← All reports

Changes on 2026-06-03

12 total changes in 2 runs

19:01 EST

🤖 AI Batch Analysis

### 整体摘要 版本 2.1.162 是一次侧重于稳定性和用户体验的更新,重点修复了 `claude agents` 的多项可靠性问题、Windows 平台兼容性(路径与权限)以及配置文件只读导致的启动挂起问题。 ### 核心变更主题 * **`claude agents` 可靠性大幅提升**:解决了会话附加失败、粘贴图片无效、消息队列丢失以及终端宽屏下的文本截断问题。 * **配置与跨平台修复**:修复了配置目录不可写时的静默挂起(现回退至内存模式),并修正了 Windows 下反斜杠路径及大小写不敏感的权限匹配逻辑。 * **工具链与集成优化**:修复了 LSP `workspaceSymbol` 查询无结果、MCP 超时配置(<1000ms 现被忽略)及 WebFetch 权限规则优先级问题。 * **用户体验 (UX) 细节**:启动信息更精简,错误提示更紧凑;斜杠命令补全行为改为“先填入后执行”;远程控制状态改为常驻底部显示。 ### 影响等级 **Medium**(中等) * **理由**:虽然包含大量修复,但大多为非破坏性改进。主要影响在于解决了关键的启动挂起 bug 和 Windows 兼容性问题,但斜杠命令自动补全的行为变更需要用户适应操作习惯。 ### 开发者行动项 1. **更新以解决启动挂起**:如果你的环境曾出现配置目录只读导致白屏的问题,更新后可解决(现使用内存配置并提示错误)。 2. **适应交互变化**:注意使用斜杠命令自动补全时,现在需要按 **Enter** 键执行,而非立即运行。 3. **检查 Windows 权限规则**:验证之前因反斜杠路径(`~\`)未能生效的“读取拒绝”规则是否现在正确屏蔽了 Glob/Grep 结果。 4. **审查 MCP 超时配置**:如果你设置了低于 1000ms 的 `timeout`,请将其修改为大于 1000ms 或依赖默认值,因为低值现在会被忽略。 5. **利用新的 JSON 字段**:如果使用脚本解析 `claude agents --json`,现在可以利用新增的 `waitingFor` 字段来判断会话阻塞原因。
💭 Model Reasoning
1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Input:** A diff of `CHANGELOG.md.md` (v2.1.162). * **Output Format:** Markdown, Chinese response, specific sections (Overall Summary, Key Themes, Impact level, Action items). * **Constraint:** Concise, insightful, focus on what matters to developers. 2. **Analyze the Diff (CHANGELOG.md.md - Version 2.1.162):** * *New feature 1:* `claude agents --json` now includes `waitingFor`. (API/Scripting improvement). * *New feature 2:* `--tools` explicitly listing Grep/Glob now works on native builds with embedded search. (Tooling fix/feature). * *New feature 3:* `/effort` confirms default persistence. (UX improvement). * *New feature 4:* Slash command autocomplete behavior change (fill vs run). (UX change). * *New feature 5:* Remote Control UI change (persistent footer pill). (UX). * *Rebrand:* Renamed "Windsurf" to "Devin Desktop" in menus. (Naming). * *Bug Fix 1:* Config directory read-only hang -> starts with in-memory config. (Critical stability fix). * *Bug Fix 2:* WebFetch permission rules applied correctly to built-in domains. (Security/Policy fix). * *Bug Fix 3:* Windows path handling (backslashes, case sensitivity) and Read deny rules for Glob/Grep. (Windows compatibility). * *Bug Fix 4:* Interrupt (Esc) at start of turn dropped. (Control fix). * *Bug Fix 5:* API 400 errors for classifier/MCP truncation boundaries with emoji. (Encoding fix). * *Bug Fix 6:* MCP `timeout` config < 1000ms floored to 1s -> now ignored/fallback. (Config fix). * *Bug Fix 7:* LSP `workspaceSymbol` returning no results -> accepts `query` param. (LSP/Tooling fix). * *Bug Fix 8:* `claude agents` status text truncation fixed. (UX). * *Bug Fix 9:* `claude agents` session name truncation fixed. (UX). * *Bug Fix 10:* `claude agents` attach bouncing back. (Reliability). * *Bug Fix 11:* `claude agents` Ctrl+V image paste fix. (UX). * *Bug Fix 12:* Backgrounding session losing conversation if service fails. (Data integrity). * *Bug Fix 13:* Failed