Claude Code 2.1.271：程式碼審查不再重複發文，已解決問題不重複回報

繁體中文

問題修復

程式碼審查不再重複發文，已解決問題不重複回報

在 GitHub 上提交拉取請求等待審查時，最怕遇到審查遺漏或重複回報。現在限制每個拉取請求審查一次的儲存庫中，若在排隊等候時送入新修訂，會如實審查該次修訂。當 GitHub 回報錯誤但實際已建立審查時，不再重複張貼兩三次相同的意見。針對先前已由真人標記解決的資安問題，即使後續程式碼行數挪移，追蹤審查也不會重新貼出。在應用程式中重新打開已完成的 /ultrareview 雲端工作階段（session），也不會無故重頭跑完整輪審查。

新功能

Claude Tag 串流討論不中斷記憶，通知與環境選單更清楚

在通訊軟體群組裡呼叫 Claude Tag 協助工作時，多討論串的訊息管理變得更加可靠。在持續啟用的頻道中，即使對話主要發生在討論串內，討論串活動現在能維持工作上下文，不再每隔約一小時就重置記憶。指派 Claude 監控拉取請求的討論串，在重啟後能持續接收持續整合（CI）失敗、留言與審查通知。刪除已獲回覆的討論串首則訊息時，Claude 會如期終止該處的工作。不再因為其他機器人的舊指令而扣留發文，只有指名給 Claude 的指令才生效，不確定時會主動詢問。加入熱門頻道時的卡片會說明真實原因，不再誤稱看見大量自動發文。管理設定中的環境挑選器明確標示 Anthropic 託管或自託管選項並提供編輯連結。

新功能

網頁版例行工作頁面改版，雙分頁呈現執行狀態

登入網頁版管理自動排程任務時，例行工作（routines）頁面帶來全新架構。在 claude.ai/code 頁面上，版面劃分為「你的例行工作」與「範本」兩大分頁，並以雙欄卡片直觀呈現每項例行工作的執行狀態。同時移除了先前的日曆視圖，讓清單檢視更為俐落。

新功能

雲端環境支援自訂網域白名單，離線工作階段發話即重啟

管理雲端開發環境時，現在擁有更細緻的網路存取控制與快速恢復機制。在管理員設定的雲端環境編輯器中，新增了自訂網路存取選項，提供與 claude.ai/code 環境對話框相同的允許網域清單。管理頁面清楚列出 Claude Tag 與 Claude Code 的預設環境，附上切換連結並標註建議建立的類型。雲端工作階段（session）在背後行程結束但外觀仍保持連線時，以前可能卡住約十分鐘才回應，現在只要發送訊息就會立刻重新啟動。

新功能

VSCode 支援關閉自動附加檔案，工作階段清單更穩定

在 VSCode 編輯器中使用 Claude Code 寫程式時，對話傳送內容與歷史清單獲得多項體驗改善。新增「附加開啟中檔案（Attach Open File）」設定，關閉後不再自動將當前開啟的檔案內容塞入對話，但選取的文字片段依然會附加。工作區位於 Windows 網路磁碟機或 SUBST 虛擬磁碟時，修復歷史記錄只顯示目前工作階段（session）的問題。篩選選單同時勾選作用中與開啟時，不再隱藏處於閒置狀態的已開啟工作階段。工作階段清單重新整理時，新建的聊天視窗不會跳回上一筆對話。修改設定中的 CLAUDE_CONFIG_DIR 環境變數後，側邊欄與分頁會即時套用新目錄。切換開關的開啟狀態顏色從原本的 Claude 橘色改為跟隨編輯器主題的按鈕顏色。

新功能

VSCode 掛鉤儲存防呆升級，消除主控台視窗閃爍

在 VSCode 擴充套件中微調自動化規則時，設定對話框與背景視窗干擾問題已全面改善。掛鉤（hook，特定事件發生時被觸發的小程式）與權限規則對話框不再把已成功的儲存誤報為失敗，並修正受外掛政策鎖定時畫面反白或儲存錯誤顯示顏色代碼的異常。儲存掛鉤時替換不再產生重複項目，修改標頭大小寫時能保留金鑰機密，且返回前會自動將本機設定檔加入忽略清單。設定檔本身導致儲存受阻時，彈出視窗會附帶「開啟設定檔」按鈕與錯誤原因。在 Windows 上執行背景 Git 或登入檢查時，不再頻繁閃爍黑色的主控台（console）視窗。提示快取（cache，把算過或查過的結果暫存起來直接用）時鐘懸浮文字與自動壓縮圖示的提示框顯示也已修復。

問題修復

遠端連線支援快速模式，/fast 關閉與降速備援修復

在雲端或自託管環境連線執行 Claude Code 遠端工作階段（session）時，現在也能享受快速模式的加速體驗。只要組織政策允許，遠端主機的設定或在工作階段中輸入 /fast 即可啟用快速模式。當組織停用快速模式時，輸入 /fast off 現在會如實關閉，不再錯誤回應模式不可用。跳過組織檢查啟動的工作階段在 API 拒絕快速請求後，會明確顯示原因並停止每回合重送。遇上用量額度（quota）上限或系統過載時，重試監控機制會自動切換速度。系統改以標準速度作為備援（fallback），不再直接中斷或堅持高速重試。

新功能

企業政策即時更新，支援自訂模型計費倍率

在企業組織中統籌多位成員的開發環境時，政策派發與成本拆帳迎來更精確的控管機制。在 modelPricing 管理設定與應用程式閘道器中，新增高於 1 且最高至 10 的 multiplier 參數，方便計算內部拆帳。切換帳號或金鑰時，組織政策快取（cache，把查過的結果暫存起來直接用）會即時刷新。工作階段（session）中途政策載入完成或變更後，工具與指令清單會同步更新。當企業的 managed-mcp.json 設定檔無法解析時，會在啟動時發出警告並維持專屬管制，禁止載入非核准的 MCP（Model Context Protocol，讓 AI 連接外部工具的協定）伺服器。透過本機代理轉發組織政策時遭拒絕的問題也已修復。

