Claude Code Changelog

Live streaming of foreground subagent tool calls to Remote Control; critical security fixes for file access and plugin paths; new CLI commands and enhanced usage/cost metrics; improved stability and performance.

Key bug fixes and reliability improvements deliver a more stable and consistent user experience. These updates enhance overall application performance.

Enhanced security with a new restricted mode, expanded cross-session messaging, and numerous fixes for session stability, data privacy, and agent interaction.

New tools for drafting feedback reports, profiling API costs, and managing Admin API resources. This release also brings numerous fixes for stability, performance, and user experience across various features.

A new Auto mode tab for editing classifier rules in /permissions, improved warnings for Bash allow rules, and fixes for various issues including transcript slowdowns, background session reliability, and tool call reporting.

Resolution of a critical startup crash on Linux distributions shipping glibc 2.44, including Arch Linux, CachyOS, and Fedora Rawhide. Enhanced application stability for affected systems.

New settings for a customizable model picker and detailed loop usage insights, alongside keyless organizational sign-in. This release also brings crucial reliability and performance improvements across the board.

Overall stability and performance improvements. General bug fixes and reliability enhancements for an improved user experience.

Extensive bug resolutions and overall reliability enhancements across the application. Users can anticipate a more stable and robust development experience.

New fullscreen setup for more environments and a Python project migration tool. Also, various fixes for Bedrock, session management, and improved cross-session communication on Windows.

A new `keybindingFlavor` setting offers Bash-like Ctrl+W prompt behavior. This release also features critical fixes for memory leaks, self-hosted runner stability, and Remote Control session reliability.

A new "Concise" output style for direct, results-first responses without preamble, and a fix for prompt caching issues when using an LLM gateway or custom base URL.

Cross-session idle notifications are here, allowing one session to ping another when it's next free. This release also brings crucial stability fixes for sessions and improved fullscreen mode.

An optional spellcheck setting for prompt input is now available. This release also brings improved memory and CPU usage for background cloud sessions, along with numerous user interface and permission dialog fixes.

New GitLab merge request badge and automatic session continuation at usage limits lead a release focused on enhanced security, improved stability, and user experience fixes.

New GitLab merge request support and opt-in memory limits for Bash tools on Linux. This release also includes fixes for session stability and performance.

New features include direct session messaging via @mentions, subagent forking by default, and enhanced GitLab integration for tokens and plugins. Stability and security improvements across Remote Control and core components are also delivered.

A critical fix for MCP OAuth sign-in issues, specifically addressing redirect URI mismatches. This ensures successful sign-in for users on servers utilizing pre-registered OAuth clients, such as Slack.

Introducing new plugin marketplace command sources and VS Code sidebar session groups. Key fixes include stable streaming responses, preventing crashes from narrow terminals and invalid tool calls, and improved self-hosted runner reliability.

Enhanced security for synced skills and improved cross-session messages, alongside critical fixes for interactive session redraws and Git detection on Windows.

This release brings a refreshed slash-command menu and significant performance improvements for a smoother experience. Key fixes include resolving Bash command failures on GitHub runners and correcting /tui conversation retrieval.

This update focuses on key stability improvements and addressing various reported issues. Users can expect a more reliable and consistent application experience.

Usage warnings now include spend limit details like cap, reset, and operator messages. Remote Control users can now initiate conversations by name. This release also features numerous stability and quality-of-life improvements.

Self-hosted runners for sessions on your own machines, plus cross-session messaging and agent discovery. Advanced sandbox credential masking and critical fixes for long project paths and sandbox security are also included.

New /teleport hint for local session continuation and owner wildcard marketplace entries. Enhanced security fixes prevent permission bypasses and sandbox escapes, alongside improved session management and context handling.

Enhanced agent security through stricter isolation for subagents and improved permission checks for inter-agent communication. Also, fixes for connection reliability, accurate usage reporting, and better diff views.

A new Focus view for cleaner chat interactions, improved security with critical Bash and PowerShell permission fixes, and numerous reliability enhancements across the platform.

This update delivers key bug fixes and general reliability improvements. Users will notice a more stable and responsive application experience.

