Changes on 2026-07-22

1. **Analyze the Request**: * **Role**: World-class programmer/doc expert. * **Task**: Analyze documentation changes as a single batch. * **Input**: Diff of 7 pages (hooks.md, sub-agents.md, get-started.md, claude-on-amazon-bedrock.md, claude-on-vertex-ai.md, claude-in-microsoft-foundry.md, mcp-connector.md). * **Output Format**: Markdown (Overall Summary, Key Themes, Impact Level, Action Items). * **Language**: Chinese. 2. **Analyze the Diffs**: * **Page 1: `hooks.md`**: * Change: "Agents" -> "Subagents". * Change: Added a comment block `{/* min-version: 2.1.218 */}` explaining that frontmatter hooks in project subagents now require workspace trust acceptance. Before 2.1.218, they could run without trust (security/permissions fix). * **Page 2: `sub-agents.md`**: * Change: Update "Tools" description from "all tools" to "every tool [available to subagents]". Link added. * Change: Added a comment block `{/* min-version: 2.1.216 */}` detailing new security checks for Bash commands. Specifically, commands redirecting git into the main checkout (via `git -C`, `--git-dir`, etc.) or complex commands will fail. This only applies to Bash, not PowerShell. This prevents subagents from accessing areas outside their allowed scope (linked main checkout). * Change: The table structure remains, but the content updates reflect specific version notes (min-version tags). * **Page 3-7: `get-started.md`, `claude-on-amazon-bedrock.md`, `claude-on-vertex-ai.md`, `claude-in-microsoft-foundry.md`, `agents-and-tools/mcp-connector.md`**: * Change: Java SDK version downgrade from `2.50.0` to `2.48.0`. * *Self-Correction/Note*: It looks like a version rollback or correction. It's unusual to see a version go *down* in docs unless the higher version was yanked/broken or incorrect. 3. **Synthesize Themes**: * **Theme 1: Security & Sandboxing (Subagents)**. This is the meat of the update. * Hooks need workspace trust. * Bash commands in subagents have stricter git access prevention (preventing access to main checkouts). * Version tags (`2.1.216`, `2.1.218`) indicate recent patches tightening security. * **Theme 2: Dependency Corrections**. * Java SDK versions reverted from 2.50.0 to 2.48.0 across multiple integration docs (Bedrock, Vertex, Foundry, MCP). 4. **Drafting the Content (Iterative Refinement in Chinese)**: * *Summary*: 此次更新主要集中在增强子代理的安全隔离机制，以及修正多个 Java SDK 的版本依赖。 * *Key Themes*: * **增强的安全隔离**: * 工作区信任：项目子代理的 Frontmatter hooks 现在仅在接受工作区信任对话框后运行 (v2.1.218)。 * Bash 命令限制：增强了对子代理中 Bash 命令的检查，禁止通过特定标志（如 `git -C`）或复杂命令重定向到主检出版本，从而防止越界访问 (v2.1.216)。 * **依赖版本回退**: * Java SDK 相关库（Bedrock, Vertex, Foundry, MCP）的版本号从 `2.50.0` 回退至 `2.48.0`，建议开发者同步更新。 * *Impact*: * High? No. Breaking? Maybe for existing workflows relying on hooks in untrusted folders or specific bash command structures. * Let's go with **High** (because of the behavior change in security permissions) or **Medium**. Given the "subagent" scope is specific, and version rollback is specific to Java, maybe "Medium". * *Re-evaluation*: If someone has a subagent automation relying on git commands hitting the main checkout, it will break. That's significant for power users. But the Java version rollback is also critical for stability if 2.50.0 was broken. * Decision: **Medium**. (Security tightening is good but might break things; version fix is crucial). * *Action Items*: * Review subagent hooks: Ensure workspace trust is enabled/accepted for projects using subagent hooks. * Check subagent Bash commands: Look for `git -C`, `--git-dir`, or complex git commands in subagents that try to access the main checkout. * Update Java dependencies: Downgrade from `2.50.0` to `2.48.0`. 