新功能

自託管 Runner 支援退場標記，超大設定檔改存磁碟快照

自行架設伺服器執行自託管執行器（self-hosted runner）時，主機維護與大型專案設定更可靠。執行 claude self-hosted-runner --drain-marker-file <path> 後，只要指定檔案存在且收到 SIGTERM 退場信號，執行器會向伺服器通報 host drain，但這僅作 telemetry 通報，不代表額外調度行為。當主機設定資料夾超過 64 MiB 時，過去工作階段可能默默遺失整份 host config，包含 settings、skills、plugins、MCP servers。現在新增 --host-config-snapshot disk|memory，可把快照放在磁碟或記憶體。

新功能

沙盒指令獨立設定連線網域，自動模式審查行內指令

在自動模式搭配沙盒環境執行終端機操作時，網路安全界限變得更加嚴密。Bash、PowerShell 與 Monitor 指令在沙盒模式下新增個別指令的 allowed_domains 網域清單，指令所需的網路主機會與指令一同審查並僅向該指令開放，其他連線一律拒絕。技能或斜線指令中內嵌的 ! 開頭 Shell 指令，改為遵循預設模式的權限規則而非由分類器全權判定，若無規則適用的指令則作為需審查的工具呼叫執行。

問題修復

Bash 權限檢查涵蓋萬用字元，防堵跨目錄讀取漏洞

透過命令列執行 Bash 工具時，權限防護系統對檔案路徑與參數的識別更加縝密。當 fmt 或 column 等指令跟隨未識別的選項時，不再遺漏其讀取的檔案路徑。萬用字元（wildcard，用星號代表多個檔案的符號）位於指令模式或選項值中時，例如 grep -v dir/* file 這類指令展開後的檔案都會納入權限檢查。Shell 變數宣告旗標不再造成指令被偽裝誤判。包含兩次目錄切換、子 Shell 或 cd 與 git 連接的指令，在啟用禁止讀取工作目錄外部設定時，不會跳過確認提示。

新功能

外掛安裝支援 SHA-256 驗證，確保指令未被竄改

在自動化腳本中安裝或更新擴充外掛時，現在有比盲目確認更安全的核可機制。claude plugin install 與 claude plugin update 都新增 --accept-command <sha256> 參數，用來接受前一次 --json run 顯示的指令雜湊值。這會精確比對那次顯示的 command hash，取代原本直接用 -y 放行。

新功能

子代理支援精簡模式，動態工作流程達額度自動暫停

多代理拆專案時，子代理（subagent）少載規則，回報可控。自訂/外掛子代理可用 frontmatter/--agents JSON 裡的 omitClaudeMd，不載入使用者、專案、本機 CLAUDE.md，但載入管理政策。雲端工作階段（session）重啟工作單元後，不再拒絕子代理工具呼叫。動態工作流程遇到用量額度（quota）上限會暫停，重置後接續。自動模式改用 hand-back call 回報並由安全分類器審查。Pro 預設規模改為 small，medium 上限從 15 降到 10。claude-api 技能讓串流自訂工具啟用 eager_input_streaming，並用 user.define_outcome 開始 deliverable-shaped Managed Agents work。

問題修復

/config 支援全螢幕滑鼠滾輪，路徑與選取標籤優化

在終端機介面中調整系統偏好或輸入路徑時，互動操作變得更符合現代使用習慣。全螢幕 /config 設定面板加入滑鼠支援，可透過滾輪滾動清單、點擊數值直接切換，游標下方的列也會高亮顯示。/add-dir 路徑輸入支援左右方向鍵移動游標，按下 Enter 鍵只會加入實際輸入的路徑，不會誤加反白的自動補齊項。主提示詞外的文字欄位不再將開頭的驚嘆號 ! 移至末尾。修正全螢幕渲染殘留背景顏色的破圖。提示詞中的 IDE 選取標籤改為可隨文字換行的 [⧉ …] 膠囊標籤，按 Backspace 即可移除。

新功能

MCP 支援工具簡稱搜尋，長任務定時回報防止中斷

透過 MCP（Model Context Protocol，讓 AI 連接外部工具的協定）擴充工具庫時，伺服器連線與呼叫流程更為穩健。工具搜尋支援 Claude 以簡短名稱選取工具，不再強制使用完整前綴名稱。當 MCP 伺服器密集發送清單變更通知時，能有效防止 CPU 負載過高與重複請求。claude mcp serve 執行中的工具呼叫每 30 秒發送一次進度更新，避免因長時間無輸出而被閒置逾時中斷。雲端平台中標記為 alwaysLoad 的 MCP 伺服器在中途連線完成後，下一回合即可直接使用。恢復工具全來自 MCP 的單次提示詞工作階段（session）時不再報錯。

問題修復

MCP 授權註冊邏輯修復，快捷鍵不再中斷重新連線

連接需要身分驗證的第三方 MCP 伺服器時，授權交握與遠端遙控操作更加穩定。修正 MCP（Model Context Protocol，讓 AI 連接外部工具的協定）OAuth 授權用戶端註冊處理，解決拒絕同意時強制重新註冊、誤用其他轉址 URI，以及多重寫入刪除有效憑據或保留錯誤設定的問題。按下 Ctrl+O 快捷鍵不再意外取消正在排隊的 MCP 伺服器重新連線。修復在開啟對話紀錄視圖時，從遠端操控（Remote Control）發送 /mcp 指令失敗的異常。

