Claude Code 2.1.273：閘道請求標頭登場，企業流量分流更精準

繁體中文

新功能

閘道請求標頭登場，企業流量分流更精準

要讓企業閘道分辨每次模型呼叫的用途，現在可以讓請求自己帶上提示線索。

Claude Code 現在支援在發送請求時附帶五組提示標頭，包含請求類別、代理類型、前次工具耗時、上下文壓縮狀態以及是否已完成壓縮，方便後端閘道進行流量分析與路由分配。若要在發送請求時啟用這組標頭，只需設定 CLAUDE_CODE_GATEWAY_HINT_HEADERS=1。當你在內部網路透過自建閘道控管模型呼叫時，這項設定能協助企業系統精確掌握每次呼叫的用途與狀態。

新功能

遠端操控支援分支，背景建立獨立工作階段

想在手機上測試另一種修改方向，現在能直接建立對話分支。

透過 claude --remote-control 或 /remote-control 啟動遠端操控後，你可以直接在 Claude 應用程式中按下分支，電腦便會在背景自動開啟一個獨立的工作階段繼續執行。當你在外面用手機遠端除錯時，這正是保留當前進度並平行嘗試新解法的好時機。

新功能

外部工具斷線主動通知，登入過期提示指令

對話進行到一半，外部工具突然斷線，你現在不會被蒙在鼓裡。使用 MCP（Model Context Protocol，讓 AI 連接外部工具的協定）時，若工具伺服器在對話中途斷線且自動重連放棄，終端機會主動跳出通知並提示執行 /mcp 指令檢查。同時，若 MCP 伺服器的登入憑證在工作階段中途過期，錯誤訊息也會清楚說明如何重新驗證。

新功能

雲端平台自動模式預設本機分類，切換開關同步備妥

在雲端託管平台上執行自動模式，現在反應更直接。於 Bedrock、Vertex AI 與 Foundry 平台上，自動模式（auto mode，讓 Claude 自行判斷執行步驟的模式）目前改為預設採用本機分類器來辨識指令。若企業需要改用雲端平台端提供的伺服器端分類器，可以設定 CLAUDE_CODE_AUTO_MODE_SERVER=1。

新功能

遙測指標補齊工具名稱，問題回報過濾隱私資料

查效能、登入外掛、回報問題三件事，這版把該留下與該遮掉的資料分清楚。開啟 OTEL_LOG_TOOL_DETAILS=1 記錄效能時，成本與 token（AI 處理與計費的文字單位）指標會更加完整。記錄中現在會一併標記代理、技能、外掛與 MCP（讓 AI 連接外部工具的協定）伺服器的真實名稱。以 Claude 帳號登入時，系統會一併請求存取你的 claude.ai 外掛功能。執行 /bug 與 /feedback 回報問題時，回報內容僅保留最後一次 API 請求的模型名稱、系統提示詞與工具清單等行為參數，自動排除請求中繼資料與 CLAUDE_CODE_EXTRA_BODY 欄位。

問題修復

停用技能移至回收區，企業管理規則優先套用

公司關掉技能或下發管理規則後，Claude Code 不該繼續沿用舊權限。企業在 claude.ai 關閉技能功能後，先前已同步的技能不會繼續留在對話中，而是直接移至可復原的垃圾桶。此外，透過行動裝置管理（MDM）或 managed-settings.json 設定的 allowManagedMcpServersOnly、deniedMcpServers 與 disableClaudeAiConnectors 規則，即使遇到伺服器端管理的設定並存，也不會再被忽略，確保資安控管如實生效。

問題修復

安全邊界防護補強，指令詢問與路徑存取更嚴格

看不懂的 Bash 指令、工作目錄外的記憶路徑與 macOS 檔案路徑，這版各自回到正確的權限處理。

啟用 permissions.blockReadsOutsideWorkingDirectories 時，檢查器無法完整分析的 Bash 指令不再跳過確認提示，同時修復了子殼層在旁路模式下隱藏危險 rm 指令的漏洞；專案設定指定的記憶目錄若位於工作目錄外，也不再載入提示詞、檢索、建立索引或用於記憶擷取。macOS 上拖入截圖或具有次要路徑的檔案，不再因符號連結（symlink，指向另一個位置的捷徑檔）解析問題而被拒絕讀取。先前針對無法分析指令的過度阻擋已還原，例如 time -p make build 恢復跳出詢問提示。

