Claude Code: What's New in April 2026
A summary of all updates, new features, and bug fixes released for Claude Code during April 2026.
New Features in April 2026
- Added support for selecting a Bedrock service tier (`default`, `flex`, or `priority`) via the `ANTHROPIC_BEDROCK_SERVICE_TIER` environment variable.(v2.1.122)
- Enabled finding the source session for a created PR by pasting the PR URL into the `/resume` search box (for GitHub, GitHub Enterprise, GitLab, and Bitbucket).(v2.1.122)
- Introduced a hint in `/mcp` to remove duplicate servers when an MCP server is manually added with the same URL as a hidden `claude.ai` connector.(v2.1.122)
- Added the `claude_code.at_mention` log event for tracking @-mention resolution in OpenTelemetry.(v2.1.122)
- Added `alwaysLoad` option to MCP server config to force immediate loading of all tools from that server.(v2.1.121)
- Introduced `claude plugin prune` command to remove orphaned auto-installed plugin dependencies; `plugin uninstall --prune` now cascades.(v2.1.121)
- Added a type-to-filter search box to the `/skills` command for easier navigation of long skill lists.(v2.1.121)
- Enabled PostToolUse hooks to replace tool output for all tools via `hookSpecificOutput.updatedToolOutput`, expanding beyond MCP-only functionality.(v2.1.121)
- Enabled support for X.509 certificate-based Workload Identity Federation (mTLS ADC) for Vertex AI connections.(v2.1.121)
- Enabled `CLAUDE_CODE_FORK_SUBAGENT=1` to function correctly in non-interactive sessions for SDK and `claude -p`.(v2.1.121)
- The `--dangerously-skip-permissions` flag now suppresses prompts for writes to `.claude/skills/`, `.claude/agents/`, and `.claude/commands/`.(v2.1.121)
- Enabled iTerm2's "Applications in terminal may access clipboard" setting via `/terminal-setup` to ensure `/copy` functionality works, including from tmux.(v2.1.121)
- The terminal tab session title is now generated based on your configured `language` setting.(v2.1.121)
- Enabled `mcp_authenticate` in the SDK to support `redirectUri` for custom scheme completion and claude.ai connectors.(v2.1.121)
- Added `stop_reason`, `gen_ai.response.finish_reasons`, and `user_system_prompt` (gated by `OTEL_LOG_USER_PROMPTS`) to OpenTelemetry LLM request spans.(v2.1.121)
- Enabled Voice dictation in VSCode to respect the `accessibility.voice.speechLanguage` setting when no Claude Code language is configured.(v2.1.121)
- Enabled `/context` in VSCode to open a native token usage dialog.(v2.1.121)
- Added the non-interactive `claude ultrareview [target]` subcommand for running `/ultrareview` from CI or scripts, printing results to stdout.(v2.1.120)
- Enabled skills to reference the current effort level using the `${CLAUDE_EFFORT}` variable within their content.(v2.1.120)
- Added support for pointing the footer PR badge at a custom code-review URL via the new `prUrlTemplate` setting.(v2.1.119)
- Enabled hiding the working directory in the startup logo using the `CLAUDE_CODE_HIDE_CWD` environment variable.(v2.1.119)
- Expanded support for the `--from-pr` flag to accept GitLab merge-request, Bitbucket pull-request, and GitHub Enterprise PR URLs.(v2.1.119)
- Enabled `--print` mode to honor the agent's `tools:` and `disallowedTools:` frontmatter, aligning behavior with interactive mode.(v2.1.119)
- Enabled `--agent <name>` to honor the agent definition's `permissionMode` for built-in agents.(v2.1.119)
- Enabled auto-approval for PowerShell tool commands when operating in permission mode, matching Bash behavior.(v2.1.119)
- Introduced `duration_ms` (tool execution time) to `PostToolUse` and `PostToolUseFailure` hook inputs.(v2.1.119)
- Enabled plugins pinned by another plugin's version constraint to auto-update to the highest satisfying git tag.(v2.1.119)
- Updated slash command suggestions to highlight the characters that matched the user's query.(v2.1.119)
- Updated slash command picker to wrap long descriptions onto a second line instead of truncating them.(v2.1.119)
- Updated `owner/repo#N` shorthand links in output to use the git remote's host instead of defaulting to github.com.(v2.1.119)
- Updated OpenTelemetry events: `tool_result` and `tool_decision` now include `tool_use_id`; `tool_result` also includes `tool_input_size_bytes`.(v2.1.119)
- Added vim visual mode (`v`) and visual-line mode (`V`) with selection, operators, and visual feedback(v2.1.118)
- Introduced merged `/cost` and `/stats` into `/usage`, which remain as typing shortcuts that open the relevant tab(v2.1.118)
- Enabled creation and switching between named custom themes from `/theme`, allowing hand-editing JSON files in `~/.claude/themes/`, and theme shipping via plugin `themes/` directories(v2.1.118)
- Enabled hooks to invoke MCP tools directly via `type: "mcp_tool"`(v2.1.118)
- Added `DISABLE_UPDATES` environment variable to completely block all update paths, including manual `claude update`, offering stricter control than `DISABLE_AUTOUPDATER`(v2.1.118)
