Claude 更新履歴

Radar Ultra

42 件のリリース

2.1.314

CLIGitHub の原文

AI 要約

  • You can now block Write, Edit, and NotebookEdit from changing files outside your working directories, in every permission mode.
  • Very long conversations can be summarized after they sit idle. The summary is used for your next message if the prompt cache has expired; this also works with telemetry off and on Bedrock, Vertex, Foundry, and LLM gateways once the one-hour prompt cache is on.
  • Claude Code handles several common failures more clearly: startup failures with claude -p now show an error and exit with code 1, and sessions retry with lower effort when an organization's limit is below the model's requested effort.
  • Cloud sessions started from the web, desktop, or mobile apps now start even if a setup script exits with an error.
  • Auto mode gives clearer explanations when it blocks a command, including that Claude may retry and that you can review the call in /permissions.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

4.4212.0

App公式の原文

AI 要約

  • SSH connections are more reliable: Claude Code retries some failed setups, gives slow proxy or jump-host routes more time, and holds messages for an unreachable host until it comes back. Messages are not sent before your organization’s settings load.
  • Code now asks you to approve before starting a dev server from launch.json in Auto and Bypass permissions modes when the sandbox or managed settings restrict shell commands.
  • The built-in browser can ask before a website reaches an app on your computer or a device on your local network. This access is controlled per site; the domain-blocking setting can block devices, but not local apps.
  • You can import Cowork notes into Claude memory where device management has enabled the feature, HIPAA restrictions do not apply, and memory import is available.
  • On supported computers, provider API keys saved in Setup are encrypted, and previously saved keys are encrypted at the next launch. Versions below 4.29 cannot read them, so after rolling back to an older version or moving app data to another computer or account, you must enter the key again.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.312

CLIGitHub の原文

AI 要約

  • Claude is less likely to overwrite a large CLAUDE.md, rules file, or included file after reading only its beginning. When writing to one, it now asks to read the whole file or edit it in place.
  • Signing in through the VS Code extension no longer prevents managed settings and organization policy from being fetched for the rest of the session. If organization settings cannot be loaded at startup, the VS Code tab now explains why and offers Retry or Sign in again.
  • Per-directory and paths: skills can now be found after you use Bash cd to move deeper into a Git repository.
  • Resuming a terminal session should no longer lose its first reply in a rare case. Cloud sessions also no longer run an unchanged message twice when you resend it after a failed send.
  • Cloud routine setup now includes an effort-level choice, and cloud environment settings include a Bash editor with colors, line numbers, and a Copy button.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

4.29.0

App公式の原文

AI 要約

  • Fixed several issues that could lose your messages or stop you from continuing a chat, including drafts lost after a failed first message, queued messages lost when Claude Code crashed or stopped responding, and new chats becoming unavailable after an interrupted first reply.
  • Fixed safety and reliability problems with files and commands: deleting a Windows folder no longer risks deleting files outside it through a junction or mounted volume, and shell commands should no longer start or keep running after you press Stop.
  • Improved SSH and WSL sessions: fixed several connection and startup failures, prevented a refused SSH password from being resent, and preserved uncommitted work when archiving a session after a cancelled start.
  • If a Cowork task or Dispatch chat can’t be deleted, it now stays available with an error so you can try again. Deleting a Dispatch chat also removes its earlier copies on this computer.
  • Claude now warns you when you have used 75% of an organization-set token limit; the organization can change the warning threshold. At the limit, the warning says “Token limit reached” and cannot be closed.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.311

CLIGitHub の原文

AI 要約

  • Fixed several issues that could interrupt or stall work: long freezes around large CLAUDE.md list items, delayed interrupts, stalled file edits, and requests that were retried even when the API said not to retry.
  • Sessions that exceed 200K tokens now compact the conversation and continue instead of stopping with a 1M-context usage-credits error. Compaction also more often retries if the model does not provide a summary.
  • Improved background and remote sessions: crashed attached background sessions can restart, Remote Control reports proxy-blocked startup requests sooner, and timeout results show output so far while the command continues in the background.
  • Pro and Max launches within an hour of the last check start faster by skipping two network requests; launches that only specify options and their values also start faster.
  • Added a tip for Explanatory output style, shown when you ask Claude for explanations.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.310

CLIGitHub の原文

AI 要約

  • Requests are more likely to work with some gateways and compatible servers: Claude Code now retries without the effort-2025-11-24 beta value or mid-conversation system messages if the server rejects them.
  • Fixed a cleanup issue that could delete recent files, such as pasted images, when cleanupPeriodDays was set very high.
  • Fixed recovery after a “Prompt is too long” error: compaction now keeps less dense text, such as base64 or SVG, so the next request is less likely to fail for the same reason.
  • On Windows, fixed a case where permission and pre-tool hooks could let a tool call through if Git Bash was missing and the hook command contained only quoted strings.
  • Added the Monitor tool and plugin monitors for Bedrock, Vertex, and Foundry, including when telemetry or nonessential traffic is disabled; long foreground sleeps are also blocked in those cases.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.309

