AI 코딩 도구,
이번엔 뭐가 바뀌었나

• + Copilot harness를 통해 새로운 에이전트 기능을 추가하고 여러 VS Code 창에서 동일한 에이전트 세션에 연결할 수 있게 한다 # 에이전트
  원문

  [Agents] Copilot harness: The Copilot harness adds exciting new agent functionality to VS Code while keeping the way you work familiar. It's powered by the Copilot SDK, so its behavior and capabilities are consistent with other Copilot products, including the standalone GitHub Copilot app and the Copilot CLI. To start using the Copilot harness, select it from the harness picker in the chat input. In this release, it might already be the default selection for you. You can continue working as you normally do. The harness runs in a dedicated agent host process based on the Agent Host Protocol (AHP), so you can connect to the same agent session from multiple VS Code windows. Learn how to work with the Copilot harness, or explore the architecture and workflows in the agent host blog post. If you have feedback or requests, please file an issue.

• + HydraFusion 모델 오케스트레이션 시스템을 도입하여 작업에 따라 모델과 워크플로우를 자동으로 선택하게 한다 # 에이전트
  원문

  [Agents] HydraFusion model orchestration (Research Preview): HydraFusion is an adaptive model orchestration system that chooses the models and workflow for each coding task. It can solve a task with one model, escalate to a stronger model, or have another model critique and revise the result. This approach aims to improve result quality while balancing speed and cost, without requiring you to coordinate models yourself. Eligible users with preview features enabled can select HydraFusion in the model picker.

• + Agents window에서 연결된 원격 에이전트 호스트로 작업을 위임하는 기능을 추가한다 # 에이전트
  원문

  [Agents] Delegate tasks to remote agent hosts (Experimental): Settings: chat.remoteAgentHostsEnabled , chat.remoteSessions.tools.enabled (Agents window only) Your agent can delegate work to connected remote agent hosts from the Agents window, without you selecting a host in a picker for each task. The new built-in tools let your agent: Discover hosts, models, resource capacities, and session load with list_agent_hosts. Start a session with create_remote_session. Specify a host, or let automatic placement match an operating system (Windows, Linux, or macOS), minimum memory capacity, logical CPU count, and optional model. Among matching hosts, placement favors the host with the fewest running sessions and pending creations. Check a remote session's status and latest response with get_remote_session. Send follow-up work or report results and questions back to the originating chat with send_remote_

• + 멀티 폴더 세션 기능을 통해 여러 리포지토리나 격리된 워크트리에서 관련 작업을 한 세션에서 조정할 수 있게 한다 # 에이전트
  원문

  [Agents] Multi-folder sessions (Experimental): Multi-folder sessions let you coordinate related work across repositories or isolated worktrees in one session. Previously, every chat in a multi-chat session shared the same folder and checkout. In this release, each chat can use its own folder or worktree without changes leaking between chats. Each chat uses its folder for its terminal, tasks, changes, pull request, and Agent merge state. Chats that use the same folder share that state. In the sessions list, hover over a session to see a summary across all its folders, or hover over a nested chat to see details for only its folder. Example use cases include: Implement a feature across repositories: Ask the main chat to work in one repository and create a peer chat for a second repository. Compare approaches in separate worktrees: Ask the main chat to create peer chats that each use a fresh worktr

• + 세션 내에서 작업이 완료된 채팅을 마크하고 보관하여 세션의 나머지 부분을 활성 상태로 유지하게 한다 # 에이전트
  원문

  [Agents] Archive chats in a session: Keep the rest of a session active when one chat finishes its work. Because sessions can contain multiple chats, select Mark as Done to archive only the finished peer chat, not the entire session. To restore it with its title and transcript intact, select the Done filter in the sessions list.

• + 위임된 작업 시 세션 계획 또는 결과물 여부와 워크스페이스 및 워크트리 선택을 분리하여 결정하게 한다 # 에이전트
  원문

  [Agents] Choose the session relationship and workspace for delegated work: Make two separate decisions when you ask an agent to delegate work. First, decide whether the task belongs to the current session's plan or deliverable. Then specify a workspace only if the task needs repository files, and request a worktree only if file changes need isolation. Related work can still use a different workspace. Create a peer chat for related work: Ask the agent to create a peer chat in the current session. To reuse the current workspace and checkout, say so in your prompt. Because the peer chat shares the checkout, use this option for research, planning, or related work that doesn't need isolated file changes. Create an independent session for unrelated work: Ask the agent to create an independent session with no workspace when a separate deliverable doesn't need repository files. The new session doesn't