- Enabled WSL on Windows to inherit Windows-side managed settings via the `wslInheritsWindowsSettings` policy key(v2.1.118)
- Updated Auto mode to include `"$defaults"` in `autoMode.allow`, `autoMode.soft_deny`, or `autoMode.environment` to add custom rules alongside the built-in list instead of replacing it(v2.1.118)
- Added a "Don't ask again" option to the auto mode opt-in prompt(v2.1.118)
- Added `claude plugin tag` command to create release git tags for plugins with version validation(v2.1.118)
- Enabled `--continue`/`--resume` to find sessions that added the current directory via `/add-dir`(v2.1.118)
- Enabled `/color` to sync the session accent color to claude.ai/code when Remote Control is connected(v2.1.118)
- Enabled the `/model` picker to honor `ANTHROPIC_DEFAULT_*_MODEL_NAME`/`_DESCRIPTION` overrides when using a custom `ANTHROPIC_BASE_URL` gateway(v2.1.118)
- Enabled forking of subagents on external builds via the CLAUDE_CODE_FORK_SUBAGENT=1 setting.(v2.1.117)
- Agent frontmatter mcpServers are now loaded for main-thread agent sessions when using the --agent flag.(v2.1.117)
- The /resume command now offers to summarize stale, large sessions before re-reading them, aligning with existing --resume behavior.(v2.1.117)
- Plugin install now attempts to install missing dependencies when a plugin is already installed, instead of stopping.(v2.1.117)
- Plugin dependency resolution is improved: errors now indicate "not installed" with an install hint, and claude plugin marketplace add now auto-resolves missing dependencies.(v2.1.117)
- Enabled agent frontmatter `hooks:` to fire when running as a main-thread agent via `--agent`(v2.1.116)
- Introduced inline progress indicators for the thinking spinner ("still thinking", "thinking more", "almost done thinking")(v2.1.116)
- Added support for `/config` search to match option values (e.g., searching "vim" finds the Editor mode setting)(v2.1.116)
- Enabled `/doctor` to be opened while Claude is responding without waiting for the current turn to finish(v2.1.116)
- Enabled `/reload-plugins` and background plugin auto-update to automatically install missing plugin dependencies from previously added marketplaces(v2.1.116)
- Added a hint surfacing in the Bash tool when `gh` commands hit GitHub's API rate limit(v2.1.116)
- Added `sandbox.network.deniedDomains` setting to block specific domains even when a broader `allowedDomains` wildcard would otherwise permit them(v2.1.113)
- Enabled Remote Control clients to query `@`-file autocomplete suggestions(v2.1.113)
- Improved `/ultrareview` with faster launch via parallelized checks, inclusion of diffstat in the launch dialog, and an animated launching state(v2.1.113)
- Enabled Claude Opus 4.7 xhigh model, accessible via `/effort`, `--effort`, and the model picker, allowing users to tune speed versus intelligence.(v2.1.111)
- Introduced Auto mode for Max subscribers when using Opus 4.7.(v2.1.111)
- Added an interactive slider for `/effort` when called without arguments, supporting arrow-key navigation and Enter confirmation.(v2.1.111)
- Added "Auto (match terminal)" theme option selectable via `/theme` that automatically matches the user's terminal dark/light mode.(v2.1.111)
- Added `/less-permission-prompts` skill to scan transcripts and propose a prioritized allowlist for read-only Bash and MCP tool calls in `.claude/settings.json`.(v2.1.111)
- Added `/ultrareview` skill for comprehensive code review using parallel multi-agent analysis, invoked via no arguments for the current branch or `/ultrareview <PR#>` for a specific GitHub PR.(v2.1.111)
- Auto mode no longer requires the `--enable-auto-mode` flag.(v2.1.111)
- Rolling out PowerShell tool support on Windows; configurable via `CLAUDE_CODE_USE_POWERSHELL_TOOL`. On Linux/macOS, enable with `CLAUDE_CODE_USE_POWERSHELL_TOOL=1` (requires `pwsh`).(v2.1.111)
- Improved `/setup-vertex` and `/setup-bedrock` to show the actual `settings.json` path when `CLAUDE_CONFIG_DIR` is set, seed model candidates from existing pins, and offer a "with 1M context" option.(v2.1.111)
- The `/skills` menu now supports sorting by estimated token count by pressing `t`.(v2.1.111)
- Added `Ctrl+Y` to restore text deleted by `Ctrl+U` (which now clears the entire input buffer).(v2.1.111)
- Added `Ctrl+L` to force a full screen redraw in addition to clearing the prompt input.(v2.1.111)
- The transcript view footer now shows shortcuts `[` (dump to scrollback) and `v` (open in editor).(v2.1.111)
- Added /tui command and tui setting to switch to flicker-free rendering in the same conversation via /tui fullscreen.(v2.1.110)
- Introduced the push notification tool, allowing Claude to send mobile push notifications when Remote Control and "Push when Claude decides" config are enabled.(v2.1.110)
- Added the `autoScrollEnabled` config option to disable conversation auto-scroll in fullscreen mode.(v2.1.110)