CLIGitHub の原文

AI 要約

  • Sandbox and permission checks were fixed across many cases, including preventing commands or edits from changing protected configuration files and making sure risky commands prompt or are refused when required.
  • Worktree isolation was tightened: some commands that should have been blocked are now refused, and worktrees no longer fetch Git objects on demand. Partial clones fetch objects first when needed.
  • Several session and messaging problems were fixed: queued messages, agent messages waiting to be delivered, Desktop-hosted messages, and renamed sessions are now handled correctly in the cases described.
  • Claude Code now handles several workflows and tools more reliably, including rewind history, nested skills, artifact updates, and hooks on Windows.
  • The VS Code panel now recovers from an incomplete cached copy of its files; the release also fixes empty history entries and improves editing responsiveness in the panel.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.308

CLIGitHub の原文

AI 要約

  • Bug fixes and reliability improvements, though the notes don’t specify what was fixed.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

3.6556.2

App公式の原文

AI 要約

  • This release bundles Claude Code version 2.1.308.
  • The release notes report no user-facing changes in General, Code, Cowork, or 3P.
  • There are no user-facing changes noted.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

3.6556.1

App公式の原文

AI 要約

  • Fixed a Windows Reset App Data issue that could delete files outside Claude’s data folder.
  • Made SSH sessions more reliable: running tasks are no longer ended or archived prematurely, messages sent during connection loss are delivered, session history loads fully, and refused sign-ins are no longer retried in the background.
  • Added Storage settings to show how much disk space Claude uses and help free space by removing inactive chats, Cowork tasks, and Code sessions.
  • Fixed several Cowork task issues: plugins and skills are now found in scheduled and other background tasks, and tasks started from suggestions are no longer silently dropped when a workspace fails to start.
  • Folder permissions are now saved separately for each computer, SSH host, and WSL distribution. Bypass permissions folders start in their default mode until you choose and confirm it again; on SSH and WSL, earlier more-permissive choices no longer carry over.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

3.1623.4

App公式の原文

AI 要約

  • If your organization requires managed permission rules, plugins now ask before using tools they had pre-approved.
  • Requests can now continue without tool search if a gateway or cloud provider rejects that search.
  • A local MCP server with ${VAR} only in its environment values now runs when isLocalDevMcpEnabled is false or allowedPluginMcpServers is set. Servers with variables in their command or arguments are still stopped.
  • Fixed GitHub MCP server write tools closing their form without creating anything.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.307

CLIGitHub の原文

AI 要約

  • MCP connections are more reliable: Claude Code now recovers from sleep-related connection failures, waits for SDK servers’ latest tools when needed, and explains macOS keychain lock issues. A later run can connect after you unlock the keychain without signing in again.
  • HTTP MCP tool calls are no longer retried indiscriminately after a connection closes mid-call; retries are limited to read-only or idempotent tools.
  • Claude Code can recover from unparseable tool calls by retrying up to twice before ending the turn with an error.
  • Some stability and session issues are fixed, including autocompact errors after reading large results and cleared conversations returning after a runner session restarts.
  • Code Review adds @claude review stop (also off or pause) to end an @claude review always subscription on a pull request without starting another review.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.306

CLIGitHub の原文

AI 要約

  • Fixed a problem where claude -p could say “Not logged in” in a cloud session if started without that session’s environment variables.
  • In sessions with many MCP tools, fixed an API error being mislabeled as “prompt is too long” and prevented an unnecessary compaction.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

3.1623.1

App公式の原文

AI 要約

  • This release bundles Claude Code version 2.1.304.
  • The release notes report no user-facing changes in General, Code, Cowork, or 3P.
  • No user-facing changes are listed for this release.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.305

CLIGitHub の原文

AI 要約

  • On Linux, sandboxed commands can now create user namespaces for tools such as Chromium and rootless Podman. Note: this lets those commands get around the sandbox’s file protections.
  • Requests can now succeed when a provider or gateway rejects tool search: Claude Code retries without tool search, addressing failures introduced in version 2.1.304.
  • Unattended retry mode can now recheck whether a usage limit has been lifted before its reset, using the new CLAUDE_CODE_RETRY_WATCHDOG_USAGE_LIMIT_RECHECK_MS setting.
  • Several reliability issues are fixed: resumed workflows no longer rerun completed agents, cloud-provided MCP servers are no longer replaced by same-named servers from your settings, and new worktrees in bare repositories use the default branch rather than sometimes using local HEAD.
  • /usage now shows one-line fixes that identify what setting caused each warning; press f to apply the fixes. The same fixes are available in VS Code and /doctor.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.304

CLIGitHub の原文

AI 要約

  • Added a safety check before resuming some inactive conversations on Pro and Max: if resuming would use a noticeable share of your five-hour usage limit, you can resume anyway or start a new conversation and refer Claude to the previous one.
  • Added a guided /auto-mode-setup that looks at your project and recent sessions to help auto mode understand your environment. It is also available on Bedrock, Vertex, Foundry, the Claude apps gateway, and sessions with telemetry turned off.
  • Added /claude-api managed-agents-secure to compare a Managed Agents agent’s settings with Anthropic’s security guide; it does not change the settings.
  • Fixed a risky rm -r case: removing a mounted drive or a folder at the top of locations such as /Volumes, /media, or WSL’s /mnt/c now gets the safety prompt.
  • Improved cloud sessions so a new conversation no longer waits for local MCP servers from synced plugins to start, unless a server is marked alwaysLoad: true.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

2.1.303

CLIGitHub の原文

AI 要約

  • The release includes bug fixes and reliability improvements.

公式ノートをもとに AI がまとめたものです。原文が優先されます。

Radar Ultra
OpenAI とは関係のない非公式ツールです。公開投稿からまとめており、元の投稿を正式な情報として参照してください。