• + Agents window에서 새 세션 생성 시 선택적 환영 문구와 사용자 이름을 포함하는 기능을 추가한다 # 에이전트
  원문

  [Agents] Personalize new-session welcome phrases (Experimental): Settings: sessions.chat.experimental.welcomePhrases , sessions.chat.experimental.welcomeName , accessibility.verbosity.newSessionWelcome (Agents window only) Add some personality to the Agents window with an optional welcome heading when you create a new session in the Agents window. The heading shows one of five phrases, such as What are we building? or Let's ship something, and can append your first name from your GitHub profile. Welcome phrases are off by default. When enabled, you can use Set Welcome Name from the heading, its context menu, or the Command Palette to choose a different name. The name syncs across devices. Clear it to use your GitHub first name when available, or a generic phrase when your profile has no name. When screen reader optimized mode is active, VS Code announces the heading once when the composer appea

• + 세션 컴포저 레이아웃을 개선하여 제어 항목을 그룹화하고 단축키를 추가한다 # 에이전트
  원문

  [Agents] Improved session composer controls (Experimental): Settings: sessions.chat.experimental.newSessionComposerLayout , sessions.chat.unifiedWorkspacePicker.enabled (Agents window only) The experimental session composer layout reduces the need to move between separate parts of the Agents window. In a new session, the workspace, branch, worktree, and harness controls are grouped above the chat input. Enable both settings to use the layout. The unified workspace picker also has commands for moving directly to its controls. Use ⌘K ⌘F (Windows, Linux Ctrl+K Ctrl+F) to focus the workspace picker or ⌘K ⌘H (Windows, Linux Ctrl+K Ctrl+H) to focus the harness picker. Hovers show your configured keybindings, and each picker's context menu provides a Configure Keybinding action. These keyboard improvements require only the unified workspace picker, not the experimental composer layout. Starting with t

• + 에이전트 오케스트레이션의 한도를 높여 복잡한 워크플로우가 중단되지 않도록 개선한다 # 에이전트
  원문

  [Agents] More reliable agent orchestration (Experimental): Setting: chat.agentHost.agentOrchestrationLimits Coordination-heavy agent workflows are less likely to stop before the work is complete. This release raises the process-wide limits for agent-created sessions, chats, inter-session messages, and recursive session creation. For example, an agent can classify CI failures, delegate each fix to a separate session, and coordinate those sessions without exhausting the message limit before the work is complete. Reaching a limit blocks new orchestration actions but does not interrupt work that is already running. The setting defaults to on, and changes apply without restarting the Agent Host. Set it to off to remove the orchestration limits. Confirmation requirements and input validation continue to apply.

• + URL을 통해 에이전트 세션과 프롬프트가 포함된 상태로 Agents window를 여는 기능을 추가한다 # 에이전트
  원문

  [Agents] Open a new session in the Agents window from a URL: Open the Agents window from another tool or webpage with a workspace and an editable prompt ready for review. Use a URL with the agents/new path and the optional prompt and workspace query parameters. For example, this URL opens a new session with no workspace and the prompt already entered: VS Code places the decoded prompt in the New Session composer without sending it. To select a workspace, add the workspace parameter with a URL-encoded folder URI. If the URL omits workspace, the draft uses No Workspace. An existing occupied composer remains unchanged, so opening a link does not overwrite work in progress. Learn more about opening VS Code with URLs. Note: Use vscode-insiders:// instead of vscode:// for VS Code Insiders.

외 16개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Gemini 3.8 Flash, 3.5 Flash Lite 모델을 지원한다 # 에이전트
  원문

  [What's Changed] Feat/gemini 3.8 flash 3.5 flash lite by @DavidAPierce in https://github.com/google-gemini/gemini-cli/pull/29443

외 18개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Enterprise plan Standard 또는 Usage-Based Chat seat을 가진 멤버가 Claude를 직접 메시지할 수 있게 하며, Claude Code가 포함된 시트가 더 이상 필요하지 않다 # 에이전트
  원문

  [What's changed] [Claude Tag] Added direct messages with Claude for members on an Enterprise plan Standard or Usage-Based Chat seat who also have Cowork; a seat that includes Claude Code is no longer required

• + Bedrock 및 Vertex AI 세션에서 관리자가 기본 모델에 대한 접근 권한을 제거할 때, 실패하는 대신 동일한 등급의 이전 모델로 전환하도록 개선한다 # 에이전트
  원문

  [What's changed] Improved Bedrock and Vertex AI sessions to switch to an older available model of the same tier, instead of failing, when an admin removes access to the default model; session titles and summaries now fall back with it

• + Claude Desktop과 같은 앱에서 호스팅된 세션에 `/btw` 질문을 할 때 현재 진행 중인 턴을 인식하도록 개선한다 # 에이전트
  원문

  [What's changed] Improved Remote Control: a `/btw` side question asked of a session hosted by an app such as Claude Desktop now sees the turn in progress, not only the last finished one