5. **Final Polish (translating to professional Chinese)**: * Summary: 本次文档更新重点在于强化子代理的安全隔离机制，特别是针对工作区信任和 Bash 命令的执行限制，同时修正了 Java SDK 的推荐版本号。 * Themes: * 安全加固: 子代理 hooks 现在严格依赖工作区信任，且 Bash 命令被禁止访问主检出版本。 * 版本修正: Java 相关 SDK 版本回退至 2.48.0。 * Impact: Medium. 行为变更可能导致现有的自动化脚本失败，且依赖版本回退对 Java 项目至关重要。 * Action Items: 1. 验证涉及子代理 Hooks 的项目是否已配置工作区信任。 2. 检查并修改子代理中试图通过 Git 操作主检出 的 Bash 命令。 3. Java 项目将 SDK 依赖从 2.50.0 降级至 2.48.0。 6. **Formatting Output**: Ensure Markdown structure matches requirements. * *Self-correction on "Impact"*: The hooks change is a security breaking change (hooks won't run if trust not accepted). The bash change is a functional breaking change (commands will error). The java version is a fix. I will label it "Medium/High" leaning towards High for those affected, but generally Medium. Let's stick to "Medium" as not everyone uses Subagents or Java. 7. **Final Output Generation**.

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Task:** Analyze a batch of 7 changed pages, summarize changes, explain relevance to developers. * **Format:** Markdown, Concise. * **Language:** Chinese. 2. **Analyze the diffs (Page by Page):** * **Page 1: `interactive-mode.md`** * *Change 1:* Updated `Ctrl+W` shortcut description. Added `Option+Delete` for macOS. * *Change 2:* Refactored and expanded the note about background task termination due to memory pressure. * Moved version check `{/* min-version: 2.1.193 */}` to the end. * Added condition: "no turn or subagent running". * Added new logic for v2.1.218: Commands owned by subagents are terminated after 60 minutes (configurable via `CLAUDE_SUBAGENT_BG_SHELL_MAX_MS`). * Added logic for v2.1.218 regarding commands moved to background with `Ctrl+B` not being covered by limits before this version. * **Page 2: `mcp.md`** * *Change:* Updated the startup notice description for servers needing authentication. * Moved version specific text to conditional comments. * Clarified behavior in v2.1.218: It counts only servers signable *in Claude Code*. * Clarified pre-v2.1.218 behavior: It included `claude.ai` connectors (which users couldn't sign into from the CLI). * **Page 3-7: `get-started.md`, `claude-on-amazon-bedrock.md`, `claude-on-vertex-ai.md`, `claude-in-microsoft-foundry.md`, `agents-and-tools/mcp-connector.md`** * *Change:* Updated Java SDK dependencies from `2.48.0` to `2.50.0` across various modules (standard, bedrock, vertex, foundry, mcp). 3. **Synthesize Key Themes:** * **Theme 1: Dependency Updates.** The Java SDK has been bumped to 2.50.0 in several integration guides. * **Theme 2: Interactive Mode Refinements.** Keyboard shortcut improvements for macOS and detailed versioning/history for background task management logic. 4. **Determine Impact Level:** * *Java SDK:* Version bump (2.48 -> 2.50). Usually patch/minor updates, not breaking unless stated otherwise. Low to Medium impact depending on how strict their dependency management is, but generally just a "keep up to date" signal. * *Interactive Mode/Background Tasks:* Added constraints (subagent timeouts) and fixed macOS shortcuts. This is helpful info but not usually breaking existing workflows unless they relied on the old specific behavior of background task longevity. * *MCP Authentication:* Clarification of behavior between versions. Low impact, mostly informational to prevent confusion. * *Overall:* **Low to Medium**. It's mostly dependency updates and documentation clarifications/fixes. No major API breaks highlighted in the diffs. 