replies in agents view now queued. (Reliability). * *Bug Fix 14:* Cross-session messaging breaking with deep TMPDIRs. (Reliability). * *Bug Fix 15:* Opening running background session stalling. (Performance). * *UX Polish:* Quieter startup, clearer warnings, pinned launch prompts, compact failed turn errors. (UX). * *Security/Performance:* Endpoint-security scanning wait logic, spawn failure error reporting. (Reliability). * *Cleanup:* Removed specific startup messages. 3. **Synthesize Analysis:** * **Overall Summary:** This release (v2.1.162) is a heavy "quality of life" and stability update. It focuses heavily on fixing `claude agents` reliability, Windows compatibility, and better error handling (especially regarding permissions and startup states). * **Key Themes:** * *Agents CLI Reliability:* Multiple fixes for attaching, pasting, message queuing, and display truncation in the `claude agents` view. * *Cross-Platform Stability:* Critical fix for startup hangs on read-only configs (in-memory fallback) and Windows path handling (backslashes/case). * *Tooling & Integration:* Fixes for LSP workspace symbols, MCP timeout handling, and WebFetch permission precedence. * *UX Polish:* Quieter startup, clearer error messages, and persistent UI indicators (Remote Control). * **Impact Level:** **Medium**. While there are many fixes, they are mostly non-breaking. However, the startup hang fix and Windows path fixes are significant for stability. The autocomplete change (fill vs run) is a behavioral change users need to adapt to. * **Action Items:** * Update to v2.1.162 to prevent startup hangs if your config directory has issues. * Be aware of the slash command autocomplete behavior change (press Enter to execute). * If using Windows, verify permission rules now work with backslash paths. * Adjust MCP server timeout configurations if setting values below 1000ms (they are now ignored). * Check `claude agents --json` output if you rely on tracking blocked sessions via `waitingFor`. 4. **Drafting the Output (in Chinese):** * *Summary:* 2.1.162 版本是一个侧重于稳定性和用户体验的更新,主要修复了 `claude agents` 的可靠性问题、Windows 路径兼容性以及启动时的挂起问题。 * *Themes:* * `claude agents` 稳定性:解决了附加、粘贴、消息队列丢失和终端显示截断等问题。 * 跨平台与配置:修复了配置目录只读导致的启动挂起(现在回退到内存配置),以及 Windows 反斜杠路径匹配问题。 * 工具与集成:修复了 LSP `workspaceSymbol`、MCP 超时配置(<1000ms 被忽略)及 WebFetch 权限规则。 * UX 细节:优化了启动时的警告信息、错误显示方式及斜杠命令的自动补全行为。 * *Impact:* Medium. (Reasoning: High number of reliability fixes, but mostly non-breaking. Autocomplete behavior change requires user adjustment.) * *Action Items:* * 更新版本以修复只读配置目录导致的启动挂起。 * 注意斜杠命令自动补全的行为变化(现在填入而非直接执行,需按 Enter 运行)。 * 如果在 Windows 上使用,检查权限规则现在是否正确匹配反斜杠路径。 * 如果 MCP 服务器配置了小于 1000ms 的超时时间,请调整配置(现在会被忽略)。 * 利用 `claude agents --json` 中新增的 `waitingFor` 字段来监控会话阻塞原因。 5. **Final Review:** Does it meet "world-class programmer" standards? Yes, it identifies the technical nuances (like the timeout behavior change and the specific Windows path fixes) rather than just listing the changelog. 6. **Final Output Generation.** (Proceed to generate markdown).