• + 자동 모드에서 서브에이전트의 실행이 보고서를 호출자에게 전달하는 즉시 종료되도록 개선한다 # 에이전트
  원문

  [What's changed] Improved subagents in auto mode: a subagent's run now ends as soon as it hands its report back to its caller, instead of taking extra turns that reach no one

• + 백그라운드에서 실행 중인 세션에 대해 `/resume` 및 `claude --resume` 명령을 개선하여 세션을 열고 다음 턴으로 프롬프트를 전송한다 # 에이전트
  원문

  [What's changed] Improved `/resume` and `claude --resume` on a session that is running in the background: they now open that session instead of refusing, and a prompt given with `claude --resume <id> "prompt"` is sent to it as its next turn

• + Code Review의 pull request 리뷰와 `/ultrareview`를 `disableWorkflows`가 켜져 있을 때도 실행하며, 관리자 설정에 의해 덮어쓰기가 되지 않는 경우에만 제외한다 # 에이전트
  원문

  [What's changed] Changed Code Review's pull request reviews and `/ultrareview` to run when `disableWorkflows` is on, unless the machine running the review has it set by its own administrator (MDM or the managed-settings file)

• + Bedrock, Vertex 및 Mantle 시작 시 모델 체크 과정에서 자신을 Claude Code로 식별하도록 수정한다 # 에이전트
  원문

  [What's changed] Changed Bedrock, Vertex and Mantle start-up model checks to identify themselves as Claude Code, like other Claude Code requests

• + Claude가 패널에서 언제든 Problems 패널의 현재 오류와 경고를 읽을 수 있는 온디맨드 진단 도구를 추가한다 # 에이전트
  원문

  [What's changed] [VSCode] Added an on-demand diagnostics tool so Claude in the panel can read the Problems panel's current errors and warnings at any time, not only right after it edits a file

• + 대규모 풀 리퀘스트나 REVIEW.md가 심볼릭 링크인 경우 등 REVIEW.md가 적용되지 않았을 때 Code Review 체크 실행 시 알림을 표시한다 # 에이전트
  원문

  [What's changed] [Code Review] Improved the Code Review check run to say when your repository's REVIEW.md wasn't applied, for example on a very large pull request or when REVIEW.md is a symbolic link

외 127개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Claude Sonnet 5.5 (`claude-sonnet-5-5`)를 추가한다 # 에이전트
  원문

  [What's changed] Added Claude Sonnet 5.5 (`claude-sonnet-5-5`), now the default Sonnet model on the Anthropic API — 1M context, $2/$10 per Mtok with $0.20/Mtok cache reads

• + Ultracode를 `/effort`에서 별도 토글로 변경하여 모든 노력 수준에서 켜져 있게 하며 xhigh 노력를 강제하지 않는다 # 에이전트
  원문

  [What's changed] Changed Ultracode into its own toggle in `/effort` (Tab, or `/effort ultracode [on|off]`): it no longer forces xhigh effort and stays on at any effort level

• + 작업 디렉토리 밖의 읽기 요청 시 'Yes, but ask again next time' 응답을 추가한다 # 에이전트
  원문

  [What's changed] Added a "Yes, but ask again next time" answer to auto mode's prompt before a read outside the working directories, so you can allow that one read and still be asked about later ones

• + `/claude-api`에서 `hillclimb`가 너무 작은 재표현을 무시하고, 추가 페이지를 단일 로컬 파일로 빌드한다 # 에이전트
  원문

  [What's changed] Improved `/claude-api`: `hillclimb` no longer spends rounds on prompt rewordings too small for the eval to measure, and an extra page you ask for beside `report.html` is built as one local file that loads nothing from the network

• + 권한 모드 설정이 없는 경우, 모든 계획과 제공자의 세션에서 대화형 터미널 및 VS Code 세션이 자동 모드로 시작되도록 변경한다 # 에이전트
  원문

  [What's changed] Changed interactive terminal and VS Code sessions to start in auto mode when no permission mode is configured, on every plan and provider; `permissions.defaultMode` still overrides it

• + Claude Tag에서 스레드, 채널 기본값 또는 DM에 대해 "Opus (latest)"와 같은 모델 패밀리 선택을 추가한다 # 에이전트
  원문

  [What's changed] [Claude Tag] Added model family choices such as "Opus (latest)" for a thread, a channel default or your DM, so the choice follows the newest model in that family

• + Claude Tag에서 조직 전체 한도에 포함되는 지출을 분석 지출 예측 차트에 추가하고 한도 사용량을 표시한다 # 에이전트
  원문

  [What's changed] [Claude Tag] Added the spend that counts toward your organization-wide limit to the analytics spend projection chart, with how much of the limit is used

외 93개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + `availableModelsMatch` 관리형 설정을 추가하여, `exact` 설정 시 `availableModels`에 명시된 모델 버전만 허용하도록 한다 # 에이전트
  원문