新功能

子代理狀態完整回傳，長對話排程運作更順暢

背景子代理跑到一半時，最重要的是結果別消失、排程別跑錯地方。

串流回覆若缺少 token（AI 處理與計費的文字單位）用量或模型編號，子代理與背景代理不會再被判定為失敗而遺失結果。透過 SDK 或 --output-format stream-json 輸出時，即使子代理中途移至背景執行，後續訊息與最終報告也不會被丟棄。/tui 重啟時不會再因已結束的代理團隊成員而卡住。複製 .claude/scheduled_tasks.json 至新工作樹（worktree，Git 平行開多份副本的功能）時，排程任務不會在錯誤的工作階段執行。長時間對話中掛鉤進度與子代理更新也不再重新處理整段紀錄。

問題修復

思考歷程精準保留，上下文用量計算不超標

長對話裡，思考歷程與上下文用量都需要被精準保存。執行 /login、/upgrade 或 /extra-usage 時，系統不會再丟棄對話中先前的思考歷程，避免下次請求被迫全面重新寫入提示詞快取（cache，把算過的結果暫存起來直接用）。上下文計量器與自動壓縮不再將諮詢工具對話誤算為兩倍大小，解決了壓縮功能在視窗過半時就過早觸發的問題。遠端操控客戶端連線至 Claude Desktop、VS Code 或 JetBrains 工作階段時，查詢上下文用量不會再遭拒絕。壓縮狀態列上的旋轉圖示也不再出現重複的刪節號。

新功能

Artifacts 發布連線自動重送，資料庫支援單欄更新

Artifact 發布、頁面讀取與資料庫更新這幾條路，這版都減少了卡住後重做的機會。

在雲端或遠端操控工作階段中，自動模式上傳對話附件時，Artifact 工具不會再無故暫停等待核准。更新資料庫時可單獨刪除單一欄位而無需重寫整份文件；發布請求若在抵達伺服器後斷線，系統會安全重送，避免發布失敗或產生重複版本。遇到不支援的檔案類型時，終端機會顯示簡潔單行提示並告知 Claude 支援的格式；讀取頁面時會清楚說明該頁面的資料庫規則與權限，且操作後不再誤跳外掛提示。

新功能

雲端閘道授權錯誤指引具體，憑證與代理診斷更清晰

遇到雲端平台與網路連線異常時，診斷訊息現在會指出具體原因。在 Bedrock、Vertex AI 與 Foundry 遇到 401/403，或 Claude 應用程式閘道遇到 403 時，訊息不再盲目要求執行 /login，而是明確點出需要重新整理哪一組憑證，或指引你聯絡閘道管理員。當雲端工作階段因登入過期或被撤銷而無法建立時，提示會正確引導執行 /login 登入。工作階段內的 SSL 憑證與代理伺服器連線錯誤，現在會列出錯誤代碼與修復方式，例如針對未受信任的企業憑證提示設定 NODE_EXTRA_CA_CERTS。

新功能

GitHub 流程錯誤精準定位，拉取請求診斷更明確

只看到泛用錯誤碼的 GitHub 流程，最容易把你卡在登入、SAML 或 Webhook 問題外面。執行 /install-github-app 時，遇到 SAML 單一登入阻擋不會再被誤報為缺少管理員權限；雲端工作階段遇到 IP 白名單、應用程式停用或 SAML 限制時，會直接顯示具體成因。執行 /autofix-pr 時，若 gh pr view 失敗會呈現 GitHub 工具原生的登入或限流錯誤，Webhook 派送失敗也會具體說明原因。執行 /web-setup 遇到 GitHub token（授權令牌）被拒時會列出可能原因與解法，連線失敗也會明確指出代理伺服器或憑證問題。

問題修復

Shell 模式驚嘆號正常輸入，版本庫排除檔不再殘留

你在 Shell 模式下想打 ! grep …，句首那個驚嘆號卻憑空消失——這個問題修好了。切換至 Shell 模式後，主提示列不會再吃掉句首輸入的 ! 驚嘆號符號，讓你能夠順利輸入包含否定邏輯的命令。另一個修好的是：在長時間運作的工作階段中，若專案版本庫的 .git 目錄已被刪除或移走，Claude Code 不會再無故於原地重新建立多餘的 .git/info/exclude 存根檔案。