- Added an option to show Claude's last response as commented context in the `Ctrl+G` external editor (enabled via /config).(v2.1.110)
- Enabled session recap for users with telemetry disabled (Bedrock, Vertex, Foundry, DISABLE_TELEMETRY).(v2.1.110)
- The Write tool now informs the model when you edit the proposed content in the IDE diff before accepting.(v2.1.110)
- Added recap feature to provide context when returning to a session, configurable in /config and manually invocable with /recap(v2.1.108)
- Enabled the model to discover and invoke built-in slash commands like /init, /review, and /security-review via the Skill tool(v2.1.108)
- Added "verbose" indicator when viewing the detailed transcript (Ctrl+O)(v2.1.108)
- Added a warning at startup when prompt caching is disabled via DISABLE_PROMPT_CACHING* environment variables(v2.1.108)
- Added the `path` parameter to the `EnterWorktree` tool, allowing users to switch into an existing worktree of the current repository.(v2.1.105)
- Added background monitor support for plugins via a top-level `monitors` manifest key that automatically arms at session start or on skill invoke.(v2.1.105)
- Added PreCompact hook support, enabling hooks to block compaction by exiting with code 2 or returning `{"decision":"block"}`.(v2.1.105)
- /proactive is now an alias for /loop.(v2.1.105)
- Added \/team-onboarding command to generate a teammate ramp-up guide from local Claude Code usage.(v2.1.101)
- Enabled OS CA certificate store trust by default, allowing enterprise TLS proxies to work without extra setup.(v2.1.101)
- Introduced auto-creation of a default cloud environment for \/ultraplan and other remote-session features, removing the need for prior web setup.(v2.1.101)
- Added interactive Google Vertex AI setup wizard accessible from the login screen for guided configuration of GCP authentication, project, region, credentials, and model pinning.(v2.1.98)
- Added Monitor tool for streaming events from background scripts.(v2.1.98)
- Added subprocess sandboxing with PID namespace isolation on Linux when `CLAUDE_CODE_SUBPROCESS_ENV_SCRUB` is set.(v2.1.98)
- Added `CLAUDE_CODE_SCRIPT_CAPS` environment variable to limit per-session script invocations.(v2.1.98)
- Added `--exclude-dynamic-system-prompt-sections` flag to print mode for improved cross-user prompt caching.(v2.1.98)
- Added `workspace.git_worktree` to the status line JSON input when operating inside a linked git worktree.(v2.1.98)
- Added W3C `TRACEPARENT` environment variable to Bash tool subprocesses when OTEL tracing is enabled for correct trace parenting.(v2.1.98)
- Enabled LSP client identification to language servers via `clientInfo` in the initialize request.(v2.1.98)
- Enabled Accept Edits mode to auto-approve filesystem commands prefixed with safe environment variables or process wrappers.(v2.1.98)
- Added focus view toggle (Ctrl+O) in NO_FLICKER mode showing prompt, one-line tool summary with edit diffstats, and final response(v2.1.97)
- Added refreshInterval status line setting to re-run the status line command every N seconds(v2.1.97)
- Added workspace.git_worktree to the status line JSON input, set when the current directory is inside a linked git worktree(v2.1.97)
- Added ● N running indicator in /agents next to agent types with live subagent instances(v2.1.97)
- Added syntax highlighting for Cedar policy files (.cedar, .cedarpolicy)(v2.1.97)
- Updated /claude-api skill to cover Managed Agents alongside the Claude API(v2.1.97)
- Added support for Amazon Bedrock powered by Mantle.(v2.1.94)
- Added compact "Slacked #channel" header with a clickable channel link for Slack MCP send-message tool calls.(v2.1.94)
- Added support for the `keep-coding-instructions` frontmatter field for plugin output styles.(v2.1.94)
- Added `hookSpecificOutput.sessionTitle` to `UserPromptSubmit` hooks for setting the session title.(v2.1.94)
- Enabled plugin skills declared via `"skills": ["./"]` to use the skill's frontmatter `name` for the invocation name instead of the directory basename.(v2.1.94)
- Added interactive Bedrock setup wizard accessible from the login screen for guided AWS authentication, region configuration, credential verification, and model pinning.(v2.1.92)
- Added per-model and cache-hit breakdown to the /cost command for subscription users.(v2.1.92)
- Enabled Remote Control session names to default to the user's hostname (e.g. myhost-graceful-unicorn), which can be overridden.(v2.1.92)
- Pro users now see a footer hint when returning to a session after prompt cache expiration, indicating the approximate number of uncached tokens for the next turn.(v2.1.92)
- Added MCP tool result persistence override via `_meta["anthropic/maxResultSizeChars"]` annotation, allowing up to 500K characters for results like DB schemas.(v2.1.91)