  [What's changed] Added `availableModelsMatch` managed setting: with `"exact"`, an `availableModels` entry allows only the model version it names, so new releases stay blocked until listed

• + `availableModels`에서 허용되더라도 특정 모델을 차단할 수 있는 `deniedModels` 관리형 설정을 추가한다 # 에이전트
  원문

  [What's changed] Added `deniedModels` managed setting to block specific models, even when `availableModels` allows them

• + 구형 모델용 프롬프트 패턴을 가진 CLAUDE.md 파일, 스킬, 에이전트 및 명령을 점검하기 위해 `/doctor prompt-audit` (또는 `/checkup prompt-audit`)를 추가한다 # 에이전트
  원문

  [What's changed] Added `/doctor prompt-audit` (also `/checkup prompt-audit`) to audit your CLAUDE.md files, skills, agents and commands for prompting patterns written for older models

• + Claude Code 설정에 대한 `prompt-audit`를 개선하여, 오래된 경로, 오래된 명령 및 상충하는 지침 파일이 리포트를 이끌고, Claude Code 문서에 포함된 생각 키워드가 유지되도록 한다 # 에이전트
  원문

  [What's changed] Improved `prompt-audit` on Claude Code configuration: stale paths, stale commands and contradicting instruction files now lead the report, and thinking keywords that Claude Code documents are kept

• + "--system-prompt" 및 "--append-system-prompt"가 텍스트와 -file 형식을 동시에 수용하도록 변경한다 # 에이전트
  원문

  [What's changed] Changed `--system-prompt` and `--append-system-prompt` to accept their text and `-file` forms together; the file's text comes first

• + 실행 중인 cloud session에 리포지토리를 추가할 때, GitHub 계정이 읽기만 가능하고 푸시할 수 없는 비공개 리포지토리를 읽기 전용으로 연결한다 # 에이전트
  원문

  [What's changed] [Cloud sessions] Improved adding a repository to a running cloud session: a private repository your GitHub account can read but not push to now attaches for reading

• + Opus의 컨텍스트 윈도우가 이미 1M이므로 `/model` 선택기에서 Opus 행과 기본 모델의 이름에서 "(1M context)"를 제거한다 # 에이전트
  원문

  [What's changed] Changed the `/model` picker's Opus row and the Default model's name to drop "(1M context)" where Opus already has a 1M context window; the window is unchanged

• + 새로운 routine 스케줄을 정각에서 몇 분 지난 시점으로 기본값으로 설정하며, 정각에 설정된 routine은 몇 분 늦게 시작될 수 있음을 알린다 # 에이전트
  원문

  [What's changed] [Cloud sessions] Changed new routine schedules to default to a few minutes past the hour, with a note that routines set exactly on the hour can start several minutes late

외 86개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + 텔레메트리가 꺼져 있을 때 Anthropic API 직접 연결 시 기본적으로 서버 측 분류기를 사용하도록 auto 모드를 변경한다 # 에이전트
  원문

  [What's changed] Changed auto mode to use the server-side classifier by default on a direct Anthropic API connection when telemetry is off (`CLAUDE_CODE_AUTO_MODE_SERVER=0` opts out)

• + 기존에 리포지토리가 있는 실행 중인 클라우드 세션에 다른 GitHub 소유자의 리포지토리를 연결하는 기능을 추가한다 # 에이전트
  원문

  [What's changed] [Cloud sessions] Added attaching a repository from a different GitHub owner, such as a fork's upstream, to a running cloud session that already has one, including sessions started from Slack

외 84개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + 첫 번째 프롬프트가 매우 큰 세션에서 "Prompt is too long" 복구 시 해당 프롬프트가 요약에 제외되지 않고 자체적으로 요약되도록 개선한다 # 에이전트
  원문

  [What's changed] Improved "Prompt is too long" recovery in sessions dominated by one very large first prompt: that prompt is now summarized on its own instead of being left out of the summary

• + `--agents` 옵션이 인라인 JSON뿐만 아니라 JSON 파일 경로를 수용하며, 빈 `prompt`를 허용하도록 개선한다 # 에이전트
  원문

  [What's changed] Improved `--agents` to accept the path to a JSON file (with `-p`) as well as inline JSON, and to allow an empty `prompt`

• + `/batch`가 WorktreeCreate 훅을 통해 에이전트 작업용 워크트리(worktree)를 제공받는 곳에서 실행되도록 개선한다 # 에이전트
  원문

  [What's changed] Improved `/batch` to run where a WorktreeCreate hook provides the agent worktrees, not only inside a git repository

• + 사용하지 않는 필수 필드를 제거하여 긴 연구 브리프(research brief)에서 `/deep-research`의 신뢰성을 개선한다 # 에이전트
  원문

  [What's changed] Improved `/deep-research` reliability on long research briefs by removing unused required fields from the scope step's output