5. **Draft the Content (Iterative refinement in Chinese):** * *Summary:* This batch updates the Java SDK to version 2.50.0 and refines documentation regarding interactive mode shortcuts and background task management behavior (specifically regarding subagents and macOS). * *Key Themes:* * Java SDK 升级至 v2.50.0（覆盖 Bedrock, Vertex AI, Foundry, MCP 等模块）。 * 完善了交互模式下的快捷键说明（macOS `Option+Delete`）。 * 详细化了后台任务终止的逻辑，特别是针对 subagent 的限制和版本差异说明（v2.1.218）。 * 澄清了 MCP 服务器身份验证通知在不同版本间的行为差异。 * *Impact:* **Low**. These are dependency version bumps and documentation clarifications. No breaking API changes are evident in the diffs. * *Action Items:* * Update Java dependencies to `anthropic-java:2.50.0` and related modules. * Review the new environment variable `CLAUDE_SUBAGENT_BG_SHELL_MAX_MS` if you manage subagent background tasks. * Be aware that background tasks created via `Ctrl+B` now have specific coverage limits defined in v2.1.218. 6. **Final Polish (Formatting):** Ensure the markdown structure matches the prompt's requirements. * *Summary:* 本次文档更新主要将 Java SDK 升级至 v2.50.0，并完善了交互模式下 macOS 快捷键及后台任务管理行为的详细说明。 * *Themes:* * **依赖更新**: Java SDK (含 Bedrock, Vertex, Foundry, MCP 模块) 全面升级至 `2.50.0`。 * **交互模式优化**: 增补了 macOS 上 `Option+Delete` 的快捷键说明。 * **后台任务逻辑细化**: 增加了 v2.1.218 版本关于 Subagent 后台任务超时终止（可配置）的说明，以及旧版本行为的修正。 * **MCP 认证通知**: 澄清了启动时 MCP 服务器认证提示在不同版本间的过滤逻辑。 * *Impact:* **Low**. 主要是依赖库的小版本更新和文档说明的精确化修正，未涉及破坏性变更。 * *Action Items:* * **更新依赖**: 将项目中的 `anthropic-java` 及相关模块依赖版本更新至 `2.50.0`。 * **检查配置**: 如果使用了 Subagent 的后台任务，请关注新增的环境变量 `CLAUDE_SUBAGENT_BG_SHELL_MAX_MS`。 * **了解行为变更**: 注意 v2.1.218 后，后台任务（特别是通过 `Ctrl+B` 移动的）有了更明确的终止限制规则。 7. **Final Review against constraints:** * Single batch analysis? Yes. * Concise? Yes. * Markdown? Yes. * Chinese? Yes. * Focus on developers? Yes. 8. **Output Generation.** (Proceed to generate output based on step 6).

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Format:** Markdown, concise, insightful. * **Input:** 3 pages (`costs.md`, `monitoring-usage.md`, `CHANGELOG.md.md`). * **Language:** Chinese (as requested by "请用中文回复"). 2. **Analyze Page 1: `costs.md` (Cost Analysis)** * *Change 1:* Modified the description of `/usage`. Added that it flags behaviors like "long context" or "cache misses" if they account for 10%+ of usage. * *Change 2:* Added a new section "Why usage climbs in a long session". Explains long context accumulation, cache misses (mentioning subscription vs. API key lifetimes), scheduled tasks, agent teammates, and compaction costs. * *Key takeaway:* Better visibility into *where* tokens are going (diagnostics) and education on why idle/open sessions burn credits. 3. **Analyze Page 2: `monitoring-usage.md` (Telemetry/OTel)** * *Change 1:* Added section "Managed endpoints govern signal-specific endpoints". * *Detail:* If a generic `OTEL_EXPORTER_OTLP_ENDPOINT` is set in *managed settings*, it overrides signal-specific endpoints (like `OTEL_EXPORTER_OTLP_LOGS_ENDPOINT`) from user/shell settings. Same applies to headers/keys/certs. * *Exceptions:* If signal-specific vars are *in* the managed settings file, they stick. Exporter selectors/protocol vars still follow normal precedence. * *Change 2:* Updated "Configuration details" text to reflect this override logic. * *Change 3:* Updated Tracing section to reflect that managed settings govern the endpoint too. * *Key takeaway:* Stronger control for administrators (managed settings) over telemetry routing to prevent data leakage or misconfiguration. 