- Added `disableSkillShellExecution` setting to completely disable inline shell execution within skills, custom slash commands, and plugin commands.(v2.1.91)
- Added support for multi-line prompts when using `claude-cli://open?q=` deep links, correctly handling encoded newlines.(v2.1.91)
- Enabled plugins to ship executables under the `bin/` directory and invoke them directly as bare commands from the Bash tool.(v2.1.91)
- Added /powerup, which provides interactive lessons teaching Claude Code features with animated demos(v2.1.90)
- Added CLAUDE_CODE_PLUGIN_KEEP_MARKETPLACE_ON_FAILURE environment variable to retain the existing marketplace cache when git pull fails, beneficial for offline use(v2.1.90)
- Added "defer" permission decision to PreToolUse hooks, allowing headless sessions to pause and resume tool calls for re-evaluation.(v2.1.89)
- Enabled flicker-free alt-screen rendering via the CLAUDE_CODE_NO_FLICKER=1 environment variable for virtualized scrollback.(v2.1.89)
- Introduced PermissionDenied hook that fires after auto mode classifier denials, allowing the model to retry by returning {retry: true}.(v2.1.89)
- Added named subagents to @ mention typeahead suggestions.(v2.1.89)
- Enabled non-blocking MCP connection mode using MCP_CONNECTION_NONBLOCKING=true for -p mode, and bounded server connections to 5s.(v2.1.89)
- Auto mode denied commands now show a notification and are visible in /permissions → Recent tab for retrying.(v2.1.89)
- Introduced /buddy for April 1st, a small creature that watches you code.(v2.1.89)
Bug Fixes in April 2026
- Fixed OAuth authentication failing with a 401 retry loop when the environment variable CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 is set.(v2.1.123)
- Fixed `/branch` producing forks that failed with a `tool_use ids were found without tool_result blocks` error when the source session included entries from rewound timelines.(v2.1.122)
- Fixed `/model` failing to show the Effort option for Bedrock application inference profile ARNs and preventing those ARNs from receiving `output_config.effort`.(v2.1.122)
- Fixed Vertex AI / Bedrock returning `invalid_request_error: output_config: Extra inputs are not permitted` errors during session-title generation and other structured-output queries.(v2.1.122)
- Fixed Vertex AI `count_tokens` endpoint returning 400 errors for users operating behind proxy gateways.(v2.1.122)
- Fixed `spinnerTipsOverride.excludeDefault` not correctly suppressing the default time-based spinner tips.(v2.1.122)
- Fixed ToolSearch failing to find MCP tools that connected after the session started in nonblocking mode.(v2.1.122)
- Fixed `!exit` / `!quit` commands in bash mode terminating the CLI instead of executing as a shell command.(v2.1.122)
- Fixed images sent to newer models being incorrectly resized to 2576px per side instead of the maximum allowed 2000px.(v2.1.122)
- Fixed remote control session idle status redrawing too frequently (twice per second), which could flood `tmux -CC` control pipes and cause terminal pauses.(v2.1.122)
- Fixed assistant messages appearing blank in some sessions due to a stale view preference.(v2.1.122)
- Fixed a malformed hooks entry in `settings.json` no longer invalidating the entire settings file.(v2.1.122)
- Fixed voice mode keybindings bound to Caps Lock showing an error because terminals do not deliver Caps Lock as a standard key event.(v2.1.122)
- Fixed unbounded memory growth (multi-GB RSS) that occurred when processing many images in a single session.(v2.1.121)
- Fixed a memory leak in `/usage` that consumed up to ~2GB of memory on machines with large transcript histories.(v2.1.121)
- Fixed a memory leak that occurred when long-running tools failed to emit a clear progress event.(v2.1.121)
- Fixed the Bash tool becoming permanently unusable if the directory Claude was started in was deleted or moved during a session.(v2.1.121)
- Fixed `--resume` crashing on startup in external builds.(v2.1.121)
- Fixed `--resume` failing on large sessions when a transcript line was corrupted by an unclean shutdown; the corrupt line is now skipped.(v2.1.121)
- Fixed `thinking.type.enabled is not supported` error when using Bedrock application inference profile ARNs.(v2.1.121)
- Fixed Microsoft 365 MCP OAuth failures caused by duplicate or unsupported `prompt` parameters.(v2.1.121)
- Fixed scrollback duplication when pressing Ctrl+L or triggering a redraw in non-fullscreen mode across tmux, GNOME Terminal, Windows Terminal, and Konsole.(v2.1.121)
- Fixed claude.ai MCP connectors silently disappearing if the connector-list fetch encountered a transient authentication error at startup.(v2.1.121)
- Fixed "Always allow" rules for built-in tools in remote sessions failing to persist across worker restarts.(v2.1.121)
- Fixed `NO_PROXY` setting not being respected by all HTTP clients when set via `managed-settings.json` in native builds.(v2.1.121)