新功能

VS Code 隱藏停用回饋，Windows 異常橫幅消除

VS Code 回饋入口該隱藏就隱藏，Windows 回合結束也不該冒出假錯誤。若企業組織已設定停用產品回饋，VS Code 介面中的「回報問題」選項以及 /bug 與 /feedback 指令都不會再開啟回報表單。在 Windows 平台下，每輪對話執行完成後，畫面頂端不會再無故彈出顯示錯誤碼 4294967295 的紅色行程結束橫幅。此外，透過 --add-dir 加入掛載的網路磁碟機時，系統也強化了通用命名規則（UNC）路徑的網路權限檢查。

新功能

網頁版排程管理更新，未儲存變更彈窗確認

例行排程頁面現在會先問清楚，才丟掉你輸入的名稱、提示詞與修改內容。

管理員移除並重新加入組織連接器後，例行排程（routine）不會再失去連線權限或錯誤呼叫舊有的連接器。在新增例行排程頁面或編輯對話框中，若已輸入名稱、提示詞或修改內容，關閉前會跳出確認視窗詢問是否捨棄未儲存的變更。詳情頁面也重新規劃：導覽路徑支援選單與重新命名，頂部整合開關與立即執行按鈕，設定旁則並排顯示歷次執行紀錄。

新功能

網頁版環境建立更穩定，直接導向設定流程

新使用者進網頁版前，環境建立、隱私設定與初次引導都少走彎路。從組織設定建立自託管環境時，修復了伺服器錯誤導致殘留未完成環境的問題。管理員的「分享雲端工作階段」設定位置調整至「資料與隱私」分頁下方，方便資料與隱私管理員直接進行設定與控管。在 Mac 與 Windows 平台上尚未配置雲端環境的新使用者，進入網頁版時不再強制顯示滿版桌面應用程式下載畫面，而是直接引導進入設定流程。

問題修復

Slack 智慧監看公開頻道，討論串對話與記憶管理修復

Slack 裡的 Claude Tag 若沉默、重開任務或漏列記憶，團隊很難信任它接手協作。

Claude 現在會主動監看與對話相依的公開頻道（例如重大事件頻道），無需手動呼叫；在舊討論串回覆時，也不會再導致正在進行的任務重新啟動而遺失未推送的工作進度。組織共用私密頻道中的排程任務能持續在原討論串中穩定運作。Enterprise Grid 斷線但其中一個工作區仍連線時，重新安裝應用程式後不再在幾分鐘內無故靜音，新增頻道管理員也支援 Enterprise Grid 共用頻道、尚未使用過 Claude 的頻道與歷史私密頻道。管理員記憶頁面現在會列出 Claude 自行建立的頻道記憶，方便管理員隨時檢視、編輯與刪除。

問題修復

雲端服務連線強化，AWS 與身分驗證修復

帳號剛完成授權更新、AWS 端點沒有區域、第三方回傳小寫授權類型時，連線不該因此斷在半路。帳號 token（授權令牌）更新後，系統不會再誤跳找不到環境的通知而遺失訊息。AWS 連線全面支援 Budgets、Savings Plans、WAF Classic 與 Import/Export 等無需指定區域的端點，Global Accelerator 請求也能正確完成簽章；若連線因主機名稱未指定區域等原因無法完成簽章，系統會清楚說明成因與修復指引。OAuth 與 JWT 連線在第三方回傳小寫令牌類型時，請求現在會統一以標準格式發送。

問題修復

程式碼審查更精準，相依性說明與留言機制修復

程式碼審查最怕小變更被重審一整輪，或留言重試後重複／漏發。

將基礎分支合併至先前帶有「其他發現」的拉取請求時，系統會改跑輕量級的後續審查而非全量重審；擁有者或儲存庫名稱包含大寫字母的儲存庫遇到空推送或相同內容時也會自動略過。REVIEW.md 不會再因 @-mention、跨行程式碼或反引號 HTML 標籤而整份被忽略，只會扣下連到已變更檔案的行。修改建議現在會主動指明相依程式碼需維持運作的行為，受影響次要位置的留言改以完整句子呈現。執行 /ultrareview --post 重試時能確保留言恰好發布一次並標明提交版本。

English

FEATURE

Gateway Hint Headers Introduced for More Precise Enterprise Traffic Routing

To help enterprise gateways tell what each model call is for, requests can now carry their own hint signals.