新功能

產出物渲染為精美排版文件，監聽數量上限翻倍

利用產出物（artifacts）功能產出文件或程式碼時，預覽效果與檔案同步能力獲得顯著增強。發佈為產出物的 Markdown 檔案現在會渲染為具備標題列、排版樣式與語法高亮的精美文件頁面。單一工作階段（session）可同時監聽的發佈產出物上限從 5 個提升至 10 個，方便追蹤其他地方的重新發佈。背景工作階段會如實監聽自己發佈的產出物。透過 /artifacts 附加的產出物在以 --resume 恢復工作階段後不會消失。發佈錯誤訊息更清楚，未指定檔案時會提示先寫入檔案，並列出所有支援的功能規格。

問題修復

工作階段恢復記憶精準，模型切換不誤報快取遺失

中斷工作後想接續先前的開發進度時，各項工作階段（session）恢復與狀態同步問題已獲得修復。使用 /resume 或 /teleport 恢復工作階段時，不會殘留前次對話的檔案讀取追蹤，確保權限狀態乾淨。使用 --resume 恢復工作階段時，即使模型系列與預設不同，也能正確保留百萬 token（AI 處理與計費的文字單位）的上下文視窗。使用 /model 切換回對話原本運行的模型時，不再誤報即將遺失快取（cache，把查過的結果暫存起來直接用）。對話壓縮後，仍在運行的背景指令不會被重複啟動第二份。跨工作階段傳送訊息時提供明確遞送狀態通知，而在高度較矮的終端機上使用全螢幕選單也能完整列出工作階段。

新功能

掛鉤執行顯示耗時進度，支援 Esc 取消等待

設定自動化掛鉤（hook，特定事件發生時被觸發的小程式）時，執行狀態反饋更加即時透明。當 SessionStart、UserPromptSubmit、PreToolUse 或 SessionEnd 掛鉤執行時，狀態轉輪會明確顯示執行狀態與已耗費時間。在等待 SessionStart 掛鉤完成期間，可隨時按下 Esc 鍵取消當前提示詞。修復互動式 /hooks 選單在掛鉤比對名稱包含 __proto__ 或 constructor 等保留屬性時發生的當機問題。

新功能

轉輪提示長思考狀態，引導下載桌面與行動版

等待 AI 運算複雜任務時，狀態列提供更清晰的思考進度與多裝置應用指引。思考時間超過 45 秒時，狀態轉輪會顯示陷入沉思。自輸出 token（AI 處理與計費的文字單位）限制恢復時，則顯示接續思考。輸入 /desktop 指令可取得桌面版應用程式下載連結，並針對雲端平台使用者提供專屬引導提示。輸入 /mobile 指令時顯示單一 QR Code 網址，自動導向手機對應的應用程式商店。修正轉輪提示推薦目前帳號類型無法使用或工作階段中已停用指令的問題。

問題修復

跨作業系統檔案監聽更穩固，Git 鎖定檔殘留修復

在 macOS、Linux 與 Windows 等多種環境下開發時，底層系統檔案與終端機操作更流暢。Linux 上沙盒指令啟動失敗後，不再殘留 .git/config.lock 鎖定檔而導致工作階段（session）後續 Git 指令失效。macOS 系統檔案事件飽和時，設定檔監聽器會自動改用輪詢機制。系統會以此作為備援（fallback），即時偵測外部變更。修正 Windows 加密虛擬磁碟回報 inode 0 時自訂代理與斜線指令無法載入的問題。登出後未於清理天數內重新整理的同步技能，下次啟動時會移至可復原的垃圾桶。修復特殊終端機的鍵位以及退出時終端機查詢字串滲漏至命令列的異常。

新功能

Claude in Chrome 提示休眠狀態，遠端連線減少殘留

透過瀏覽器擴充套件或遠端操控（Remote Control）與 Claude 互動時，連線錯誤訊息更加貼近實際狀況。雲端工作階段無法連線至本機瀏覽器時，提示訊息會先說明電腦可能處於休眠狀態，再建議安裝擴充功能。工具搜尋不可用時，不再指示模型透過 ToolSearch 載入工具。在不穩定網路環境下建立遠端操控失敗時，減少在 claude.ai 上殘留的空白工作階段（session）。

新功能

終端機長文渲染不再卡頓，監聽任務設有逾時上限

面對龐大程式碼變更與長時間監控任務時，命令列工具的反應速度與資源管理更有保障。終端機渲染效能改善，大量差異比對與長對話紀錄顯示更加流暢。介面減少掉格卡頓（jank，短暫畫面停頓現象）。啟動時略過內建模型資料的重複驗證，略微縮短啟動時間。Monitor 監聽機制改為強制設定時間上限（一般工作階段最多 30 分鐘，單次提示詞模式為 10 分鐘），到期前提醒 Claude 重新武裝，取代過去無逾時的常駐選項。Windows 環境下暫存輸出路徑達到 260 字元上限時，修復 PowerShell 指令報錯 Exit code 1 且無輸出的問題。

新功能

LLM 閘道相容純文字回應，切換目錄精確重計技能

串接大型語言模型閘道器或切換工作目錄時，資料格式解析與環境清理更為精準。當大型語言模型（LLM）閘道器將非串流回覆以 text/plain 純文字格式返回時，對話回合不再報錯為空回應或格式損毀。執行 /cd 切換目錄後，/reload-skills 回報的技能數量與斜線選單完全一致。引用 PDF 檔案時若無法計算頁數，會明確標註「頁數未知」，不再根據檔案大小猜測。Claude 應用程式閘道器與第三方雲端平台（Bedrock、Vertex AI、Foundry）的工作階段，不再刷新多餘且未使用的 claude.ai 登入狀態。

English

BUGFIX

Code Review Stops Duplicate Posts and Won't Re-report Resolved Findings

When submitting pull requests on GitHub for review, missing reviews or duplicate reports are frustrating. In repositories configured to review once per PR, if a new commit arrives while waiting in the queue, it now properly reviews that requested commit. When GitHub reports an error for a review that was actually created, it no longer repeatedly posts the same findings two or three times. For security findings previously resolved by a human, follow-up reviews will not re-post them even if subsequent code line numbers shift. Reopening a completed /ultrareview cloud session in the Claude app will no longer restart the entire review unprompted.