• + auto mode에서 서버 측 분류기 검토를 거치는 경우, 읽기 전용 및 샌드박스 셸 명령도 해당 검토를 기다리며 차단될 수 있다 # 에이전트
  원문

  [What's changed] Changed auto mode so that, where its classifier review runs server-side, read-only and sandboxed shell commands also wait for that review and are blocked when it flags them

• + `CLAUDE_CODE_AUTO_MODE_SERVER` 환경변수가 Anthropic API 직접 연결 시에도 적용되도록 변경한다 # 에이전트
  원문

  [What's changed] Changed `CLAUDE_CODE_AUTO_MODE_SERVER` to also apply on a direct Anthropic API connection: `0` opts out of the server-side auto mode classifier (the local classifier then counts toward usage), `1` opts in

• + 클라우드 세션에서 컴포저의 모델 메뉴에 Fast 모드 전환 스위치를 추가한다 # 에이전트
  원문

  [What's changed] [Claude Code on the web] Added a Fast mode switch to the composer's model menu in cloud sessions, shown when your plan includes fast mode and the selected model supports it

• + Slack 스레드에서 루틴을 물어볼 때 기본적으로 해당 스레드의 예약된 작업만 보여주도록 변경한다 # 에이전트
  원문

  [What's changed] [Claude Tag] Changed the routine list Claude gives when asked in a Slack thread to show that thread's own scheduled tasks by default instead of every routine in the channel

• + 많은 중간 크기 파일이 감지되도록 CLAUDE.md 시작 알림에 지침 파일들을 함께 카운트하도록 개선한다 # 에이전트
  원문

  [What's changed] Improved the large CLAUDE.md startup notice to also count instruction files together, so many mid-sized files and @-imports are caught

외 167개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + 원격 호스트의 Dev Container에서 에이전트 세션을 실행할 수 있도록 `chat.agentHost.devContainer.enabled` 설정을 추가한다 # 에이전트
  원문

  [Agents] Run agent sessions in Dev Containers on remote hosts: Setting: chat.agentHost.devContainer.enabled (Agents window only) Let agents build and test your remote project with the right tools and dependencies, without duplicating toolchain setup on your laptop or the remote host. This release extends Dev Container sessions from local folders to projects on SSH, Tunnel, and WSL hosts. To get started, enable chat.agentHost.devContainer.enabled and select Use Dev Container from the folder menu in the Agents Window. The remote folder must have a supported Dev Container configuration, and Docker must be available on the remote host. Note: Dev Container sessions are rolling out gradually, so the setting might not be enabled by default for you yet. You can enable the setting manually to try the feature now.

• + 에이전트 세션 목록을 더 빠르게 로드하고 새로고침하도록 개선한다 # 에이전트
  원문

  [Agents] Faster session list loading: The Agents Window opens and refreshes large session lists faster. The agent host now keeps lightweight session and chat metadata in a central catalog, instead of opening every conversation database each time the list is built. Full conversation content stays isolated in the individual session and chat databases. The improvement grows with the number of sessions, because the previous approach did work in proportion to your session count. Measured with around 645 sessions on a development machine:

• + 많은 세션이 있는 경우에도 에이전트 세션 목록을 더 빠르게 로드하고 새로고침하도록 개선한다 # 에이전트
  원문

  [Agents] Faster session list loading: VS Code loads and refreshes large agent session lists faster. The agent host keeps lightweight session and chat metadata in a central catalog instead of opening every conversation database each time the list is built. Full conversation content remains isolated in the individual session and chat databases. The improvement grows with the number of sessions because the previous approach did work in proportion to your session count. Measured with around 645 sessions on a development machine: If you have few sessions, expect a smaller difference. Sessions created before this release are migrated automatically in the background.

• + 에이전트 세션 목록에 Compact View를 활성화하여 더 많은 세션을 표시하도록 한다 # 에이전트
  원문

  [Agents] Compact sessions list: Fit more sessions in the sessions list by enabling Compact View in the sessions list view of the Agents Window. Compact rows show the session title at rest and reveal workspace details when you hover over or focus the row. A row expands when the session needs input or approval, so these requests remain visible. Progress also appears on the row for the chat that owns the work. When you collapse a session, the parent row summarizes progress from its hidden chats.

• + 세션 목록에서 세션이나 중첩된 채팅을 직접 이름 변경하도록 한다 # 에이전트
  원문

  [Agents] Rename sessions and chats in place: Rename a session or nested chat directly in the sessions list. Double-click its title, use the Rename context menu action, or focus the row and press F2 for a session or F2 for a nested chat. Inline validation prevents blank titles, and canceling restores the previous title.