Claude Code now supports sending five hint headers with requests, covering request class, agent type, previous tool durations, context compaction status, and whether compaction was completed, making it easier for backend gateways to analyze traffic and route requests. To enable these headers when sending requests, simply set CLAUDE_CODE_GATEWAY_HINT_HEADERS=1. When managing model calls via custom gateways on internal networks, this setting helps enterprise systems accurately track the purpose and state of each invocation.

FEATURE

Remote Control Adds Session Forking to Launch Independent Background Sessions

When you want to test an alternative direction on mobile, you can now fork the conversation directly.

After launching Remote Control with claude --remote-control or /remote-control, you can tap fork directly in the Claude app, and your computer will automatically spawn an independent background session to continue execution. When debugging remotely from your phone on the go, this lets you preserve current progress while testing new approaches in parallel.

FEATURE

Proactive Disconnection Alerts for External Tools and Commands for Expired Logins

When an external tool drops mid-session, you are no longer left guessing. When using MCP (Model Context Protocol, a protocol connecting AI to external tools), if a tool server disconnects mid-session and automatic reconnection gives up, the terminal proactively displays a notification pointing you to the /mcp command. Additionally, if an MCP server's login credentials expire mid-session, the error message clearly explains how to re-authenticate.

FEATURE

Cloud Platforms Default Auto Mode to Local Classifier with Switch Available

Running auto mode on cloud-hosted platforms is now more responsive. Across Bedrock, Vertex AI, and Foundry, auto mode currently defaults to using the local classifier to identify commands. If organizations need to switch to the cloud platform's server-side classifier, they can set CLAUDE_CODE_AUTO_MODE_SERVER=1.

FEATURE

Telemetry Metrics Complete Tool Names and Problem Reports Filter Private Data

For performance checks, plugin login, and issue reports, this release separates the data that should be kept from the data that should be hidden. When enabling OTEL_LOG_TOOL_DETAILS=1 to track performance, cost and token metrics become more complete. Logs now include the real names of agents, skills, plugins, and MCP servers. Signing in with a Claude account now also requests access to your claude.ai plugins. When running /bug and /feedback to report issues, submitted reports retain only behavioral parameters from the last API request—such as model name, system prompt, and tools list—while automatically omitting request metadata and CLAUDE_CODE_EXTRA_BODY fields.

BUGFIX

Disabled Skills Move to Trash and Enterprise Management Rules Take Precedence

After a company turns off skills or rolls out managed rules, Claude Code should not keep using old permissions. When an organization turns off skills in claude.ai, previously synced skills no longer remain in sessions and instead move directly to the recoverable trash. Furthermore, allowManagedMcpServersOnly, deniedMcpServers, and disableClaudeAiConnectors rules configured via Mobile Device Management (MDM) or managed-settings.json are no longer ignored even when server-managed settings coexist, ensuring security controls take full effect.

BUGFIX

Security Boundary Protections Reinforced with Stricter Command Prompts and Path Access

Unreadable Bash commands, memory paths outside the working directory, and macOS file paths each return to the right permission handling.

When permissions.blockReadsOutsideWorkingDirectories is enabled, Bash commands that the permission checker cannot fully analyze no longer skip confirmation prompts, and a flaw where a subshell in bypass mode hid a dangerous rm command has been fixed; memory directories specified by project settings outside the working directory are no longer loaded into the prompt, recalled, indexed, or used for memory extraction. On macOS, dragged-in screenshots or files reported under a secondary path are no longer denied due to symlink resolution issues. Previous over-blocking on unanalyzable commands has been reverted, restoring confirmation prompts for commands like time -p make build.

FEATURE

Sub-Agent States Delivered Reliably and Long-Session Scheduling Runs Smoother

When background sub-agents are mid-run, the important part is that results do not disappear and scheduling does not run in the wrong place.

If a streamed reply omits token usage or carries no model ID, sub-agents and background agents are no longer marked as failed with lost results. When outputting via the SDK or --output-format stream-json, remaining messages and final reports from a sub-agent are no longer dropped even if it moves to the background mid-run. /tui no longer hangs on restart due to an agent-team member that has already finished. When .claude/scheduled_tasks.json is copied to a new worktree, scheduled tasks no longer run in the wrong session. Hook progress and sub-agent updates in long sessions no longer re-process the entire conversation history on every update.