FEATURE

Claude Tag Preserves Thread Context and Clarifies Notifications and Environment Menus

When invoking Claude Tag in chat groups to assist with work, managing messages across multiple threads is now far more reliable. In channels where Claude stays continuously active, even if conversations mostly take place within threads, thread activity now preserves working context instead of resetting memory roughly once every hour. A thread asking Claude to monitor a pull request continues to receive CI failure, comment, and review notifications even after restarting Claude. When the first message of a thread that Claude already replied to is deleted, Claude terminates its work there as expected. Claude no longer holds back posts due to earlier instructions addressed to other bots; only instructions addressed to Claude take effect, and it proactively asks when unsure. The card posted upon joining a busy channel now explains the real reason rather than incorrectly claiming it sees a lot of automated posts. The Environment picker in admin settings clearly labels Anthropic-hosted or self-hosted options and provides links to edit or create them.

FEATURE

Web Routines Page Redesigned with Dual Tabs Displaying Run Status

When managing automated scheduled tasks on the web, the Routines page introduces a fresh layout. On claude.ai/code, the layout is split into two major tabs, "Yours" and "Templates", using two-column routine cards that intuitively display the run status of each routine. The previous calendar view has also been removed to keep the list view streamlined.

FEATURE

Cloud Environments Add Custom Domain Whitelisting, Offline Sessions Restart on Message

Managing cloud development environments now offers finer network access controls and a rapid recovery mechanism. In the Cloud environments editor within admin settings, a Custom network access option has been added, offering the same allowed-domains list available in the claude.ai/code environment dialog. The admin page clearly lists the default environments for Claude Tag and Claude Code, complete with links to change them and labels indicating recommended types to create. When a cloud session's background process exits while the session still appears connected, it previously took about ten minutes to respond; now sending a message restarts it immediately.

FEATURE

VSCode Supports Disabling Auto-Attached Files and Stabilizes Session List

When coding with Claude Code in VSCode, message attachments and history lists receive several user experience improvements. A new "Attach Open File" setting allows turning off the automatic inclusion of currently open files in messages, while explicitly selected text is still attached. Fixed an issue where session history only showed the current session when the workspace was located on a Windows mapped network drive or SUBST virtual drive. When both Active and Open are checked in the filter menu, open idle sessions are no longer hidden. When the session list refreshes, newly created chat windows no longer switch back to the previous conversation. After changing the CLAUDE_CONFIG_DIR environment variable in settings, the sidebar and open tabs immediately apply the new directory. The "on" state color of toggle switches has also changed from Claude orange to the editor theme's button color.

FEATURE

VSCode Upgrades Hook Save Guards and Eliminates Flashing Console Windows

When fine-tuning automation rules in the VSCode extension, settings dialogs and background window disturbances are now resolved. The Hooks and Permission rules dialogs no longer falsely report successful saves as failed, and issues causing the screen to go blank under plugin policy locks or showing color codes in save errors are fixed. Saving hooks on replace no longer produces duplicates, retyping header names in different capitalization preserves secrets, and the local settings file is added to gitignore before the save returns. When the settings file itself blocks saving, the popup includes an "Open settings file" button alongside the error reason. Background Git, ripgrep, or sign-in checks on Windows no longer cause black console windows to flash frequently. Prompt cache clock hover text delays and duplicate tooltips on the auto-compact icon are also fixed.

BUGFIX

Remote Connections Add Fast Mode, Fixing /fast Disabling and Fallback Speeds

When running Claude Code Remote sessions in cloud or self-hosted environments, users can now enjoy the accelerated experience of fast mode. Where organization policy allows, fast mode can be enabled via host settings or by typing /fast directly in the session. When the organization disables fast mode, typing /fast off now turns it off properly instead of returning an error stating the mode is unavailable. Sessions started by skipping organization checks will display the rejection reason and stop resending fast requests every turn once the API rejects them. When encountering usage quota limits or system overloads, the retry watchdog automatically switches speeds. The system falls back to standard speed rather than failing the turn or insisting on fast retries.

FEATURE

Enterprise Policies Update Instantly with Custom Model Pricing Multipliers

Managing development environments across organizations now features more precise policy distribution and cost chargebacks. A multiplier parameter above 1 and up to 10 has been added to the modelPricing managed setting and the Claude apps gateway pricing block to facilitate marked-up internal chargeback rates. When switching accounts, organizations, or API keys, cached organization policies refresh immediately. When policies finish loading after startup or change mid-session, tool and command lists update synchronously. When an enterprise managed-mcp.json file cannot be parsed, it issues a warning at startup while maintaining exclusive control, blocking unapproved MCP servers from loading. Fixed an issue where organization policies were rejected when fetched through third-party local proxies configured via ANTHROPIC_UNIX_SOCKET.

FEATURE

Self-Hosted Runners Support Drain Markers and Store Oversized Configs on Disk

When operating self-hosted runners on your own servers, host maintenance and large project setups are significantly more reliable. Running claude self-hosted-runner --drain-marker-file <path> allows the runner to report its exit to the server as a host drain upon receiving a SIGTERM signal if the specified file exists, though this functions strictly as telemetry reporting without triggering extra scheduling. When the host config directory exceeds 64 MiB, sessions previously suffered from silently losing the entire host config, including settings, skills, plugins, and MCP servers. A new --host-config-snapshot disk|memory option now allows placing config snapshots on disk or in memory.

FEATURE

Sandboxed Commands Configure Domain Access Independently, Auto Mode Reviews Inline Commands

When executing terminal operations in auto mode with sandboxing, network security boundaries are much tighter. Bash, PowerShell, and Monitor commands in sandboxed mode now include per-command allowed_domains lists, ensuring required network hosts are reviewed alongside each command and opened exclusively for it while all other hosts are refused. Inline ! shell commands embedded inside skills or slash commands now follow default-mode permission rules rather than being evaluated entirely by the classifier; commands not covered by any rule execute as reviewed tool calls.