- Fixed the managed settings approval prompt incorrectly exiting the session upon acceptance; settings are now applied and the session continues.(v2.1.121)
- Fixed `/usage` returning "rate limited" due to a stale OAuth token; tokens now refresh automatically.(v2.1.121)
- Fixed invalid legacy enum values in `settings.json` from invalidating the entire settings file.(v2.1.121)
- Fixed `/usage` dialog content being clipped when no-flicker mode was disabled.(v2.1.121)
- Fixed `/focus` showing "Unknown command" when the fullscreen renderer was off; it now explains how to enable fullscreen rendering.(v2.1.121)
- Fixed embedded grep/find/rg shell wrappers failing if the running binary was deleted mid-session; now falls back to installed tools.(v2.1.121)
- Fixed a regression where pressing Esc during a stdio MCP tool call would close the entire server connection.(v2.1.120)
- Fixed interactive overlays like `/rewind` not responding to keyboard input after launching the tool with `claude --resume`.(v2.1.120)
- Fixed terminal scrollback duplication occurring during resizing, dialog dismissal, or long sessions in non-fullscreen mode.(v2.1.120)
- Fixed telemetry settings (`DISABLE_TELEMETRY` / `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC`) failing to suppress usage metrics for API and enterprise users.(v2.1.120)
- Fixed false-positive "Dangerous rm operation" permission prompts in auto mode for multi-line bash commands containing both a pipe and a redirect.(v2.1.120)
- Fixed long selection menus clipping below the terminal in fullscreen mode; the focused option now remains visible during scrolling.(v2.1.120)
- Fixed the Write tool output collapsing instead of expanding when clicking "+N lines" in fullscreen mode.(v2.1.120)
- Fixed the slash command picker jumping while typing and improved highlighting to only match contiguous substrings.(v2.1.120)
- Fixed the `/plugin` marketplace failing to load when one entry used an unrecognized source format (the entry is now shown with an update prompt upon installation attempt).(v2.1.120)
- Fixed the `find` command in the Bash tool exhausting open file descriptors on large directory trees, which caused host-wide crashes on macOS/Linux native builds.(v2.1.120)
- Fixed pasting CRLF content (from Windows clipboards or Xcode console) from inserting an extra blank line between every line.(v2.1.119)
- Fixed multi-line paste losing newlines when using terminals that employ kitty keyboard protocol sequences inside bracketed paste.(v2.1.119)
- Fixed Glob and Grep tools disappearing on native macOS/Linux builds when the Bash tool was denied via permissions.(v2.1.119)
- Fixed scrolling up in fullscreen mode snapping back to the bottom every time a tool finished execution.(v2.1.119)
- Fixed MCP HTTP connections failing with "Invalid OAuth error response" when servers returned non-JSON bodies during OAuth discovery requests.(v2.1.119)
- Fixed the Rewind overlay incorrectly showing "(no prompt)" for messages that contained image attachments.(v2.1.119)
- Fixed auto mode overriding plan mode when conflicting "Execute immediately" instructions were present.(v2.1.119)
- Fixed async `PostToolUse` hooks that emit no response payload from writing empty entries to the session transcript.(v2.1.119)
- Fixed the spinner remaining active when a subagent task notification became orphaned in the queue.(v2.1.119)
- Fixed `@`-file Tab completion from replacing the entire prompt when used inside a slash command with an absolute path.(v2.1.119)
- Fixed a stray `p` character appearing at the prompt on startup when running in macOS Terminal.app via Docker or SSH.(v2.1.119)
- Fixed `${ENV_VAR}` placeholders in `headers` for HTTP/SSE/WebSocket MCP servers from not being substituted before requests.(v2.1.119)
- Fixed MCP OAuth client secret provided via `--client-secret` not being sent during token exchange for servers requiring `client_secret_post`.(v2.1.119)
- Fixed the Enter key in the `/skills` dialog from closing the dialog instead of pre-filling `/<skill-name>` into the prompt.(v2.1.119)
- Fixed the `/agents` detail view from mislabeling built-in tools unavailable to subagents as "Unrecognized".(v2.1.119)
- Fixed MCP servers originating from plugins from failing to spawn on Windows when the plugin cache was incomplete.(v2.1.119)
- Fixed `/export` from showing the current default model instead of the model the conversation actually used.(v2.1.119)
- Fixed the verbose output setting from not persisting correctly after a restart.(v2.1.119)
- Fixed `/usage` progress bars from overlapping with their corresponding "Resets …" labels.(v2.1.119)
- Fixed plugin MCP servers from failing when `${user_config.*}` references an optional field that was left blank.(v2.1.119)
- Fixed list items containing a sentence-final number from wrapping that number onto its own line.(v2.1.119)
- Fixed `/plan` and `/plan open` from not acting on the existing plan when entering plan mode.(v2.1.119)