📚 Claude Code GitHub

16:14 EST

🤖 AI Batch Analysis

# 文档变更分析 ## 总体摘要 此次更新主要增强了错误处理的标准化,引入了新的 `overloaded` 错误类别,并调整了 Sub-agent 的默认行为(`/fork` 命令默认开启)。同时,文档明确了不同平台(特别是 Bedrock)的工具和 API 限制差异,并新增了关于终端渲染问题的故障排除指南。 ## 关键变更主题 * **错误处理增强**:在 `system/api_retry` 事件和 `StopFailure` hooks 中新增了 `"overloaded"` 错误类别,用于标识服务过载情况。 * **Sub-agent 行为变更**:从 v2.1.161 开始,`/fork` 命令默认启用;若需将其设为模型的默认生成行为,仍需通过环境变量开启(实验性功能)。 * **平台限制差异化**:明确区分了 Claude API (10MB) 与 Bedrock/Vertex AI (5MB) 的图片上传限制,并指出 Bedrock 不支持 WebSearch 工具。 * **可观测性优化**:新增 `OTEL_METRICS_INCLUDE_RESOURCE_ATTRIBUTES` 环境变量,允许开发者更精细地控制 OpenTelemetry 指标的基数。 * **用户体验与故障排除**:新增针对编辑器集成终端(如 VS Code)中文本乱码的修复建议;MCP 连接器列表在未登录状态下将自动折叠。 ## 影响等级:中等 **理由**:虽然大部分更改为文档澄清或非破坏性功能增强,但引入新的错误类型(`overloaded`)意味着依赖事件流或 Hooks 的开发者可能需要更新处理逻辑。此外,`/fork` 命令的默认启用改变了 CLI 的交互界面,尽管不破坏现有代码,但属于行为层面的变更。 ## 开发者行动建议 1. **更新错误处理逻辑**:检查处理 `system/api_retry` 事件和 `StopFailure` hooks 的代码,确保能正确识别并处理新增的 `"overloaded"` 错误类型。 2. **注意图片上传限制**:在使用 Claude API 时可利用新的 10MB 图片上限,但在 Bedrock 或 Vertex AI 上仍需保持在 5MB 以下。 3. **验证 Bedrock 功能集**:若使用 Bedrock,请知悉目前无法使用 WebSearch 工具,需调整相关工作流。 4. **配置 OTEL 指标**:如果担心自定义资源属性导致指标基数过高,评估设置 `OTEL_METRICS_INCLUDE_RESOURCE_ATTRIBUTES=false`。 5. **检查终端渲染问题**:若在 VS Code 或 Cursor 的集成终端中遇到乱码,尝试运行 `/terminal-setup` 或禁用 GPU 加速。
💭 Model Reasoning
1. **Analyze the Request:** * **Role:** World-class programmer analyzing documentation changes. * **Task:** Analyze a batch of 11 documentation pages as a single unit. * **Output Format:** Markdown, concise, insightful, Chinese. * **Structure:** 1. Overall Summary (1-2 sentences). 2. Key Themes (bullets). 3. Impact Level (Low/Medium/High/Breaking) with reasoning. 4. Action Items for developers (bullets, max 5). 2. **Analyze the Diffs (Page by Page):** * **Page 1: `amazon-bedrock.md`** * *Change:* Added a note that the WebSearch tool is not available on Bedrock. * *Impact:* Clarifies feature availability for Bedrock users. * **Page 2: `chrome.md`** * *Change:* Updated troubleshooting text. Instead of "Chrome extension not detected", it now refers to checking the "setup-issues line lists `chrome`". * *Impact:* Better UX/Debugging guidance. * **Page 3: `headless.md`** * *Change:* Updated `system/api_retry` event schema. Added `"overloaded"` to the `error` category list. * *Impact:* API integration developers need to handle this new error type in their event listeners. * **Page 4: `hooks.md`** * *Change:* Updated table for "Event" matchers. Added `"overloaded"` to `StopFailure` error types. Changed `CwdChanged` and `MessageDisplay` to move around slightly (though the diff looks truncated in the provided context, looking at `StopFailure` is the key). * *Impact:* Consistent error handling updates (`overloaded`). * **Page 5: `hooks-guide.md`** * *Change:* Similar to `hooks.md`. Added `"overloaded"` to `StopFailure` error types. * *Impact:* Consistent error handling updates. * **Page 6: `mcp.md`** * *Change:* Added info about collapsing unused connectors in v2.1.161. * *Impact:* UI/UX improvement for MCP connector management. * **Page 7: `monitoring-usage.md`** * *Change:* 1. Added `OTEL_METRICS_INCLUDE_RESOURCE_ATTRIBUTES` env var. 2. Clarified behavior of custom resource attributes vs standard attributes. 3. Updated standard attributes table (diff truncated in provided context, but shows expansion). * *Impact:* Enhanced observability/control for monitoring setups. * **Page 8: `settings.md`** * *Change:* The diff is massive (+96 / -95). Looking at the snippet, it seems to be a table definition update (likely just reformatting or slight description updates). The snippet provided is cut off in the middle (`...set CLAUDE_CODE_DISABLE_THINKING...`). Without the full diff, I must infer it's likely documentation refinement or schema updates. Given the other diffs, it's likely formatting consistency or adding minor descriptions. * *Impact:* Documentation clarity. * **Page 9: `sub-agents.md`** * *Change:* 1. Updated "Fork the current conversation" section. 2. Changed min-version requirement: From v2.1.117 to v2.1.161. 