BUGFIX

Bash Permission Checks Cover Wildcards to Prevent Cross-Directory Read Leaks

When running Bash tools from the command line, permission guards identify file paths and arguments with greater precision. When commands like fmt or column are followed by unrecognized options, the file paths they read are no longer missed. When wildcards sit inside a command's pattern or option value, such as in grep -v dir/* file, all expanded files are now included in permission checks. Shell variable declaration flags can no longer misrepresent the command being run. Commands containing two directory changes, a subshell, or a cd and git chain will not bypass confirmation prompts when permissions.blockReadsOutsideWorkingDirectories is enabled in bypass or auto mode.

FEATURE

Plugin Installations Support SHA-256 Verification to Prevent Tampered Commands

When installing or updating plugins within automated scripts, a safer verification mechanism replaces blind confirmations. Both claude plugin install and claude plugin update have added the --accept-command <sha256> parameter to accept the exact command hash displayed in a previous --json run. This matches the displayed command hash precisely instead of passing -y unconditionally.

FEATURE

Subagents Support Lean Mode, Dynamic Workflows Auto-Pause at Quota Limits

When splitting projects across multiple agents, subagents load fewer rules and provide controllable reporting. Custom and plugin subagents can use omitClaudeMd in agent frontmatter or --agents JSON to run without loading user, project, or local CLAUDE.md files, while managed policy files still load. Cloud sessions no longer reject subagent tool calls after worker restarts. Dynamic workflows pause upon hitting usage limits and resume automatically when reset, rather than dropping agents. Auto mode now reports back via dedicated hand-back calls reviewed by the safety classifier. The default dynamic workflow size on Pro plans is now small, and the medium guideline is lowered from 15 to 10 agents. The claude-api skill enables eager_input_streaming on streaming custom tools and initiates deliverable-shaped Managed Agents work with user.define_outcome.

BUGFIX

Fullscreen /config Adds Mouse Support, Optimizing Paths and Selection Pills

Adjusting system preferences or typing paths in the terminal is now far more ergonomic. The fullscreen /config panel adds mouse support, allowing the scroll wheel to navigate settings, clicking values to toggle them, and highlighting the row under the pointer. The /add-dir path input supports left and right arrow keys to move the cursor, and pressing Enter adds only the typed path instead of including highlighted autocompletions. Text fields outside the main prompt no longer shift a leading exclamation mark ! to the end of your input. Fixed a fullscreen rendering glitch where text retained a stale background color. The IDE selection indicator in prompts is now a [⧉ …] pill that wraps cleanly with text and can be removed with Backspace.

FEATURE

MCP Supports Bare Name Tool Searches and Periodic Updates for Long Tasks

Expanding toolsets via MCP now provides more reliable server connections and execution flows. Tool search allows Claude to select tools by bare names without requiring full prefixed names. When an MCP server sends rapid list change notifications, sustained high CPU usage and duplicate requests are effectively prevented. Running tool calls in claude mcp serve send progress updates every 30 seconds, preventing idle timeouts from aborting long commands that produce no output. MCP servers marked alwaysLoad in cloud platforms can be used on the next turn once connected mid-conversation. Resuming single-prompt sessions whose tools come entirely from MCP servers no longer throws errors.

BUGFIX

MCP OAuth Registration Logic Fixed, Shortcuts Won't Cancel Reconnects

Connecting to third-party MCP servers requiring authentication is now much more stable during handshakes and remote control. MCP OAuth client registration handling is fixed, resolving issues where denying consent forced re-registration, reused registrations for different redirect URIs, or, during concurrent writes, deleted valid credentials or kept a mismatched registration. Pressing Ctrl+O no longer cancels pending MCP server reconnections. Fixed an error where /mcp commands sent from Remote Control failed while the transcript view was open.

FEATURE

Artifacts Render as Styled Documents, Doubling Watched Artifact Limits

Preview quality and file synchronization for documents and code generated via artifacts receive significant upgrades. Markdown files published as artifacts now render as styled document pages with title headers, document typography, and syntax-highlighted code. The maximum number of published artifacts a single session can watch simultaneously has increased from 5 to 10, facilitating tracking of republishes made elsewhere. Background sessions now watch artifacts they publish. Artifacts attached via /artifacts no longer disappear after resuming a session with --resume. Publish error messages are clearer, prompting users to write to a file first when none is specified and listing all supported capabilities.

BUGFIX

Precise Session Resumes Prevent False Cache Loss Warnings on Model Switches

Resuming earlier development work after interruptions is now smoother thanks to multiple session recovery fixes. Resuming sessions via /resume or /teleport no longer retains previous file-read tracking, ensuring clean permission states. Resuming via --resume preserves the 1M token context window even when the resumed session's model family differs from the configured default. Switching back to the model a conversation originally ran on via /model no longer falsely warns about losing the conversation cache. After conversation compaction, running background commands are not duplicated. Cross-session messaging provides clear delivery notifications, and fullscreen menus on short terminals display session lists properly.

FEATURE

Hook Executions Display Elapsed Time, Supporting Esc to Cancel Waiting

Configuring automation hooks provides more immediate and transparent feedback. While a SessionStart, UserPromptSubmit, PreToolUse, or SessionEnd hook runs, the status spinner clearly shows its execution status alongside elapsed time. While waiting for a SessionStart hook to complete, pressing Esc cancels the prompt at any time. Fixed crashes in the interactive /hooks menu when a hook matcher is named after inherited properties like __proto__ or constructor.

FEATURE

Spinner Displays Long Thinking Status and Guides Desktop and Mobile Downloads

Waiting for AI computations during complex tasks now offers clearer progress indicators and multi-device app guidance. When thinking exceeds 45 seconds, the spinner displays "deep in thought". While recovering from output-token limits, it displays "picking the thought back up". Typing /desktop provides download links for the desktop app alongside dedicated tips for cloud platform users. Typing /mobile displays a single QR code URL that automatically routes to the appropriate app store for your phone. Fixed spinner tips recommending commands unavailable for the current account type or disabled in the session.