• + 세션 내에서 채팅이 표시되는 방식을 선택할 수 있는 `sessions.showChatTabs` 설정을 추가한다 # 에이전트
  원문

  [Agents] Choose how chats appear in a session (Preview): Setting: sessions.showChatTabs (Agents window only) An agent session can contain multiple chats, each representing a different conversation or context. When a session contains multiple chats, choose the presentation that best fits your workflow from the session header menu: Multiple shows each chat on its own tab. Single shows only the active chat and hides the tab bar. Switching presentations preserves your open chats, active chat, and conversation state. In Single mode, chats that you explicitly open to the side remain independent panes with their own header actions.

외 6개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Claude Opus 5.5(`claude-opus-5-5`)를 추가하며 기본 Opus 모델로 설정한다 # 에이전트
  원문

  [What's changed] Added Claude Opus 5.5 (`claude-opus-5-5`), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads

• + 플랜 모드로 전환하고 첫 계획 프롬프트를 전송하거나 세션의 계획을 보여주는 `/plan` 명령어를 추가한다 # 에이전트
  원문

  [What's changed] [VSCode] Added a typed `/plan` that switches to plan mode, sends a first planning prompt, or shows the session's plan

• + Claude 앱에서 볼 수 있는 클라우드 세션의 파일 저장 위치를 개선한다 # 에이전트
  원문

  [What's changed] [Claude Code on the web] Improved cloud sessions viewed in the Claude app: Claude now saves files meant for you where the app can open them

• + Code Review 체크 실행 시 REVIEW.md 지침이 크기 제한으로 인해 잘리거나 누락된 경우 파일명과 제한 수치를 표시한다 # 에이전트
  원문

  [What's changed] [Code Review] Improved the Code Review check run to say when REVIEW.md instructions were cut or left out of a review for exceeding a size limit, naming the file and the limit

• + cross-session 메시징 시작 시 `--debug-file`이 사용자가 선택한 경로에 디버그 로그를 기록한다는 설명을 포함한 경고를 개선한다 # 에이전트
  원문

  [What's changed] Improved the cross-session messaging startup warning to explain that `--debug-file` writes a debug log to a path you choose

외 109개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Claude API, Enterprise, Bedrock, Vertex, Foundry 및 gateways에서 서버측 분류기를 기본값으로 설정하고, 과금되는 대체 방식에 대해 경고를 표시한다 # 에이전트
  원문

  [What's changed] Changed auto mode for Claude API and Enterprise users, and on Bedrock, Vertex, Foundry and gateways, to default to the server-side classifier, which does not charge for classifier overhead (`CLAUDE_CODE_AUTO_MODE_SERVER=0` opts out on Bedrock, Vertex, Foundry and gateways); warns on billed fallback. See https://code.claude.com/docs/en/auto-mode-classifier-billing

외 1개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + CLAUDE.md가 없을 때 AGENTS.md를 읽는 기능을 추가한다 # 에이전트
  원문

  [What's changed] Added AGENTS.md support: in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead; change it under "Project instructions" in `/config` (not yet on Bedrock, Vertex or Foundry)

• + 검토할 내용이 없을 때 `/ultrareview`의 메시지를 개선하고, 최신 커밋을 검토하는 명령을 제공하며, 새 리포지토리의 첫 커밋을 전체 검토한다 # 에이전트
  원문

  [What's changed] Improved `/ultrareview` when there's nothing to review: messages say which case you're in, offer a command that reviews your latest commit, and a new repository's first commit is reviewed in full

외 85개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + 편집 권한이 있는 공유 아티팩트의 경우, Claude가 별도의 복사본을 발행하는 대신 기존에 업데이트한다 # 에이전트
  원문

  [What's changed] Improved the Artifact tool's guidance so Claude updates a shared artifact in place when you were given edit access to it, instead of publishing a separate copy

• + Memory 대화 상자 내에서 저장된 메모리를 조회, 편집 및 삭제할 수 있게 한다 # 에이전트
  원문

  [What's changed] [VSCode] Added viewing, editing and deleting a saved memory inside the Memory dialog

• + 에이전트 맵을 개선한다: 실행 중인 에이전트의 릴즈(pill) 수와 실패 시 빨간색으로 변하는 기능, 맵이 스크롤될 때 메인 에이전트가 시야에 유지되는 기능, 상태와 종료 시간으로 에이전트 정렬을 추가한다 # 에이전트
  원문

  [What's changed] [VSCode] Improved the agent map: the pill counts running agents and turns red after a failure, the main agent stays in view while the map scrolls, and agents sort by state then end time

• + 수정 가능한 에이전트 결과물을 편집할 때 질문 없이 데이터를 저장하고 페이지를 다시 게시하며, 공개 결과물, 첫 게시 및 삭제 시에는 여전히 질문을 요구한다 # 에이전트
  원문

  [What's changed] Changed scheduled and Run now routine runs to save data to, and republish the page of, an artifact you can edit without asking; public artifacts, first publishes and deletes still ask