This release makes the new Opus 5 model the default, offering 1M context and fast mode. It also introduces deeper nested subagents, network allowlisting for sandboxed commands, and numerous stability fixes.

Background code reviews now keep your conversation clean. Important fixes address Windows path corruption, accidental conversation discards, and improved screen reader announcements, enhancing stability and usability.

Emoji shortcode autocomplete in prompts for faster input. The release includes crucial reliability improvements, such as warnings for failed transcript writes and robust fixes for Windows auto-updates.

This release introduces a new `sandbox.filesystem.disabled` setting, fixes major slowdowns in long sessions, and prevents auto mode command denials after OAuth token expiry, improving overall stability.

Verification and code review tools now require manual invocation, providing users with greater control over when these actions occur. This update ensures a more intentional workflow.

Enhanced safety features, including new controls for difficult conversations and mandatory prompts for certain Docker commands. Extensive security and stability improvements for shell tools and background sessions are also included.

Forking now creates independent background sessions, and /subtask handles in-session delegation. New auto-mode reset and guardrails for runaway tool calls enhance safety and control, alongside numerous stability fixes.

New subagent thinking and text are now included in stream-json output. This release also features important fixes for session stability, agent control, plugin reliability, and cost-billing accuracy.

New: a live elapsed-time counter for tool calls, so you can visibly track progress. This release also brings significant fixes for stability, security, and the overall user experience.

Resolution of a blocking issue for UI dialogs within background agent sessions. This update provides seamless interaction with dialogs, such as /model, ensuring uninterrupted workflows.

Screen reader support, vim insert mode key remapping, and mouse-click support for menus headline this release. Significant fixes for background agent reliability, context window accuracy, and performance improvements are also included.

Default auto mode availability, a switch to Opus 4.8 for Bedrock/Vertex/Foundry, and vital fixes for terminal freezing, auto-updater issues, and agent stability.

Path suggestions added to /cd and a new /doctor check for trimming CLAUDE.md files. This release also brings improved git push workflows, faster background agent upgrades, and fixes for login, model picker, and long-running tool timeouts.

Major agent view improvements, with pull request linking and clearer status. A new /doctor setup checkup is available, alongside enhanced auto mode safety for critical file operations. Multiple fixes address Windows worktree issues and system stability.

Enhanced stability for remote workers with a fix for unexpected terminations during SessionStart hooks in headless sessions. Reliable operation during critical startup.

Warnings for expiring login and enhanced background session stability, including fixes for unresponsiveness, macOS stalling, and preserved subagent work. This update delivers significant reliability and performance improvements.

A new "Dynamic workflow size" setting for better control over agent counts, alongside critical fixes for remote control, session management, and installation stability.

Significant improvements to Sonnet 5 conversation sessions. A key change: the removal of mid-conversation system role harness reminders, ensuring a more natural and less interrupted conversational flow for users.

Greater user control over interactive dialogs with no auto-continue, plus a default "Manual" permission mode. Includes robust fixes for background session stability and improved accessibility.

Multi-skill slash commands now process up to five invocations. This release also brings significant reliability improvements, including better error handling for streaming responses, subagent work, and background agents, alongside clearer diagnostics for common issues.

Default background subagent operation allows continued primary session work with completion notifications. New data visualization guidance, general availability of in-browser functionality, and core stability improvements are also included.

Introduction of Sonnet 5 as the new default model. Key features include a native 1M-token context window and promotional pricing of $2/$10 per Mtok through August 31.

Clickable file attachments and organization default models are new. This release also brings significant reliability improvements for background sessions and agents, alongside numerous quality-of-life and stability fixes.

A new option for disabling mouse clicks in fullscreen, alongside voice dictation fixes for macOS and non-spaced languages, and improved Linux voice diagnostics. Also, various stability and plugin management fixes.

Introducing live file path autocomplete in bash mode and enhanced auto-mode transparency with denial reasons. Also, improvements to background agent workflows and essential bug fixes.

A new /rewind command restores conversations after /clear, complemented by a fix for scroll jumping during streaming output. This release also brings significant CPU and memory optimizations, alongside agent reliability and various interface improvements.