BUGFIX

Cross-Platform File Watching Stabilized, Stale Git Lock Files Fixed

Operating across macOS, Linux, and Windows now delivers smoother low-level file and terminal handling. On Linux, when a sandboxed command fails to start, stale .git/config.lock files are no longer left behind to break subsequent Git commands in the session. When system file-event services are saturated on macOS, the config watcher automatically falls back to polling to detect external changes. Fixed custom agents and slash commands failing to load from virtual drives reporting inode 0 on Windows. Synced skills not refreshed within cleanupPeriodDays after signing out move to the recoverable trash upon the next launch. Fixed keybindings in specialized terminals and terminal query strings leaking to the shell prompt upon exit.

FEATURE

Claude in Chrome Notes Sleep State, Remote Control Leaves Fewer Empty Sessions

When interacting with Claude via browser extensions or Remote Control, connection error messages reflect real conditions more accurately. When a cloud session cannot reach the local browser, the message notes that the computer may be asleep before suggesting an extension install. When ToolSearch is unavailable, it no longer instructs the model to load tools via ToolSearch. When Remote Control setup fails on unstable networks, fewer empty sessions are left behind on claude.ai.

FEATURE

Terminal Renders Large Diffs Smoothly, Monitor Watches Enforce Deadlines

When dealing with massive code diffs and long-running monitoring tasks, command-line responsiveness and resource management are much more reliable. Terminal rendering performance is improved so large diffs and long transcripts render faster with fewer frame drops. Startup time is slightly reduced by skipping redundant validation of built-in model data. Monitor watches now enforce a mandatory deadline (at most 30 minutes in regular sessions, 10 minutes in single-prompt -p runs) and notify Claude to re-arm, replacing the previous no-timeout persistent option. Fixed an issue on Windows where PowerShell commands failed with "Exit code 1" and no output when the session temp output path reached the 260-character limit.

FEATURE

LLM Gateways Support Plaintext Replies, Skill Counts Stay Accurate After Directory Changes

Connecting through LLM gateways or switching directories now features more accurate formatting and environment cleanup. When an LLM gateway returns non-streaming replies as text/plain, turns no longer fail with empty or malformed response errors. After changing directories via /cd, /reload-skills reports a skill count that perfectly matches the slash menu. When referencing PDF files where page count cannot be determined, it explicitly notes "page count unknown" rather than guessing based on file size. Sessions on the Claude apps gateway and third-party cloud platforms (Bedrock, Vertex AI, and Foundry) no longer refresh unused leftover claude.ai logins.

原始 CHANGELOG

Original changelog程式碼審查不再重複發文，已解決問題不重複回報

[Code Review] Fixed a pull request in a repository reviewed once per PR sometimes getting no review when a commit arrived while its review was waiting to start; it now reviews the requested commit
[Code Review] Fixed Code Review occasionally posting the same findings two or three times when GitHub reported an error for a review it had in fact created
[Code Review] Fixed follow-up reviews re-posting a security finding a person had already resolved when a later push moved the lines it was anchored to
[Code Review] Fixed reopening a finished /ultrareview cloud session in the Claude app starting the whole review over again unprompted

Original changelogClaude Tag 串流討論不中斷記憶，通知與環境選單更清楚

[Claude Tag] Fixed Claude in a channel where it stays active losing its working context about once an hour when the conversation is mostly in threads; thread activity now keeps it from being reset
[Claude Tag] Fixed a thread that asked Claude to watch a pull request no longer hearing about CI failures, comments and reviews after Claude was restarted in that thread
[Claude Tag] Fixed deleting the first message of a thread Claude had already replied in not ending Claude's work there; it now stops, as it did when a message with no replies was deleted
[Claude Tag] Fixed Claude holding back a post because of an earlier instruction addressed to a different bot or assistant; only instructions addressed to Claude bind it, and it asks when unsure
[Claude Tag] Fixed the reply-mode card Claude posts on joining a busy channel saying it "sees a lot of automated posts" when the channel is only chatty or large; the card now names the real reason
[Claude Tag] Improved the Environment picker in Claude Tag admin settings: options are labeled Anthropic-hosted or self-hosted, with links to edit that environment or create one

Original changelog網頁版例行工作頁面改版，雙分頁呈現執行狀態

[Claude Code on the web] Changed the Routines page on claude.ai/code to a new layout with Yours and Templates tabs and two-column routine cards that show run status, and removed its calendar view

Original changelog雲端環境支援自訂網域白名單，離線工作階段發話即重啟

[Claude Code on the web] Fixed a cloud session sometimes taking about ten minutes to respond after its process exited while the session still looked live; sending a message now restarts it right away
[Claude Code on the web] Added a Custom network access option to the Cloud environments editor in admin settings, with the same allowed-domains list the environment dialog on claude.ai/code offers
[Claude Code on the web] Improved the Cloud environments admin page: it shows the default environment for Claude Tag and Claude Code, with a link to change it, and marks the recommended kind to create

Original changelogVSCode 支援關閉自動附加檔案，工作階段清單更穩定

[VSCode] Added an Attach Open File setting that, when turned off, stops the open file from being added to messages; selected text is still attached
[VSCode] Fixed session history showing only the current session when the workspace is on a Windows mapped network drive or SUBST drive
[VSCode] Fixed the session list's Active filter hiding open idle sessions when Open is also checked in the filter menu
[VSCode] Fixed a new chat switching back to the previous chat when the session list refreshed
[VSCode] Fixed open tabs and the side bar keeping the old config folder until a window reload after `CLAUDE_CONFIG_DIR` changed in the `environmentVariables` setting
[VSCode] Changed the on state of toggle switches from Claude orange to the editor theme's button color