• + 텍스트를 입력하지 않고 첨부된 이미지를 전송할 수 있게 한다 # 에이전트
  원문

  [What's changed] [VSCode] Added sending an attached image without typing any text

외 91개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + 설정된 값이 없는 모델에 대해 `/code-review`가 서브에이전트를 생성하는 대신 가벼운 인라인 리뷰 프롬프트를 사용하도록 변경한다 # 에이전트
  원문

  [What's changed] Changed `/code-review` to use leaner inline review prompts for every model that has no tuned settings of its own, instead of spawning many review subagents

• + 로컬 세션에서 artifact를 감시하는 방식을 변경한다 # 에이전트
  원문

  [What's changed] Changed artifact watching in local sessions: a new version published elsewhere no longer starts a turn; Claude learns of it from a later Artifact tool result

• + 셀프 호스팅 러너가 접근 권한이 없는 읽기 전용 저장소에 대해 세션 시작 실패 대신 건너뛰도록 변경한다 # 에이전트
  원문

  [What's changed] Changed self-hosted runners to skip a read-only repository the git host refuses at the access check instead of failing the session start

외 105개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + 워크스페이스와 리포지토리 선택기를 통합하고 `sessions.chat.unifiedWorkspacePicker.enabled` 설정을 추가한다 # 에이전트
  원문

  [Agents] Unified workspace and repository picker (Experimental): Setting: sessions.chat.unifiedWorkspacePicker.enabled (Agents window only) Start agent work from one searchable list of local folders, GitHub repositories, Cloud repositories, and remote targets. Remote connection actions remain available from the Remote entry. Selecting Work in Repository uses a cloud-first workflow. A GitHub repository that is not already local is selected immediately with the Cloud harness, without a clone prompt. If you later choose a local harness, VS Code prompts you to clone the repository and preserves the cloud selection if you cancel.

• + Automations를 기본적으로 활성화하고 공유할 수 있게 기능을 추가한다 # 에이전트
  원문

  [Agents] Automations are now enabled by default and can be shared: Settings: chat.automations.enabled (Agents window only) Automations are now enabled by default, making it easier to streamline repetitive tasks. You can also export and import automations to share them across different environments or with your team. Learn more about Automations in the VS Code docs.

• + 로컬 Dev Containers에서 에이전트 세션을 실행할 수 있는 기능을 추가한다 # 에이전트
  원문

  [Agents] Run agent sessions in local Dev Containers: Setting: chat.agentHost.devContainer.enabled (Agents window only) Keep agent work aligned with your project's toolchain by running a session inside a local folder's Dev Container. The agent uses the environment and dependencies configured for the project instead of those on your local machine. When the setting is enabled, local folders with a supported Dev Container configuration automatically show a folder menu with a Use Dev Container action. Select the action to run the agent session in the folder's Dev Container. Docker must be installed on your machine. Note: Local Dev Container sessions are rolling out gradually, so the setting might not be enabled by default for you yet. You can enable the setting manually to try the feature now.

• + 에이전트 호스트에서 Codex 지원을 확장한다 # 에이전트
  원문

  [Agents] Expanded Codex support in the agent host: Settings: chat.agentHost.codexAgent.enabled , chat.editor.codex.preferAgentHost This release expands Codex support in the agent host, so you can keep the same coding session while changing where and how you work. If you are not signed in to GitHub Copilot and want to use the Codex harness just with your ChatGPT subscription, you also have to enable chat.agentHost.allowSignedOutWhenUsable . Continue across apps: move the same Codex session between the ChatGPT app and VS Code instead of starting a new conversation. Interact with desktop apps from VS Code: if the ChatGPT app is installed and configured for computer use, the Codex harness in VS Code can reuse that setup to interact with apps on your computer. This works with models from either your Copilot subscription or your ChatGPT subscription. Use VS Code tools: Codex can use the full set of t

• + 워크스페이스 내에서 Codex quick chats를 이어가는 기능을 추가한다 # 에이전트
  원문

  [Agents] Continue Codex quick chats in a workspace: Last release, we introduced continuing quick chats in a workspace for Copilot sessions. This release extends the same flow to Codex, so starting project-specific work no longer means abandoning a workspace-less Codex quick chat. Ask Codex to attach a local folder, then choose whether to use the folder directly or create an isolated worktree. After you confirm the change, the same chat and native Codex thread become a workspace session. The session retains its title, conversation history, current request, selected model, and permission mode. Codex then continues the request with access to the project files. Workspace conversion is available for idle Codex quick chats in Interactive mode and supports single-root workspace targets. If the change is canceled or cannot be applied, the original workspace-less chat remains available.