3. Behavior change: `/fork` command is now *enabled by default* in v2.1.161+. Previously required env var. 4. The env var `CLAUDE_CODE_FORK_SUBAGENT` now makes forks the *default spawn behavior* (experimental), rather than just enabling the command. * *Impact:* Significant behavioral change for subagents/forking. The `/fork` command is now available by default. * **Page 10: `troubleshooting.md`** * *Change:* Added a section on "Garbled or corrupted text in an editor's integrated terminal". Suggests running `/terminal-setup` or disabling GPU acceleration. * *Impact:* Fixes a common rendering issue. * **Page 11: `build-with-claude/vision.md`** * *Change:* 1. Updated image size limits. 2. API: 5MB -> 10MB. 3. Bedrock/Vertex: 5MB (explicitly stated). 4. Added distinction between platforms for per-image limits vs request size limits. * *Impact:* Increased upload limit for direct API usage. 3. **Synthesize Key Themes:** * **Error Handling:** Introduction of a new error type `"overloaded"` across Headless, Hooks, and Hooks Guide. * **Feature Availability/Boundaries:** Bedrock limitations (WebSearch), Vision limits (API vs Bedrock/Vertex). * **Feature Changes/Defaults:** Sub-agent forking is now default (command available), GPU acceleration troubleshooting. * **Observability:** New environment variable for OTEL resource attributes. * **UX/Refinement:** Chrome setup issues text, MCP connector collapsing. 4. **Determine Impact Level:** * Most changes are clarifications or non-breaking additions. * **Potential Breaking/Behavioral Change:** The `/fork` command is now enabled by default in v2.1.161. While not strictly breaking code, it changes the available commands and default behavior if the user relied on it being hidden. The `overloaded` error type is a new value, usually non-breaking. * **Upgrade:** Image limits increased (non-breaking). * **Verdict:** **Medium**. The changes in sub-agent behavior (defaulting `/fork`) and the addition of the `overloaded` error type require attention from developers integrating these features. The vision limit increase is a benefit. 5. **Draft Action Items:** * Handle the new `"overloaded"` error category in `system/api_retry` events and `StopFailure` hooks. * Note that `/fork` command is now available by default (v2.1.161+) without env vars. * Update vision API clients to utilize the new 10MB image limit where applicable. * Be aware that Bedrock users cannot use WebSearch tool. * Consider new `OTEL_METRICS_INCLUDE_RESOURCE_ATTRIBUTES` env var for fine-tuned metric control. 6. **Translate to Chinese & Refine:** * *Summary:* 此次更新主要集中在错误处理增强、API 限制调整以及部分功能默认行为的变更。特别是引入了新的 "overloaded" 错误类型,并默认启用了 `/fork` 命令,同时提高了视觉 API 的图片上传限制。 * *Themes:* * 错误处理标准化:新增 `overloaded` 错误类别。 * 功能可用性明确:明确了 Bedrock 不支持 WebSearch 工具,区分了不同平台的图片大小限制。 * 交互与监控优化:Sub-agent `/fork` 命令默认开启,新增 OTEL 指标控制变量。 * 故障排除增强:新增针对集成终端乱码的解决方案。 * *Impact:* Medium。新增的错误类型需要开发者在集成中处理,且 Sub-agent 的默认行为发生了变化(命令默认可用),属于非破坏性但需要注意的功能升级。 * *Actions:* * 检查并更新处理 `system/api_retry` 事件和 `StopFailure` hook 的代码,以支持新的 `"overloaded"` 错误类型。 * 留意从 v2.1.161 版本开始,`/fork` 命令将默认可用,无需设置环境变量。 * 利用 Claude API 的新图片上传上限(10MB),同时注意 Bedrock 和 Vertex AI 仍为 5MB。 * 如果使用 Bedrock,请知悉 WebSearch 工具不可用。 * 如遇集成终端显示乱码,尝试运行 `/terminal-setup` 禁用 GPU 加速。 7. **Final Review against Constraints:** * Markdown format? Yes. * Concise? Yes. * Chinese? Yes. * Developer focus? Yes. * Action items max 5? Yes. *Self-Correction on `settings.md`:* The diff is huge but truncated. I shouldn't over-speculate. It seems mostly table formatting consistency. I won't list it as a key theme unless I see a specific setting change in the visible snippet. The snippet ends at `CLAUDE_CODE_DISABLE_THINKING`, so I'll assume it's documentation cleanup. *Refining the "Fork" action:* It's important to clarify that the *command* is default, but using it as the *default spawn behavior* still needs the env var (based on the text: "Making forks the model's *default* spawn behavior is experimental... enable it by setting the same variable"). I will be precise in the summary. *Refining Vision limits:* 10MB for Claude API, 5MB for Bedrock/Vertex. Important distinction for multi-platform devs. 8. **Final Output Generation.** (Proceeding to generate output based on these steps).

📚 Claude Code

📚 Anthropic API