Original changelogVSCode 掛鉤儲存防呆升級，消除主控台視窗閃爍

[VSCode] Fixed the Hooks and Permission rules dialogs reporting a save that landed as failed, and the Hooks dialog going blank under a plugin-only policy lock or showing color codes in save errors
[VSCode] Fixed Hooks dialog saves: no duplicate hook on replace, a header name retyped in other capitals keeps its secret, and settings.local.json is gitignored before the save returns
[VSCode] Fixed console windows flashing on Windows when the extension runs background commands such as git, ripgrep, and the sign-in status check
[VSCode] Fixed the prompt cache clock's hover text appearing only after a delay, and the auto-compact icon showing the browser's own tooltip beside its popup
[VSCode] Improved the Hooks dialog: a save refused because of the settings file itself now opens a popup with an "Open settings file" button and the reason behind "Copy error"

Original changelog遠端連線支援快速模式，/fast 關閉與降速備援修復

Added fast mode in Claude Code Remote sessions (cloud and self-hosted runners): the host's fast-mode setting or `/fast` typed in the session applies where your organization allows it
Fixed `/fast off` answering "Fast mode unavailable" instead of turning fast mode off when the organization has fast mode disabled
Fixed sessions started with `CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK` re-sending fast requests every turn after the API rejected fast mode; the rejection now stands and its reason is shown
Fixed fast mode under `CLAUDE_CODE_RETRY_WATCHDOG` failing the turn on a usage-credits limit, or retrying an overload at fast speed, instead of falling back to standard speed

Original changelog企業政策即時更新，支援自訂模型計費倍率

Added support for a `multiplier` above 1, up to 10, in the `modelPricing` managed setting and the Claude apps gateway `pricing` block, for marked-up internal chargeback rates
Fixed a cached organization policy being reused after switching accounts, organizations, or API keys, and the policy not refreshing until the hourly check when the credential changes mid-session
Fixed the tool and command lists not updating when the organization policy finishes loading after startup or changes mid-session
Fixed an enterprise `managed-mcp.json` that can't be read or parsed being ignored: it now keeps exclusive MCP control (user, project and plugin servers don't load) and warns at startup
Fixed org policy being fetched through, and rejected by, third-party local proxies set via `ANTHROPIC_UNIX_SOCKET`; they are again treated like other custom gateways, including for Remote Control

Original changelog自託管 Runner 支援退場標記，超大設定檔改存磁碟快照

Added `claude self-hosted-runner --drain-marker-file <path>`: when that file exists at a SIGTERM drain, the runner reports its exit to the server as a host drain (telemetry only)
Fixed self-hosted runner sessions silently losing all host config (settings, skills, plugins, MCP servers) when the host config directory exceeds 64 MiB; added `--host-config-snapshot disk|memory`

Original changelog沙盒指令獨立設定連線網域，自動模式審查行內指令

Added per-command `allowed_domains` to Bash, PowerShell and Monitor in auto mode with sandboxing: the hosts a command needs are reviewed with it and opened for it alone; other hosts are refused
Changed auto mode so that a skill's or slash command's inline `!` shell commands follow default-mode permission rules instead of the classifier; a command no rule decides runs as a reviewed tool call

Original changelogBash 權限檢查涵蓋萬用字元，防堵跨目錄讀取漏洞

Fixed Bash permission checks missing the file that `fmt`, `column` and similar commands read when it follows an option the checker doesn't recognize
Fixed Bash permission checks skipping files a wildcard expands to when the wildcard sits in a command's pattern or option value (for example `grep -v dir/* file`)
Fixed Bash permission checks so that shell variable declaration flags cannot misrepresent the command being run
Fixed Bash commands with two directory changes, a subshell, or a `cd`+`git` chain skipping the prompt under `permissions.blockReadsOutsideWorkingDirectories` in bypass and auto mode

Original changelog外掛安裝支援 SHA-256 驗證，確保指令未被竄改

Added `--accept-command <sha256>` to `claude plugin install` and `claude plugin update` to accept exactly the command a previous `--json` run displayed, instead of `-y`

Original changelog子代理支援精簡模式，動態工作流程達額度自動暫停

Added `omitClaudeMd` to agent frontmatter and `--agents` JSON, letting custom and plugin subagents run without user, project and local CLAUDE.md files; managed policy files still load
Fixed cloud sessions rejecting every subagent tool call ("updatedInput … failed schema validation") when a workflow or agent approval was applied after the session's worker restarted
Improved dynamic workflows to pause when you hit your usage limit and continue automatically when it resets, instead of dropping the affected agents
Changed auto mode so a subagent reports back to its caller through a dedicated hand-back call that the safety classifier reviews, instead of its last message being reviewed after the fact
Changed the default dynamic workflow size to small on Pro plans and lowered the medium size guideline from 15 to 10 agents
Updated the bundled `claude-api` skill to enable `eager_input_streaming` on streaming custom tools, and to start deliverable-shaped Managed Agents work with `user.define_outcome`

Original changelog/config 支援全螢幕滑鼠滾輪，路徑與選取標籤優化

Added mouse support to the `/config` panel in fullscreen mode: the wheel scrolls the settings list, a click on a setting's value changes it, and the row under the pointer is highlighted
Fixed the `/add-dir` path input: the left and right arrow keys now move the cursor, and Enter adds only the typed path instead of also adding the highlighted completion
Fixed text fields outside the main prompt moving a leading `!` to the end of what you typed (`!foo` came out as `foo!`)
Fixed a fullscreen rendering glitch where text kept a stale background color after the box around it lost its background
Changed the IDE selection indicator in the prompt to a `[⧉ …]` pill that wraps with the text instead of squeezing multi-line prompts; delete it with Backspace to leave the selection out