- Fixed skills invoked before auto-compaction from being re-executed against the next user message.(v2.1.119)
- Fixed `/reload-plugins` and `/doctor` from reporting load errors for plugins that were disabled.(v2.1.119)
- Fixed Agent tool instances using `isolation: "worktree"` from reusing stale worktrees from prior sessions.(v2.1.119)
- Fixed disabled MCP servers from incorrectly appearing as "failed" in the `/status` command output.(v2.1.119)
- Fixed `TaskList` from returning tasks in arbitrary filesystem order instead of sorting them by ID.(v2.1.119)
- Fixed spurious "GitHub API rate limit exceeded" hints appearing when `gh` output contained PR titles that mentioned "rate limit".(v2.1.119)
- Fixed SDK/bridge `read_file` functionality from not correctly enforcing the size cap on growing files.(v2.1.119)
- Fixed the PR not being linked to the session when working within a git worktree.(v2.1.119)
- Fixed `/doctor` from incorrectly warning about MCP server entries that were overridden by a higher-precedence scope.(v2.1.119)
- Removed a false-positive warning on Windows regarding the need for a `'cmd /c' wrapper` for MCP configuration.(v2.1.119)
- Fixed voice dictation's first recording producing no output on macOS while the microphone permission prompt was visible in VSCode.(v2.1.119)
- Fixed `/mcp` menu hiding OAuth Authenticate/Re-authenticate actions for servers configured with `headersHelper`, and HTTP/SSE MCP servers with custom headers being stuck in "needs authentication" after a transient 401(v2.1.118)
- Fixed MCP servers whose OAuth token response omits `expires_in` requiring re-authentication every hour(v2.1.118)
- Fixed MCP step-up authorization silently refreshing instead of prompting for re-consent when the server's `insufficient_scope` 403 names a scope the current token already has(v2.1.118)
- Fixed an unhandled promise rejection when an MCP server's OAuth flow times out or is cancelled(v2.1.118)
- Fixed MCP OAuth refresh proceeding without its cross-process lock under contention(v2.1.118)
- Fixed macOS keychain race where a concurrent MCP token refresh could overwrite a freshly-refreshed OAuth token, causing unexpected "Please run /login" prompts(v2.1.118)
- Fixed OAuth token refresh failing when the server revokes a token before its local expiry time(v2.1.118)
- Fixed credential save crash on Linux/Windows corrupting `~/.claude/.credentials.json`(v2.1.118)
- Fixed `/login` having no effect in a session launched with `CLAUDE_CODE_OAUTH_TOKEN`; the environment token is now cleared so disk credentials take effect(v2.1.118)
- Fixed unreadable text in the "new messages" scroll pill and `/plugin` badges(v2.1.118)
- Fixed plan acceptance dialog offering "auto mode" instead of "bypass permissions" when running with `--dangerously-skip-permissions`(v2.1.118)
- Fixed agent-type hooks failing with "Messages are required for agent hooks" when configured for events other than `Stop` or `SubagentStop`(v2.1.118)
- Fixed `prompt` hooks re-firing on tool calls made by an agent-hook verifier subagent(v2.1.118)
- Fixed `/fork` writing the full parent conversation to disk per fork; it now writes a pointer and hydrates on read(v2.1.118)
- Fixed Alt+K / Alt+X / Alt+^ / Alt+_ freezing keyboard input(v2.1.118)
- Fixed connecting to a remote session overwriting your local `model` setting in `~/.claude/settings.json`(v2.1.118)
- Fixed typeahead showing "No commands match" error when pasting file paths that start with `/`(v2.1.118)
- Fixed `plugin install` on an already-installed plugin not re-resolving a dependency installed at the wrong version(v2.1.118)
- Fixed unhandled errors from file watcher on invalid paths or fd exhaustion(v2.1.118)
- Fixed Remote Control sessions getting archived on transient CCR initialization blips during JWT refresh(v2.1.118)
- Fixed subagents resumed via `SendMessage` not restoring the explicit `cwd` they were spawned with(v2.1.118)
- Fixed Plain-CLI OAuth sessions dying with "Please run /login" when the access token expires mid-session by reactively refreshing the token on 401 errors.(v2.1.117)
- Fixed WebFetch hanging on very large HTML pages by truncating input before HTML-to-markdown conversion.(v2.1.117)
- Fixed a crash when a proxy returns HTTP 204 No Content, now surfacing a clear error instead of a TypeError.(v2.1.117)
- Fixed /login having no effect when launched with the CLAUDE_CODE_OAUTH_TOKEN environment variable if that token subsequently expires.(v2.1.117)
- Fixed prompt-input undo (Ctrl+_) doing nothing immediately after typing and skipping a state on each undo step.(v2.1.117)
- Fixed NO_PROXY not being respected for remote API requests when running under Bun.(v2.1.117)
- Fixed rare spurious escape/return triggers when key names arrive as coalesced text over slow connections.(v2.1.117)
- Fixed SDK reload_plugins reconnecting all user MCP servers serially.(v2.1.117)