General bug fixes and reliability enhancements. Users can expect a more stable and dependable coding environment.

New mouse click support for menus in fullscreen mode and organizational model restrictions. This release also includes critical fixes for structured output reliability and agent stability.

New CLI commands for MCP server login/logout and automatic responses to `!` bash commands enhance scripting and interactive sessions. This release also includes key stability fixes and improved agent control.

Revised API response waiting message, now showing "Waiting for API response" instead of "No response from API." The trigger delay for this hint has also been extended to 20 seconds from 10, for a less interruptive experience.

Major enhancements to auto mode safety, blocking destructive git and infrastructure commands without explicit consent. Also, warnings for deprecated models and various reliability fixes.

Direct configuration of any setting from the prompt with /config key=value syntax. This release also brings smoother text streaming, automatic API retry on connection drops, and critical fixes for file writing on network drives and macOS.

Enhanced stability and performance, with preserved partial responses during connection drops and faster plugin loading in remote sessions. This release also addresses several workflow and UI issues.

Granular control over tool parameters in permission rules and smarter handling of nested directories and skills for complex projects. Significant stability improvements, especially for subagent interactions and diagnostics, are also included.

New conversation-language session titles and better model enforcement. Reliability improvements include Bedrock credential caching, auto mode fixes for Fable 5, and critical copy/paste functionality for tmux over SSH.

Introduction of the `enforceAvailableModels` setting, enabling stricter enforcement of model allowlists. This ensures that the Default model adheres to allowed models and prevents user/project settings from overriding managed configurations.

Detailed usage attribution in the Account & Usage dialog and a new setting for fullscreen scroll acceleration. Further improvements include better model visibility, performance boosts, and enhanced reliability.

Key bug fixes in this update. Fable 5 model names are now automatically normalized by stripping the [1m] suffix for improved clarity. A misleading "sandbox dependencies missing" startup warning on Windows when sandbox is enabled is also no longer present.

Support for nested sub-agents allows for up to five levels of agent-to-agent delegation. This release also brings crucial fixes for 1M context sessions, agent panel responsiveness, and enhanced tool loading.

Access to Fable 5, a powerful new Mythos-class model. This update also includes a fix for session transcripts not saving or appearing in resume when launched from integrated terminals.

New `--safe-mode` disables customizations for troubleshooting, and a `/cd` command moves sessions without breaking the prompt. This release also features critical fixes for UI navigation, enterprise policy enforcement, and performance across platforms.

Major bug resolutions and reliability enhancements are the focus of this update, providing a more stable and consistent development platform through fewer issues.

Overall improvements to application stability and reliability. A smoother, more dependable experience for all users.

New fallback model settings enhance reliability and resilience against API issues. This release also features improved cross-session security and critical fixes for remote sessions, terminals, and image processing errors.

Significant bug fixes and reliability improvements are the focus of this release, ensuring enhanced app stability and a more dependable user experience.

New capabilities include a command for listing installed plugins and a shortcut for copying answers. This release also features critical fixes for command-line operations, Windows compatibility, and background agent stability.

Enhanced agent session insights, including waiting status details, and critical fixes for silent startup hangs on read-only configurations. This release also brings refined permission rule application and a smoother user interface.

Independent parallel tool calls now prevent single failures from canceling a batch. Usage metrics can now be sliced by custom attributes like team or repo, alongside numerous critical reliability and user experience fixes.

Enhanced safety prompts before writing to shell configs and build-tool files, along with easier editing after grep commands. This release also includes numerous fixes for background agent reliability, performance, and user interface issues.

Internal infrastructure improvements, enhancing the system's foundational architecture. These behind-the-scenes updates ensure long-term stability and future development readiness.

Auto mode's expanded availability across Bedrock, Vertex, and Foundry for Opus 4.7 and Opus 4.8 models. This update broadens the intelligent assistance capabilities for developers using these platforms.

Effortless local plugin loading and scaffolding, plus enhanced agent management with settings.json support and mid-session worktree switching. Numerous stability and usability fixes.