4. **Analyze Page 3: `CHANGELOG.md.md` (Version 2.1.218)** * *Theme:* Huge list of fixes and improvements. * *Highlights:* * `/code-review` runs as background subagent (doesn't fill chat). * Accessibility fixes (screen reader announcements). * Path corruption fix (Windows `\u`). * `/usage` improvements (matches costs.md changes). * MCP connection error details. * Gateway spend metering fix (Bedrock ARNs). * Deep research no longer auto-launches (needs manual invoke). * Agent/Skill behavior changes (backgrounding, trust, security). 5. **Synthesize Themes:** * *Cost & Performance Visibility:* Deep dive into why sessions are expensive (long context, cache misses) + better diagnostics in `/usage`. * *Telemetry/Enterprise Control:* Managed settings now strictly override user telemetry endpoint configs. * *Stability & UX:* Many bug fixes (Windows paths, crashes, screen readers) and UX improvements (backgrounding code review). * *Agent/Skill Config:* Changes to how agents/skills run (backgrounding, permissions). 6. **Determine Impact Level:** * *Managed Settings Change:* This is technically a behavior change (breaking) for users who relied on overriding a managed endpoint with a shell var. However, it's likely intended for security/compliance. Impact: Medium/High for enterprise admins. * *Cost Visibility:* High for users on tight budgets (explains "mystery" costs). * *Path Corruption:* High for Windows users (prevents breaking functionality). * *Overall:* Medium-High. Significant bug fixes and important clarification on cost/telemetry behavior. 7. **Draft Action Items:** * Check `/usage` for cost anomalies. * Admins: Review managed OTEL config. * Windows users: Update to fix path corruption. * Developers: Update agent/skill configs regarding `background: false`. 8. **Translate to Chinese & Refine (Self-Correction during drafting):** * *Summary:* 本次更新聚焦于成本可观测性增强、企业级遥测管控收紧，以及大量稳定性修复。 * *Themes:* * 成本诊断：新增 `/usage` 的行为标记（如长上下文）和文档解释。 * 遥测管控：受管设置强制优先级，防止用户环境变量绕过管理员配置。 * 稳定性/UX：修复了 Windows 路径编码、Agent 行为变更（Code Review 后台运行）及无障碍功能。 * *Impact:* **Medium/High**. 企业配置和成本计量逻辑有变，加上关键 Bug 修复。 * *Actions:* * 关注 `/usage` 的新提示以优化 Token 消耗。 * 管理员需注意 OTEL 环境变量的优先级变化。 * Windows 用户应升级以修复路径乱码问题。 * 调整 Agent/Skill 配置（默认后台运行）。 9. **Final Polish:** Ensure tone is professional and concise. Use the requested markdown format. * *Reviewing costs.md change:* "It also flags behaviors such as long context or cache misses..." -> This is v2.1.218 feature. * *Reviewing monitoring-usage change:* "Managed endpoints govern signal-specific endpoints" -> v2.1.217 feature (based on min-version tag). * *Reviewing Changelog:* Covers v2.1.218. *Wait, a slight inconsistency:* The `monitoring-usage.md` change mentions `{/* min-version: 2.1.217 */}`. The `costs.md` change doesn't specify a version but correlates with Changelog 2.1.218 (usage flags). I should treat the changelog as the primary source for "what changed" but acknowledge the docs explain the "why". *Structure:* 1. Overall Summary. 2. Key Themes (Cost Insights, Enterprise Control, Stability). 3. Impact (Medium-High). 4. Action Items. *Translation check:* "Managed endpoints govern signal-specific endpoints" -> 受管端点统管特定信号端点。 "Cache misses" -> 缓存未命中。 "Background subagent" -> 后台子代理。 *Final Output Generation.*