BUGFIX

Thinking History Preserved Accurately and Context Usage Metering Avoids Premature Triggers

In long conversations, thinking history and context usage both need to be preserved accurately. Running /login, /upgrade, or /extra-usage no longer discards previous thinking history from the conversation, avoiding forced full prompt-cache rewrites on subsequent requests. The context meter and auto-compact no longer count advisor-tool turns at twice their actual size, fixing an issue where compaction triggered prematurely halfway through the window. Remote Control clients connected to Claude Desktop, VS Code, or JetBrains sessions are no longer refused when querying context window usage. The compaction status line spinner no longer displays duplicate ellipses.

FEATURE

Artifacts Publishing Re-sends Automatically and Databases Support Single-Field Updates

Artifact publishing, page reads, and database updates now make it less likely that you have to redo work after getting stuck.

In cloud or Remote Control sessions, the Artifact tool no longer pauses unnecessarily for approval when auto mode uploads chat attachments. Database updates can now remove a single field without rewriting the entire document, and publishing requests whose connections drop after reaching the server are re-sent safely to avoid failures or duplicate versions. When encountering unsupported file types, the terminal displays a clean single-line error while informing Claude of supported formats; reading a page clearly explains the database rules and capabilities for that page, and operations no longer trigger false-positive plugin suggestions.

FEATURE

Cloud Gateway Auth Errors Provide Specific Guidance with Clearer Certificate and Proxy Diagnostics

Diagnostic messages now pinpoint specific causes during cloud platform and network connection issues. When encountering 401 or 403 errors on Bedrock, Vertex AI, or Foundry, or 403 errors on Claude app gateways, messages no longer blindly prompt to run /login; instead, they clearly specify which credential to refresh or guide you to contact your gateway administrator. When a cloud session cannot be created due to an expired or revoked login, the prompt correctly directs you to run /login. In-session SSL certificate and proxy connection errors now list error codes and remediation steps, such as pointing to NODE_EXTRA_CA_CERTS for untrusted corporate certificates.

FEATURE

GitHub Workflow Errors Pinpoint Causes with Clearer Pull Request Diagnostics

When a GitHub workflow only shows a generic error code, it can leave you stuck outside the real login, SAML, or webhook problem. Running /install-github-app no longer misreports SAML single sign-on blocks as missing administrator permissions, and cloud sessions encountering IP allowlists, suspended app installations, or SAML restrictions directly show the specific cause. When running /autofix-pr, failures with gh pr view display native GitHub CLI errors (such as sign-in or rate limits), while webhook delivery failures state the specific reason. Running /web-setup with a refused GitHub token lists likely causes and fixes, and connection failures clearly point out proxy or TLS certificate issues.

BUGFIX

Exclamation Marks Allowed in Shell Mode and Repository Exclude Stubs No Longer Linger

You are in shell mode, you type ! grep …, and the leading exclamation mark vanishes — that one is fixed. When already in shell mode, the main prompt no longer drops a leading ! exclamation mark, so you can type negated commands. The other fix: during long-running sessions, if a repository's .git directory has been deleted or moved away, Claude Code no longer recreates a stub .git/info/exclude file in its place.

FEATURE

VS Code Hides Disabled Feedback and Windows Error Banner Eliminated

VS Code feedback entry points should stay hidden when disabled, and Windows should not show fake errors after a turn completes. If an organization has disabled product feedback, the "Report a problem" option in VS Code along with /bug and /feedback commands will no longer open a report form. On Windows, a red exit banner showing error code 4294967295 no longer pops up at the top of the screen after completed turns. Furthermore, when adding mapped network drives via --add-dir, the system strengthens network permission checks for Universal Naming Convention (UNC) paths.

FEATURE

Web Routine Management Updated with Confirmation Dialog for Unsaved Changes

Routine pages now ask first before throwing away the name, prompt, or edits you typed.

After an administrator removes and re-adds an organization connector, routines no longer lose access permissions or erroneously invoke old connectors. When a routine name, prompt, or edit has been entered on the New routine page or Edit routine dialog, closing prompts a "Discard unsaved changes?" confirmation. The detail page has also been redesigned: breadcrumbs now include a menu and rename option, the on/off switch and Run now button are integrated at the top, and run history is displayed alongside routine settings.