Resolution of an issue preventing successful API calls when using Opus 4.8, specifically addressing unintended modifications to thinking blocks. This ensures smoother operation for Opus 4.8 users.

Introducing dynamic workflows for orchestrating multi-agent tasks, Opus 4.8 with high effort defaults, and a more affordable fast mode. This release brings significant new capabilities, performance enhancements, and numerous stability fixes for background operations.

This release introduces a new option to skip Git LFS downloads, alongside major enhancements to background session stability and usability, plus fixes for memory, update processes, and authentication.

Direct application of code review findings to your working tree for reuse, simplification, and efficiency. This release also brings improved session stability, clearer thinking displays, and several user interface refinements.

Significant internal infrastructure enhancements for improved system stability and efficiency. These behind-the-scenes updates create a stronger foundation for future performance and features.

A new detailed usage breakdown by category helps you understand costs. This release also includes critical security fixes, stability improvements for Bash, and various user experience enhancements.

An important fix addressing a recent regression where the Bash tool consistently returned exit code 127 for some users. This update's benefit: improved overall command execution reliability.

Pinned background sessions now persist, restarting for updates and prioritized under memory pressure. A new /code-review command finds correctness bugs with GitHub integration, alongside improvements to the auto-updater, diff rendering, and critical fixes for PowerShell, enterprise, and paste functionality.

New /code-review command with optional effort levels and improved AskUserQuestion behavior in auto mode. This release also includes critical fixes for Windows PowerShell and background session issues.

Enhanced agent management through tab title attention counts and JSON output for scripting, plus better plugin discovery with pre-installation command previews. This release also includes numerous UI and stability fixes.

Resumable background sessions and improved terminal stability are here. This update also brings faster startup, more robust tooling, and many quality-of-life enhancements.

Plugin dependency enforcement and direct worktree editing are now available, alongside new cost estimates in the plugin marketplace. This release also brings significant improvements to background session consistency and stability, plus several key interface fixes.

Extensive new configuration flags for dispatched background agents enhance control over automated tasks. General improvements include Opus 4.7 as the fast mode default and numerous reliability fixes across agents, plugins, and macOS.

Hook-driven desktop notifications and window titles are now possible. This release also introduces a 'Summarize up to here' option, plus numerous agent management and stability fixes across the board.

Improvements to Agent tool subagent type matching for flexible, case- and separator-insensitive input. This release also includes critical fixes for silently hanging goal commands and enhanced background service reliability.

A new agent session view and a /goal command for setting persistent completion conditions are introduced. This release also includes numerous bug fixes and usability improvements.

Internal stability enhancements. This release focuses on foundational improvements for a more reliable experience.

An essential update addressing the extension's failure to activate on Windows. This fix significantly improves the initial setup experience for Windows users.

Enhanced auto mode control with hard-deny rules and improved multi-server authentication reliability. Significant stability fixes across sessions, plugins, and UI interactions.

New worktree base reference options for flexible code management, including a default change, highlight this update. Key fixes include improved network proxy support, better stability for concurrent sessions, and enhanced remote control.

A new option to disable the fullscreen terminal mode is available. This release also brings significant stability improvements, including graceful shutdown on interrupt and fixes for various terminal display and interaction issues.

Key stability improvements, with a fix for VS Code extension activation on Windows and a resolution for Mantle endpoint authentication failures due to a missing API key header.

New capabilities for direct plugin loading via URL and background updates for package manager installations are here. This release also includes important fixes for session stability and core conversational workflows.

Support for .zip plugin archives via --plugin-dir for streamlined plugin management. This release also includes enhanced MCP server details, a simplified model picker, and many fixes across stability, UI, and tool functionality.

A powerful project state purge command and enhanced OAuth login for various environments are introduced. This release also brings significant reliability fixes, security updates, and better Windows shell integration.

A critical fix for OAuth authentication. Users disabling experimental betas can now authenticate successfully without encountering 401 retry loops.

This release introduces Bedrock service tier selection and the ability to find sessions by pasting PR URLs. It also includes important fixes for branching, structured output, and tool discovery.