Original changelogMCP 支援工具簡稱搜尋，長任務定時回報防止中斷

Fixed resumed `claude -p` sessions whose tools all come from MCP servers failing with "At least one tool must have defer_loading=false"
Fixed sustained high CPU usage and repeated tool-list requests when an MCP server sends `list_changed` notifications in a tight loop
Fixed tool search returning no match when Claude selects an MCP tool by its bare name instead of its full `mcp__server__tool` name
Improved `claude mcp serve`: a running tool call now sends a progress update every 30 seconds, so clients show it is still running and idle timeouts don't abort a long command that prints nothing
Improved Foundry and Claude Platform on AWS sessions: an `alwaysLoad` MCP server that finishes connecting mid-conversation is usable on the next turn without a tool-search round trip

Original changelogMCP 授權註冊邏輯修復，快捷鍵不再中斷重新連線

Fixed MCP OAuth mishandling client registrations: denying consent forced a new one, one for another redirect URI was reused, and a concurrent write could delete a valid one or keep a mismatched one
Fixed Ctrl+O cancelling pending MCP server reconnects, and `/mcp` sent from Remote Control failing while the transcript view is open

Original changelog產出物渲染為精美排版文件，監聽數量上限翻倍

Fixed artifacts attached with `/artifacts` disappearing from the session after `--resume`
Fixed background sessions (`claude --bg`, `claude agents`) not watching the artifacts they publish for republishes made elsewhere
Improved Markdown files published as artifacts: they now render as styled document pages (title header, document typography, syntax-highlighted code)
Improved Artifact tool publish errors: a publish with no file now says to write the page to a file first, and an unsupported file type is reported before a missing favicon
Improved the Artifact tool's error when a page declares a capability its contract version lacks: it now lists every supported capability and notes when a newer contract version has it
Improved artifact watching: a session can now watch up to 10 published artifacts at once for republishes made elsewhere, up from 5

Original changelog工作階段恢復記憶精準，模型切換不誤報快取遺失

Fixed cross-session messages held by the receiving session's permission-mode policy leaving no trace: headless senders now get a delivery notice, and `SendMessage` results no longer imply it was read
Fixed Claude starting a second copy of a background command (such as a watch task or dev server) that was still running after the conversation was compacted
Fixed `/model` warning about losing the conversation cache when switching back to the model the conversation actually ran on
Fixed `/resume` and `/continue` showing only 1-2 sessions in fullscreen mode on short terminals
Fixed `/resume` and `/teleport` keeping the previous conversation's file-read tracking, so Claude could edit files the resumed conversation had never read
Fixed `--resume` dropping the 1M context window (`[1m]`) when the resumed session's model family differs from the configured default model

Original changelog掛鉤執行顯示耗時進度，支援 Esc 取消等待

Fixed the interactive `/hooks` menu crashing when a hook matcher is named after an inherited object property such as `__proto__` or `constructor`
Improved hook feedback: while a SessionStart, UserPromptSubmit, PreToolUse or SessionEnd hook runs, the spinner says so with elapsed time, and Esc cancels a prompt waiting on a SessionStart hook

Original changelog轉輪提示長思考狀態，引導下載桌面與行動版

Added a spinner tip pointing Bedrock, Vertex AI, Foundry and LLM gateway users to the Claude desktop app; the claude.ai desktop app tip now suggests `/desktop`, which offers to download the app
Fixed spinner tips suggesting commands that aren't available for your account type or are disabled in your session
Improved the spinner status during long thinking: it now reads "deep in thought" after 45s, and shows "picking the thought back up" while recovering from the output-token limit
Improved `/mobile` to show a single QR code for claude.ai/mobile, which opens the right app store for your phone

Original changelog跨作業系統檔案監聽更穩固，Git 鎖定檔殘留修復

Fixed a stale `.git/config.lock` breaking `git checkout -b`, `git push -u` and `git config` for the rest of a session after a sandboxed command failed to start (Linux)
Fixed settings file changes made outside the session going unnoticed on macOS machines whose system file-event service is saturated; the watcher now falls back to polling
Fixed custom agents, slash commands and output styles beyond the first not loading from a virtual drive that reports inode 0, such as an encrypted vault mounted as a Windows drive
Fixed skills synced from claude.ai staying on disk indefinitely after signing out; copies not refreshed within `cleanupPeriodDays` now move to the recoverable trash at the next launch
Fixed Delete in st and Alt+arrow keys in rxvt-unicode not working in attached background sessions
Fixed the terminal's replies to capability queries (`^[[?1;2c`) appearing at the shell prompt or in an editor when Claude Code exits, is suspended, or opens an editor right after starting

Original changelogClaude in Chrome 提示休眠狀態，遠端連線減少殘留

Fixed the Claude in Chrome prompt telling the model to load tools through ToolSearch when ToolSearch is unavailable
Improved Remote Control to leave fewer empty sessions on claude.ai when setup fails on a flaky network
Improved the Claude in Chrome message in cloud sessions when the browser can't be reached: it now says the computer may be asleep before it suggests an install

Original changelog終端機長文渲染不再卡頓，監聽任務設有逾時上限

Improved terminal rendering performance: large diffs and long transcripts render faster, with fewer slow frames
Improved startup time slightly by skipping a redundant validation of built-in model data on every launch
Changed Monitor watches to always have a deadline (at most 30 minutes; 10 in single-prompt `-p` runs) and notify Claude to re-arm, replacing the no-timeout `persistent` option
Windows: Fixed PowerShell commands failing with "Exit code 1" and no output when the session's temp output path reaches 260 characters

Original changelogLLM 閘道相容純文字回應，切換目錄精確重計技能

Fixed turns failing with "API returned an empty or malformed response" when an LLM gateway returns the non-streaming reply as `text/plain`
Fixed `/reload-skills` reporting a skill count that disagreed with the slash menu after `/cd`
Improved PDF @-mentions to say "page count unknown" instead of a page count guessed from the file size when pdfinfo cannot count the pages
Changed Claude apps gateway, Bedrock, Vertex AI, and Foundry sessions so that they no longer refresh a leftover claude.ai login that the session does not use