FEATURE

Web Environment Setup Made More Reliable with Direct Routing to Configuration

Before new users enter the web version, environment creation, privacy settings, and first-run guidance now take fewer detours. When creating a self-hosted environment from organization settings, a server error that left half-created environments behind has been fixed. The administrator "Share cloud sessions" setting has been moved under Data and privacy, allowing Data and privacy admins to manage it directly. New users on Mac and Windows without a configured cloud environment no longer see a mandatory full-page desktop app download screen when visiting the web version, heading straight to setup instead.

BUGFIX

Claude Tag Watches Public Slack Channels with Thread and Memory Fixes

If Claude Tag in Slack goes silent, restarts work, or leaves out memory, teams have a hard time trusting it to take over collaboration.

Claude now proactively watches related public channels (such as incident channels a conversation depends on) without needing manual mentions; replying in an older thread no longer restarts in-progress tasks or loses unpushed work. Scheduled tasks in organization-shared private channels now continue running reliably in their original threads. Reinstalling the app under Enterprise Grid when the grid is disconnected but one of its workspaces stays connected no longer causes Claude to go silent within minutes, and adding channel managers is now supported on Enterprise Grid shared channels, channels where Claude has not yet been used, and legacy private channels. The admin Memory page now lists channel memory created by Claude, allowing administrators to review, edit, and delete it at any time.

BUGFIX

Cloud Service Connectivity Reinforced with AWS and Authentication Fixes

Connections should not break midway just because an account was freshly authorized, an AWS endpoint has no region, or a third party returns a lowercase auth type. After an account token refreshes, messages are no longer dropped with false notices claiming an environment could not be found. AWS connections fully support region-less endpoints such as Budgets, Savings Plans, WAF Classic, and Import/Export, and Global Accelerator requests now sign properly; if a request cannot be signed due to hostnames lacking a region, the system clearly explains the cause and remediation. When third-party providers return lowercase token types for OAuth client-credentials and JWT-bearer connections, requests now standardize on the standard Bearer scheme.

BUGFIX

Code Review Precision Improved with Dependency Notes and Comment Delivery Fixes

In code review, few things hurt more than a small change getting a full re-review, or retrying comments only to duplicate or miss them.

When merging a base branch into a pull request whose prior review noted "Additional findings," the system runs a lighter follow-up review instead of a full re-review; repositories whose owner or name contains a capital letter now skip empty or content-identical pushes. REVIEW.md files are no longer ignored in their entirety due to @-mentions, multiline code spans, or backticked HTML tags. Instead, only lines that link to changed files are withheld. Suggested fixes now state what dependent code must keep working, and review comments for secondary affected locations use full sentences. Retrying /ultrareview --post ensures findings comments post exactly once while naming the reviewed commit.

原始 CHANGELOG

Original changelog閘道請求標頭登場，企業流量分流更精準

Added `x-claude-code-request-class`, `x-claude-code-agent-type`, `x-claude-code-prev-tool-durations`, `x-claude-code-compaction` and `x-claude-code-context-compacted` request headers for LLM gateways; opt in with `CLAUDE_CODE_GATEWAY_HINT_HEADERS=1`

Original changelog遠端操控支援分支，背景建立獨立工作階段

Added forking a session started with `claude --remote-control` or `/remote-control` from the Claude app; the fork runs as a background session on your computer

Original changelog外部工具斷線主動通知，登入過期提示指令

Added a notification when an MCP server disconnects mid-session and automatic reconnection gives up, pointing at `/mcp`
Improved the error shown when an MCP server's sign-in expires mid-session to say how to re-authenticate (`/mcp`)

Original changelog雲端平台自動模式預設本機分類，切換開關同步備妥

Changed auto mode on Bedrock, Vertex and Foundry to use the local classifier by default for now; set `CLAUDE_CODE_AUTO_MODE_SERVER=1` to use the platform's server-side classifier

Original changelog遙測指標補齊工具名稱，問題回報過濾隱私資料

Changed `OTEL_LOG_TOOL_DETAILS=1` to also include real agent, skill, plugin and MCP server names on cost and token metrics
Changed sign-in with a Claude account to also request access to your claude.ai plugins
Changed `/bug` and `/feedback` reports to include only model-behavior params (model, system prompt, tools) from the last API request, omitting request metadata and `CLAUDE_CODE_EXTRA_BODY` fields