- Fixed Bedrock application-inference-profile requests failing with 400 when backed by Opus 4.7 with thinking disabled.(v2.1.117)
- Fixed MCP elicitation/create requests auto-cancelling in print/SDK mode when the server finishes connecting mid-turn.(v2.1.117)
- Fixed subagents running a different model than the main agent incorrectly flagging file reads with a malware warning.(v2.1.117)
- Fixed Opus 4.7 sessions showing inflated /context percentages and autocompacting too early because the context window size was miscalculated.(v2.1.117)
- Fixed broken column alignment for Devanagari and other Indic scripts rendering in the terminal UI(v2.1.116)
- Fixed Ctrl+- not triggering undo in terminals using the Kitty keyboard protocol (iTerm2, Ghostty, kitty, WezTerm, Windows Terminal)(v2.1.116)
- Fixed Cmd+Left/Right not jumping to line start/end in terminals that use the Kitty keyboard protocol (Warp fullscreen, kitty, Ghostty, WezTerm)(v2.1.116)
- Fixed Ctrl+Z hanging the terminal when Claude Code is launched via a wrapper process (e.g. `npx`, `bun run`)(v2.1.116)
- Fixed scrollback duplication in inline mode that occurred when resizing the terminal or during large output bursts(v2.1.116)
- Fixed modal search dialogs overflowing the screen at short terminal heights, hiding the search box and keyboard hints(v2.1.116)
- Fixed scattered blank cells and disappearing composer chrome in the VS Code integrated terminal during scrolling(v2.1.116)
- Fixed an intermittent API 400 error related to cache control TTL ordering that occurred during parallel request setup(v2.1.116)
- Fixed `/branch` rejecting conversations with transcripts larger than 50MB(v2.1.116)
- Fixed `/resume` silently showing an empty conversation on large session files instead of reporting the load error(v2.1.116)
- Fixed `/plugin` Installed tab showing the same item twice when it appears under Needs attention or Favorites(v2.1.116)
- Fixed `/update` and `/tui` not working after entering a worktree mid-session(v2.1.116)
- Fixed security sandbox auto-allow bypassing the dangerous-path safety check for `rm`/`rmdir` targeting critical system directories like `/` or `$HOME`(v2.1.116)
- Fixed a crash in the permission dialog when an agent teams teammate requested tool permission(v2.1.114)
- Fixed concurrent-call timeout handling where a message for one tool call could silently disarm another call's watchdog(v2.1.113)
- Fixed Cmd-backspace / `Ctrl+U` to correctly delete from the cursor to the start of the line(v2.1.113)
- Fixed markdown tables breaking when a cell contains an inline code span with a pipe character(v2.1.113)
- Fixed session recap auto-firing while composing unsent text in the prompt(v2.1.113)
- Fixed `/copy` "Full response" not aligning markdown table columns correctly for pasting into GitHub, Notion, or Slack(v2.1.113)
- Fixed messages typed while viewing a running subagent being hidden from its transcript and misattributed to the parent AI(v2.1.113)
- Fixed Bash `dangerouslyDisableSandbox` running commands outside the sandbox without triggering a permission prompt(v2.1.113)
- Fixed `/effort auto` confirmation message to correctly state "Effort level set to max"(v2.1.113)
- Fixed the "copied N chars" toast overcounting emoji and other multi-code-unit characters(v2.1.113)
- Fixed `/insights` crashing with `EBUSY` on Windows(v2.1.113)
- Fixed exit confirmation dialog mislabeling one-shot scheduled tasks as recurring; it now correctly shows a countdown(v2.1.113)
- Fixed slash/@ completion menu not sitting flush against the prompt border in fullscreen mode(v2.1.113)
- Fixed `CLAUDE_CODE_EXTRA_BODY` `output_config.effort` causing 400 errors on subagent calls to models that don't support effort and on Vertex AI(v2.1.113)
- Fixed prompt cursor disappearing when `NO_COLOR` is set(v2.1.113)
- Fixed `ToolSearch` ranking so pasted MCP tool names surface the actual tool instead of description-matching siblings(v2.1.113)
- Fixed compacting a resumed long-context session failing with "Extra usage is required for long context requests"(v2.1.113)
- Fixed `plugin install` succeeding when a dependency version conflicts with an already-installed plugin; it now correctly reports `range-conflict`(v2.1.113)
- Fixed "Refine with Ultraplan" not showing the remote session URL in the transcript(v2.1.113)
- Fixed SDK image content blocks that fail to process from crashing the session; they now degrade to a text placeholder(v2.1.113)
- Fixed Remote Control sessions not streaming subagent transcripts(v2.1.113)
- Fixed Remote Control sessions not being archived when Claude Code exits(v2.1.113)
- Fixed `thinking.type.enabled is not supported` 400 error when using Opus 4.7 via a Bedrock Application Inference Profile ARN(v2.1.113)
- Fixed an issue where "claude-opus-4-7 is temporarily unavailable" error appeared when using auto mode.(v2.1.112)
- Fixed permission prompts being triggered by read-only bash commands using glob patterns (e.g., `ls *.ts`) and commands starting with `cd <project-dir> &&`.(v2.1.111)
