: `common-workflows.md`** * Change 1: Text description changed from `` `--worktree` `` to `` `--worktree` (`-w`) ``. Added short flag documentation. * Change 2: Code examples changed from `claude -w feature-auth` to `claude --worktree feature-auth`. * Change 3: Code examples changed from `claude -w bugfix-123` to `claude --worktree bugfix-123`. * Change 4: Code examples changed from `claude -w` to `claude --worktree`. * *Observation:* The text now mentions the short flag `-w`, but the examples *switch* from using `-w` to using the long form `--worktree`. * **Page 2: `vs-code.md`** * Change 1: Text description changed from `` `--worktree` `` to `` `--worktree` (`-w`) ``. Added short flag documentation. * Change 2: Code example changed from `claude -w feature-auth` to `claude --worktree feature-auth`. * *Observation:* Same pattern as Page 1. 3. **Synthesize the Meaning:** * **What changed?** The documentation was updated to explicitly mention the short flag `-w` is available. However, all code examples were refactored to use the long form `--worktree` instead of the short form `-w` (which was used in the previous version). * **Why?** Likely for better readability in examples (explicit is better than implicit), while acknowledging the shortcut exists in the prose. This standardizes the documentation style to favor the more descriptive flag in code blocks. 4. **Drafting the Content (Internal Draft - English -> Chinese):** * *Overall Summary:* Documentation updated to standardize Git worktree examples, switching from short flag `-w` to long flag `--worktree` in code blocks, while adding short flag notation in the text description. * *Chinese:* 文档更新了 Git worktree 的使用示例，将代码块中的短标志 `-w` 统一替换为长标志 `--worktree`，同时在文本描述中增加了短标志的说明。 * *Key Themes:* * Consistency in command-line examples. * Better documentation (mentioning aliases). * No functional change to the tool itself. * *Chinese:* 示例代码一致性；命令选项文档完整性；无功能变更。 * *Impact Level:* Low. It's just syntax preference in docs. The tool supports both. * *Chinese:* 低。仅文档层面的语法规范化，不影响现有命令的使用。 * *Action Items:* * None strictly required for users. * Developers updating docs should follow this pattern. * *Chinese:* 无需修改现有代码/脚本；编写新文档时遵循此风格（优先使用长标志）。 5. **Refining the Output (Applying "World-class programmer" persona):** * *Refining the "Why it matters":* It matters because explicit flags (`--worktree`) are self-documenting for people reading the docs, whereas the docs now acknowledge that power users can type `-w` to save time. * *Drafting the Final Response in Chinese:* 1. **Overall Summary**: 文档更新了 `--worktree` 功能的说明，在文本中显式标注了短选项 `-w`，但将所有代码示例从使用 `-w` 改为使用更明确的 `--worktree`，以提高可读性。 2. **Key Themes**: * **代码示例规范化**: 统一使用长格式标志 (`--worktree`) 替代短格式，增强代码示例的自解释性。 * **选项文档完整化**: 在段落描述中补充了 `-w` 作为别名，兼顾了详细说明与便捷性提示。 3. **Impact Level**: **低** (Low)。 * **Reasoning**: 这仅是文档展示风格的一致性调整。CLI 工具本身的功能和参数支持未发生任何变化，旧命令（使用 `-w`）依然有效。 4. **Action items**: * **无需操作**: 如果你已经在脚本中使用 `claude -w`，无需修改，它们继续有效。 * **最佳实践**: 在未来的脚本或文档编写中，建议参考此更新优先使用 `--worktree` 以提高代码可读性，除非需要极简的输入。 6. **Final Polish:** Check against the markdown format and constraints. Looks good.