Original changelog停用技能移至回收區，企業管理規則優先套用

Fixed skills synced from claude.ai staying available after your organization turns Skills off; they now move to the recoverable trash
Fixed `allowManagedMcpServersOnly`, `deniedMcpServers` and `disableClaudeAiConnectors` set via MDM or `managed-settings.json` being ignored when server-managed settings are also present

Original changelog安全邊界防護補強，指令詢問與路徑存取更嚴格

Fixed Bash commands the permission checker cannot fully analyze skipping the prompt under `permissions.blockReadsOutsideWorkingDirectories`, and a subshell hiding a dangerous `rm` in bypass mode
Fixed `permissions.blockReadsOutsideWorkingDirectories`: a memory directory chosen by a repository's settings is no longer loaded into the prompt, recalled, indexed, or used by memory extraction
Fixed Read on macOS refusing a dragged-in screenshot, or any file the system reports under a second path, with "symlink resolution changed after permission was checked"
Reverted a 2.1.268 change that checked Read and Edit deny rules on Bash lines the permission checker can't analyze (`eval`, `env -C`); commands like `time -p make build` prompt again instead of being denied

Original changelog子代理狀態完整回傳，長對話排程運作更順暢

Fixed sub-agents and background agents being reported as failed, with their result never delivered, when the final streamed reply omitted token usage or carried no model id
Fixed SDK and `--output-format stream-json` output dropping a subagent's remaining messages and final report after it is moved to the background mid-run (e.g. by `CLAUDE_AUTO_BACKGROUND_TASKS`)
Fixed `/tui` refusing to restart because of an agent-team teammate that had already finished its work and was no longer shown in the agents panel
Fixed saved scheduled tasks running in the wrong session after `.claude/scheduled_tasks.json` was copied into another folder, such as a new worktree
Improved responsiveness in long sessions: hook progress and sub-agent activity no longer re-process the whole conversation on every update

Original changelog思考歷程精準保留，上下文用量計算不超標

Fixed `/login`, `/upgrade`, and `/extra-usage` discarding earlier thinking from the conversation, which forced a full prompt-cache rewrite on the next request
Fixed the context meter and auto-compact counting advisor-tool turns at roughly twice their real context size, which made auto-compact fire at about half the real window
Fixed Remote Control clients attached to a Claude Desktop, VS Code or JetBrains session being refused when they ask for the session's context window usage
Fixed the spinner showing a doubled ellipsis ("……") on compaction status lines such as "Running PreCompact hooks…"

Original changelogArtifacts 發布連線自動重送，資料庫支援單欄更新

Fixed auto mode stopping for approval when the Artifact tool uploads a file you attached to the chat in a cloud or Remote Control session
Fixed a false-positive spinner tip suggesting the frontend-design plugin after reading or publishing Artifacts
Improved the Artifact tool's error when a publish includes a file type artifacts don't serve: Claude is told which types are served and what to do instead, and the terminal shows one plain line
Improved the Artifact tool's page read to state the capabilities and database rules the artifact service holds for the page, for anyone who can publish to it
Improved artifact database writes: an update can now remove a single field instead of rewriting the whole document
Improved artifact publishing: a publish whose connection drops after reaching claude.ai is now re-sent safely instead of failing or creating a duplicate version

Original changelog雲端閘道授權錯誤指引具體，憑證與代理診斷更清晰

Fixed 401/403 errors on Bedrock, Vertex and Foundry, and Claude apps gateway 403s, telling you to run `/login`; the message now names the credential to refresh or points to your gateway administrator
Improved the error when a cloud session can't be created because your Claude login expired or was revoked: it now tells you to run `/login`
Improved the in-session SSL certificate and proxy connection errors to name the error code and what to fix, such as `NODE_EXTRA_CA_CERTS` for an untrusted corporate CA

Original changelogGitHub 流程錯誤精準定位，拉取請求診斷更明確

Fixed `/install-github-app` reporting a SAML single sign-on block as "admin permissions required"
Improved the cloud-session GitHub error for an IP allow list, a suspended app installation or SAML single sign-on to show the cause instead of a generic install hint
Improved `/autofix-pr`: when `gh pr view` fails it now shows gh's own error (sign-in, SAML, rate limit) instead of a generic exit-code line
Improved `/autofix-pr` to say why GitHub webhook delivery couldn't be set up for the PR (for example, no linked GitHub account) instead of a generic warning
Improved `/web-setup` errors: a refused GitHub token now lists the likely reasons and the fix, and a connection failure names a configured proxy or TLS certificate problem