• + 에이전트 세션에서 pull request를 생성하는 기능을 추가한다 # 에이전트
  원문

  [Agents] Create pull requests from agent sessions: Setting: chat.agentMerge.enabled (optional, for agent merge only) Create pull requests from Agent Host sessions in the Agents window using one form to review and edit generated titles and descriptions, choose draft status, and configure available merge options. Create the pull request directly or send the request to your agent, with your preferred options remembered for next time. Note: The Agent Merge option is experimental and is only available when the setting above is enabled.

• + Voice Mode에서 병렬 에이전트 작업을 탐색하고 모니터링하며 세션 간 이동이 가능해진다 # 에이전트
  원문

  [Chat] Improved Voice Mode session awareness (Experimental): Navigate and monitor parallel agent work without leaving Voice Mode. Voice Mode can find recent agent sessions, switch between them by label, and report each session's status.

• + 필요할 때 에이전트 SDK를 다운로드하도록 안내를 개선한다 # 에이전트
  원문

  [Agents] On-demand agent SDK downloads: The Claude and Codex agents in the Agents window rely on an SDK that is not shipped with VS Code and is downloaded when you first need it. Previously, the offer to download it appeared only as part of signed-out account setup, so if you were signed in it would download at first message. The offer now appears whenever the SDK is missing for transparency that a download is needed to continue. When models are already available for that agent, the notification presents two options: select Download or send a message to download the SDK as part of the turn. The wording updates in place if models become available while the notification is showing.

• + 완료된 세션을 정리하는 기능을 추가한다 # 에이전트
  원문

  [Agents] Keep completed sessions organized (Preview): Settings: chat.agentSessions.archiveNudge.enabled , chat.agentSessions.autoMarkAsDoneMergedSessionsAfterDays , chat.agentSessions.autoDeleteArchivedMergedSessionsAfterDays , sessions.markAsDoneConfetti (Agents window only) Keep finished work out of the way while preserving the conversation for later. When all of an inactive session's pull requests have merged, the Agents window can suggest marking the session as done. A first-use guide shows you where to find Mark as Done in the sessions list. Enable chat.agentSessions.archiveNudge.enabled to show these suggestions. To automate cleanup, mark inactive sessions as done after their pull requests merge, and optionally delete them after a separate grace period. Both automatic-cleanup settings are disabled by default. When all pull requests for a session are merged, select Configure Automatic Clea

외 3개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Claude 앱에서 `claude --remote-control` 또는 `/remote-control`로 시작된 세션을 포크(fork)하는 기능을 추가한다 # 에이전트
  원문

  [What's changed] 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

• + Bedrock, Vertex, Foundry에서 자동 모드 시 기본적으로 로컬 분류기를 사용하도록 변경한다 # 에이전트
  원문

  [What's changed] 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

• + Claude가 요청받지 않아도 관련 공용 채널을 스스로 감시하기 시작한다 # 에이전트
  원문

  [What's changed] [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

• + 다른 코드가 변경되는 동작에 의존하는 경우, 수정 제안이 해당 동작을 유지해야 함을 명시하도록 개선한다 # 에이전트
  원문

  [What's changed] [Code Review] Improved suggested fixes to say what the fix must keep working when other code depends on the behavior being changed

• + 두 번째 영향받는 위치를 가리키는 리뷰 코멘트를 완전한 문장으로 표현하도록 개선한다 # 에이전트
  원문

  [What's 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

외 59개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Claude Code Remote 세션에서 fast mode를 추가한다 # 에이전트
  원문

  [What's changed] 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

• + 서브에이전트가 호출자에게 보고할 때 안전 분류기가 검토하는 전용 핸드백 호출을 사용하도록 자동 모드를 변경한다 # 에이전트
  원문

  [What's changed] 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

• + Pro 플랜에서 기본 동적 워크플로우 크기를 작게 설정하고, 중간 크기 가이드라인을 15에서 10 에이전트로 낮춘다 # 에이전트
  원문

  [What's changed] Changed the default dynamic workflow size to small on Pro plans and lowered the medium size guideline from 15 to 10 agents

• + Claude Tag 관리자 설정에서 Environment picker를 개선한다 # 에이전트
  원문

  [What's changed] [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

외 92개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.

• + Bash 툴이 파일 수정을 수행할 때 변경된 파일의 차이점을 Bash 툴 결과에 포함한다 # 에이전트
  원문

  [What's changed] Added a diff of the files a Bash command changed to the Bash tool result when the Bash tool handles file edits (setting `bashEditDiffEnabled`)

• + 추론 중심의 분산 처리를 위해 Workflow 툴의 실행당 동시 에이전트 제한을 늘리는 `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` 환경변수를 추가한다 # 에이전트
  원문

  [What's changed] Added `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) to raise the Workflow tool's per-run concurrent agent limit for inference-bound fan-outs

외 96개 항목은 원문에서 볼 수 있습니다.

AI가 만든 요약입니다. 원문과 다르면 오류 제보해 주세요.