Find skills quickly with a new type-to-filter search. Critical memory leaks are resolved, alongside scrollable dialogs and more robust tool integration for a smoother experience.

A new `ultrareview` subcommand for CI/script integration and Windows users no longer require Git for Windows. This release also brings several key fixes for terminal interaction and stability.

Persistent configuration settings are now available across sessions and projects. This release also includes broader PR URL support, significant fixes for pasting issues, and improved agent and tool behavior across the board.

Vim visual mode (v/V) for advanced text selection and operators, alongside extensive custom theme options. Enhancements also include streamlined usage monitoring and critical authentication stability.

Improved model selection persistence across restarts and enhanced plugin management with automatic dependency resolution. This release also brings performance boosts for startup and native searches, alongside several key reliability fixes.

Significantly faster session loading (up to 67%), clearer inline thinking progress, and automatic plugin dependency installation. This release also brings smoother terminal scrolling and several key fixes.

A crash fix for the tool permission dialog when an agent's teammate requests tool access. This update improves application stability, particularly for agent teams.

Major updates including a native CLI for improved performance, new network domain blocking capabilities, and an even faster /ultrareview. This release also features enhanced security measures and many general improvements and bug fixes.

Restoration of Opus-4-7 model availability when using auto mode. This resolves previous 'temporarily unavailable' messages, ensuring smoother operations for all users.

New Opus 4.7 xhigh effort and Auto mode, plus multi-agent /ultrareview for comprehensive code analysis. Many UX enhancements and important bug fixes improve stability and workflow.

This release introduces mobile push notifications and flicker-free rendering with /tui, alongside a dedicated /focus command. Expect enhanced stability, improved plugin management, and numerous bug fixes across the experience.

An improved extended-thinking indicator with a rotating progress hint, offering clearer visual feedback during complex operations.

New session recap context and 1-hour prompt caching are now available. The model can also invoke built-in commands, alongside fixes for login paste, prompt cache TTL, and session naming.

Quicker thinking hints during long operations. This update provides users with earlier progress feedback, enhancing the overall experience during computationally intensive tasks.

Plugin background monitors are now supported for continuous operation. This release also brings significant improvements to system reliability, user interface clarity, and stability across various tools and workflows.

New team onboarding command and OS certificate trust for enterprise TLS. This release brings substantial improvements to session management, tool reliability, and enterprise-grade proxy support, alongside critical fixes for security and stability.

An interactive setup wizard for Google Vertex AI, new Perforce integration, and a Monitor tool for background scripts lead this release. It also includes critical security enhancements for Bash tools and fixes across streaming, UI, and settings.

Significant usability enhancements including flicker-free alt-screen rendering, improved auto-mode denials with a retry option, and faster tool connections. Many core features also received stability and accuracy fixes.

New flicker-free rendering with virtualized scrollback, advanced auto mode permission hooks, and named subagents. This update provides significant stability, performance, and UI improvements for all users.

Resolution of message delivery failures within Cowork Dispatch. This fix ensures consistently reliable communication for all users of the feature.

A new API request header for proxy session aggregation and VCS exclusion lists for Jujutsu/Sapling, alongside critical fixes for session resume, file operations, and general stability.

Conditional hooks for smarter script execution and headless integrations for user questions are now available. This release also brings expanded deep link queries, along with stability and performance fixes.

A new PowerShell tool for Windows (opt-in preview), along with better session management, improved startup, and critical bug fixes for a smoother experience.

New features include conversation transcript search, reactive environment management hooks, and credential scrubbing. This release also brings significant performance boosts and numerous reliability fixes.

New mobile tool approval prompts and a --bare flag for faster scripted calls lead this release, alongside critical fixes for authentication, voice mode, and plugin stability.

New statusline AI rate limit usage and experimental `--channels` support headline this update. Critical fixes include restoring parallel tool results and improved voice mode stability.

Introducing session remote control for browser/phone access, alongside an --console flag for API billing login. This release brings key improvements to CLI stability, memory usage, and conversational flow.

Line-by-line streaming responses and robust security enhancements. This release also includes improved terminal notifications, better sandboxing, and fixes for large session history truncation.