Original changelogShell 模式驚嘆號正常輸入，版本庫排除檔不再殘留

Fixed a long-running session recreating a stub `.git/info/exclude` after the repository's `.git` directory was removed or moved away
Fixed the main prompt dropping a `!` typed at the start while already in shell mode, so negated commands like `! grep …` can be typed

Original changelogVS Code 隱藏停用回饋，Windows 異常橫幅消除

[VSCode] Fixed "Report a problem" still appearing, and `/bug` / `/feedback` opening a report form, for organizations that have product feedback disabled
[VSCode] Fixed a red "Claude Code process exited with code 4294967295" banner appearing after completed turns on Windows
Windows: Improved the network-path permission check for UNC paths when a mapped network drive was added with `--add-dir`

Original changelog網頁版排程管理更新，未儲存變更彈窗確認

[Claude Code on the web] Fixed routines losing access to an organization connector, and still calling the old one, after an admin removed and re-added that connector
[Claude Code on the web] Added a "Discard unsaved changes?" confirmation before the New routine page or the Edit routine dialog throws away a routine name, prompt or edit you typed
[Claude Code on the web] Improved the routine detail page: menu and rename in the breadcrumb, the on/off switch and Run now at the top, and run history beside the routine's settings

Original changelog網頁版環境建立更穩定，直接導向設定流程

[Claude Code on the web] Fixed creating a self-hosted environment from organization settings occasionally failing with a server error and leaving a half-created environment behind
[Claude Code on the web] Changed the admin "Share cloud sessions" setting to live under Data and privacy instead of the Claude Code page, where Data and privacy admins can also manage it
[Claude Code on the web] Removed the full-page desktop-app download screen that new users without a cloud environment saw on Mac and Windows; they now go straight to setup

Original changelogSlack 智慧監看公開頻道，討論串對話與記憶管理修復

[Claude Tag] Fixed Claude going silent minutes after reinstalling the app when an Enterprise Grid was disconnected but one of its workspaces stayed connected
[Claude Tag] Fixed scheduled tasks set up in an organization-shared private Slack channel silently never posting; they now keep running in the thread they were created in
[Claude Tag] Fixed replying in an older Slack thread while Claude is mid-task sometimes restarting it from scratch and losing work it had not pushed yet
[Claude Tag] Fixed adding a channel manager being refused on Enterprise Grid shared channels, on channels where Claude hasn't been used yet, and on legacy private channels
[Claude Tag] Changed Claude to start watching related public channels on its own, such as an incident channel a conversation depends on, instead of only when asked
[Claude Tag] Fixed the admin Memory page not listing Slack channels Claude set up on its own even when they had saved memory; admins can now open, edit and delete that memory

Original changelog雲端服務連線強化，AWS 與身分驗證修復

[Claude Tag] Fixed Claude occasionally dropping a message with an incorrect "couldn't find a Claude Code environment" notice right after your account token refreshed
[Claude Tag] Fixed AWS connections refusing region-less endpoints such as Budgets, Savings Plans, WAF Classic and Import/Export; Global Accelerator requests now sign correctly
[Claude Tag] Improved AWS connection failures: when a request can't be signed, such as a hostname with no region, Claude is told why and how to fix it instead of a bare error
[Claude Tag] Fixed OAuth client-credentials and JWT-bearer connections failing with providers that return a lowercase token type; requests now send the standard Bearer scheme

Original changelog程式碼審查更精準，相依性說明與留言機制修復

[Code Review] Fixed merging the base branch into a PR whose earlier review listed "Additional findings" triggering a full re-review; these pushes now get the lighter follow-up review
[Code Review] Fixed a whole REVIEW.md being ignored because of an @-mention, a code span wrapped across lines, or a backticked HTML tag; only lines linking to changed files are withheld
[Code Review] Improved suggested fixes to say what the fix must keep working when other code depends on the behavior being changed
[Code Review] Improved review comments that point to a second affected location to state that location's issue in a full sentence instead of a cut-off stub
[Code Review] Fixed `/ultrareview --post` so a retry after a GitHub error posts the findings comment exactly once instead of never or twice; the comment now names the reviewed commit
[Code Review] Fixed empty or content-identical pushes being re-reviewed on GitHub repositories whose owner or name contains a capital letter; these pushes are now skipped