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Task:** Analyze a batch of 7 changed pages and provide a summary. * **Format:** Markdown, concise, insightful. * **Language:** Chinese. * **Key Components:** Overall Summary, Key Themes, Impact Level (with reasoning), Action Items (max 5). 2. **Analyze the Input Data (Page by Page):** * **Page 1: `claude-code-on-the-web.md`** * *Change 1:* Added text about idle sessions: "If Claude asks a question and the session sits idle, you can still answer when you come back..." * *Change 2:* Teleporting behavior clarification. Added: "The terminal gets its own copy of the session: new work there stays local and doesn't appear in the cloud session... To keep steering from your phone after teleporting, start `/remote-control`..." * *Theme:* Session lifecycle, teleport/local/cloud distinction. * **Page 2: `desktop.md`** * *Change:* Renaming "Remote" to "Cloud" in the Environment selector dropdown. * *Context:* "Select Remote for Anthropic-hosted" -> "Select Cloud for Anthropic-hosted". * *Theme:* Terminology change (Remote -> Cloud). * **Page 3: `iam.md`** * *Change:* Login warning timing adjustment. From "within five days" to "within three days". * *Context:* Version requirement update (`{/* min-version: 2.1.217 */}`). Before 2.1.217, it was 5 days. * *Theme:* Minor behavior refinement, documentation of version change. * **Page 4: `interactive-mode.md`** * *Change 1:* New feature: Emoji shortcodes support (`:`). Typing `:heart:` inserts emoji. * *Change 2:* PR badge hyperlink behavior change. Now renders as hyperlink even when support isn't detected (e.g., SSH/tmux) (v2.1.217). * *Theme:* New UI features (emojis), rendering improvements (hyperlinks). * **Page 5: `memory.md`** * *Change:* Added explanation of glob brace expansion budget (1,000 patterns, 4 MiB). Clarifies behavior if budget is exceeded. * *Context:* Version `v2.1.217`. Mention of potential crash/stall in older versions with many brace groups. * *Theme:* Performance/stability fix (glob expansion), technical limitation clarification. * **Page 6: `settings.md`** * *Change:* Added `emojiCompletionEnabled` setting. * *Context:* Related to Page 4. * *Theme:* Configuration for new features. * **Page 7: `vs-code.md`** * *Change:* Tab name change in session history dialog: "Remote" -> "Web". * *Theme:* Terminology consistency with Page 2 (Remote -> Cloud/Web). 3. **Synthesize Key Themes:** * **Terminology Shift:** "Remote" is being renamed to "Cloud" (desktop, VS Code). * **New Features (v2.1.217):** Emoji shortcode support (`:`) and improved hyperlink rendering for PR badges. * **Session Management:** Clarifications on how `teleport` works (local vs. cloud state) and idle session handling. * **Performance/Stability:** Glob expansion budgeting to prevent crashes. * **Timing Adjustments:** Login expiry warning window (5 days -> 3 days). 4. **Drafting the Analysis (Internal Monologue/Rough Draft):** * *Summary:* The update focuses on renaming "Remote" to "Cloud" environments for clarity, adding small UX improvements like emoji support, and refining session management behaviors (teleporting and idle states). It also addresses stability issues with glob patterns. * *Themes:* * Terminology: Remote -> Cloud. * v2.1.217 Features: Emojis, hyperlinks. * Architecture: Teleport behavior clarification. * Stability: Glob expansion limits. * *Impact:* Low to Medium. Terminology changes might confuse existing users briefly. New features are additive. Stability fix is good. * *Action Items:* * Look for "Cloud" instead of "Remote". * Update to v2.1.217 for features/fixes. * Note that teleporting creates a local copy. * Watch out for glob expansion limits if using braces. 5. **Refining and Translating to Chinese (Target Output):** * **Overall Summary:** 此次文档更新主要涉及术语统一（将 "Remote" 重命名为 "Cloud"）、引入 v2.1.217 版本的新功能（Emoji 输入支持、PR 徽章链接优化），以及会话管理和性能稳定性的细节调整（如 Teleport 行为说明和 Glob 模式扩展限制）。 * **Key Themes:** * **术语统一**: Desktop 和 VS Code 环境配置中的 "Remote" 统一更名为 "Cloud"，以更准确描述 Anthropic 托管的会话。 * **v2.1.217 新特性**: 新增 Emoji shortcode 支持（使用 `:` 触发）；PR 状态徽章现在在无法检测链接支持的环境（如 SSH/tmux）中也会渲染为链接。 * **会话与状态管理**: 明确了 `teleport` 操作会创建会话的本地副本，云端会话与终端工作流将不再实时同步（除非使用 `--remote-control`）。 * **性能与稳定性**: 增加了 Glob 模式中花括号扩展的预算限制（1000 个模式 / 4 MiB），防止复杂规则导致 CLI 崩溃。 * **权限与认证**: 登录过期警告窗口从 5 天缩短至 3 天。 * **Impact Level:** * **Level: Low (to Medium)** * **Reasoning:** * 术语更改（Remote -> Cloud）可能需要用户短暂的适应，但主要是 UI 文字层面的更新。 * 新增的 Emoji 功能属于体验提升，不影响核心功能。 * Teleport 行为的澄清有助于开发者理解云端与本地代码库的同步机制，避免混淆。 * Glob 扩展限制属于底层稳定性修复，解决了潜在的崩溃风险，影响面主要是配置复杂的用户。 * **Action Items:** * **更新认知**: 在 UI 和文档中寻找 "Cloud" 而非 "Remote" 以指代云端托管环境。 * **版本升级**: 更新至 v2.1.217 版本以获得 Emoji 支持及 Glob 扩展的稳定性修复。 * **调整工作流**: 若习惯使用 `teleport` 传输会话，需注意后续在终端的操作仅保留在本地，不再同步回云端；需配合 `--remote-control` 实现远程控制。 * **检查配置**: 如果使用了大量花括号的 Glob 匹配规则，需注意新版本引入的 1000 个模式的扩展限制。 6. **Final Polish:** Ensure formatting matches the requested markdown structure and tone is professional/insightful. (Self-correction: Make sure the "Action items" are concise and actionable). 7. **Final Output Generation** (Proceeding to generate the Chinese response based on step 5).

1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** A batch of 10 documentation pages with diffs (markdown format). * **Output:** Concise, insightful analysis in markdown format, focusing on developer impact, in Chinese. * **Structure:** Overall Summary, Key Themes, Impact Level, Action Items. 2. **Analyze the Diffs (Page by Page):** * **Page: `intro.md`** * *Change:* Removed "and prompt generator" from the description of the Workbench card. * *Theme:* De-emphasizing/removing internal tooling (prompt generator) from the main intro. * **Page: `get-started.md`** * *Change:* "Anthropic Console account" -> "Claude Console account" with a link to `platform.claude.com`. * *Theme:* Branding consistency. Shifting from "Anthropic" to "Claude" for the console. * **Page: `api/overview.md`** * *Change:* Workbench link changed from `/workbench` to `/playground`. * *Theme:* URL structure changes (Workbench -> Playground). * **Page: `api/messages.md`** * *Change:* Added detailed descriptions for refusal categories (`cyber`, `bio`, `frontier_llm`, `reasoning_extraction`). Added a new category: `general_harms`. * *Change:* Updated example JSON payloads to show realistic data (IDs, text like "The grass is green", explanation strings). * *Theme:* Improved error reporting/transparency for safety refusals. API response structure updates (data examples). Added "reasoning_extraction" context (linking to adaptive thinking). * **Page: `api/models/list.md` & `api/beta/files/upload.md`** * *Change:* Model ID list count "26 more" -> "27 more". * *Change:* Added new model/beta: `"dreaming-2026-04-21"`. * *Theme:* New feature/flag ("dreaming"). * **Page: `build-with-claude/vision.md`** * *Change:* "Anthropic Workbench" -> "Workbench". Link changed from `/workbench/` to `/playground`. * *Theme:* Branding/URL consistency (Workbench -> Playground). * **Page: `build-with-claude/search-results.md`** * *Change:* Code examples updated. `anthropic` var renamed to `client`. * *Change:* Java code refactored: removed static main wrapper, removed `ToolUseBlockParam` import (use `response.content()` logic), improved handling of tool results. * *Theme:* Code example improvements (standardization on `client`, better Java implementation patterns). * **Page: `build-with-claude/prompt-engineering/overview.md`** * *Change:* Removed mention of "Console prompting tools" from intro. * *Change:* Changed "Prompt generator" link from internal Console dashboard to an external Colab notebook (Claude Cookbook metaprompt). * *Theme:* Shift from internal tool features to external resources (Cookbooks) for prompt generation. * **Page: `build-with-claude/prompt-engineering/prompting-tools.md`** * *Change:* *Major Rewrite.* Title changed from "Console prompting tools" to "Prompting best practices". * *Change:* Content replaced entirely. Old content (Generator, Templates, Improver) is gone. New content is a comprehensive guide on prompting (Fable 5, Sonnet 5, Opus 4.8, XML tags, roles, etc.). * *Theme:* Major documentation restructuring. Deprecated the "Console Prompting Tools" page in favor of a "Best Practices" guide. Mentions new models (Claude Fable 5, Mythos 5, Opus 4.8). 3. **Synthesize Key Themes:** * **Branding & Naming:** Shift from "Anthropic Console" to "Claude Console". "Workbench" is now "Playground". * **Documentation Restructuring:** The "Prompting Tools" page was completely replaced with "Prompting Best Practices". Focus moved from UI tool features to coding guides. * **New Models & Features:** Introduction of "Claude Fable 5", "Mythos 5", and "Opus 4.8". New beta flag "dreaming-2026-04-21". * **API Refinement:** Detailed refusal reasons (new `general_harms` category, detailed explanations) and better JSON examples. * **Code Quality:** Improved Java examples (removing static wrappers, better tool handling). 4. **Determine Impact Level:** * *Breaking?* No immediate breaking API changes in the *parameters*, but the shift in documentation (removing "Prompting Tools" from console context) implies a product strategy shift. * *High?* For the documentation consumer, yes. The removal of the "Prompting Tools" page is significant if they were using those docs for UI workflows. The mention of new models (Fable 5, Opus 4.8) is a major signal of upcoming releases. * *Medium/High:* Let's go with **High** because it reveals major new model updates (Fable 5, Opus 4.8) and a significant shift in how prompt engineering is documented (moving away from UI generators to code/manual best practices). 5. **Draft Action Items for Developers:** * Update code examples if copying from Java docs (refactoring patterns). * Review new refusal categories (`general_harms`) in error handling logic. * Check out the new "Prompting best practices" guide. * Watch out for the new "dreaming" beta feature/model. * Update internal documentation links if pointing to the old "Prompting Tools" or specific Workbench URLs. 6. **Translate to Chinese & Format:** * *Summary:* 本次更新主要涉及品牌重塑（Anthropic Console -> Claude Console）、重大文档结构调整以及新模型（Claude Fable 5/Opus 4.8）的预披露，并增强了 API 错误信息的透明度。 * *Themes:* 品牌与命名, 文档重构与模型预览, API 细节完善, 代码示例优化. * *Impact:* High. * *Actions:* 审查错误处理, 更新链接, 学习新最佳实践, 关注新模型. 7. **Refine content for "World-class programmer" persona:** * Focus on the *implication* of the "Prompting tools" page replacement. It suggests the prompt generator feature might be de-emphasized or moved out of the main product docs into Cookbooks. * Highlight the specific refusal categories changes—crucial for robust API integration. * Note the "dreaming" beta flag—potential for a new feature capability. 8. **Final Polish (Self-Correction during drafting):** * *Correction:* Ensure I note that "Workbench" links are now "/playground". * *Correction:* The prompt generator moved to a Colab notebook. That's a clear shift from SaaS to open resource. 9. **Final Output Generation:** (Proceed to generate markdown in Chinese as requested).