- Fixed the tool suggesting the closest matching subcommand when invoked with a near-miss typo (e.g., `claude udpate` now suggests `claude update`).(v2.1.111)
- Fixed LSP diagnostics from before an edit appearing after the edit, preventing the model from re-reading already edited files.(v2.1.111)
- Fixed tab-completing `/resume` immediately resuming an arbitrary titled session instead of showing the session picker.(v2.1.111)
- Fixed `/context` grid rendering issues showing extra blank lines between rows.(v2.1.111)
- Fixed `/clear` dropping the session name set by `/rename`, which caused the statusline output to lose `session_name`.(v2.1.111)
- Fixed plugin dependency errors: distinguishing conflicting, invalid, and overly complex version requirements; fixed stale resolved versions after `plugin update`; `plugin install` now recovers from interrupted prior installs.(v2.1.111)
- Fixed Claude calling a non-existent `commit` skill and showing "Unknown skill: commit" for users without a custom `/commit` command.(v2.1.111)
- Fixed 429 rate-limit errors on Bedrock/Vertex/Foundry incorrectly referencing status.claude.com.(v2.1.111)
- Fixed feedback surveys appearing back-to-back after dismissing one.(v2.1.111)
- Fixed bare URLs in bash/PowerShell/MCP tool output being unclickable when the terminal wraps them across lines.(v2.1.111)
- Fixed Windows environment file application: `CLAUDE_ENV_FILE` and SessionStart hook environment files now apply correctly.(v2.1.111)
- Fixed Windows permission rules: paths with drive-letter paths are now correctly root-anchored, and paths differing only by drive-letter case are recognized as the same path.(v2.1.111)
- Fixed terminal display tearing (random characters, drifting input) in iTerm2 + tmux setups when terminal notifications are sent.(v2.1.111)
- Fixed `@` file suggestions re-scanning the entire project on every turn in non-git working directories, and fixed showing only config files in freshly-initialized git repos with no tracked files.(v2.1.111)
- Fixed MCP tool calls hanging indefinitely when the server connection drops mid-response on SSE/HTTP transports.(v2.1.110)
- Fixed non-streaming fallback retries causing multi-minute hangs when the API is unreachable.(v2.1.110)
- Fixed session recap, local slash-command output, and other system status lines not appearing in focus mode.(v2.1.110)
- Fixed high CPU usage in fullscreen when text is selected while a tool is running.(v2.1.110)
- Fixed plugin install not honoring dependencies declared in `plugin.json` when the marketplace entry omits them; /plugin install now lists auto-installed dependencies.(v2.1.110)
- Fixed skills with `disable-model-invocation: true` failing when invoked via /<skill> mid-message.(v2.1.110)
- Fixed `--resume` sometimes showing the first prompt instead of the /rename name for sessions still running or exited uncleanly.(v2.1.110)
- Fixed queued messages briefly appearing twice during multi-tool-call turns.(v2.1.110)
- Fixed session cleanup not removing the full session directory including subagent transcripts.(v2.1.110)
- Fixed dropped keystrokes after the CLI relaunches (e.g. /tui, provider setup wizards).(v2.1.110)
- Fixed garbled startup rendering in macOS Terminal.app and other terminals that don't support synchronized output.(v2.1.110)
- Hardened "Open in editor" actions against command injection from untrusted filenames.(v2.1.110)
- Fixed `PermissionRequest` hooks returning `updatedInput` not being re-checked against `permissions.deny` rules; `setMode:'bypassPermissions'` updates now respect `disableBypassPermissionsMode`.(v2.1.110)
- Fixed `PreToolUse` hook `additionalContext` being dropped when the tool call fails.(v2.1.110)
- Fixed stdio MCP servers that print stray non-JSON lines to stdout being disconnected on the first stray line (regression in 2.1.105).(v2.1.110)
- Fixed headless/SDK session auto-title firing an extra Haiku request when `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` or `CLAUDE_CODE_DISABLE_TERMINAL_TITLE` is set.(v2.1.110)
- Fixed potential excessive memory allocation when piped (non-TTY) Ink output contains a single very wide line.(v2.1.110)
- Fixed /skills menu not scrolling when the list overflows the modal in fullscreen mode.(v2.1.110)
- Fixed Remote Control sessions showing a generic error instead of prompting for re-login when the session is too old.(v2.1.110)
- Fixed Remote Control session renames from claude.ai not persisting the title to the local CLI session.(v2.1.110)
- Fixed paste not working in the /login code prompt (regression in 2.1.105)(v2.1.108)
- Fixed subscribers who set DISABLE_TELEMETRY falling back to 5-minute prompt cache TTL instead of 1 hour(v2.1.108)
- Fixed Agent tool prompting for permission in auto mode when the safety classifier's transcript exceeded its context window(v2.1.108)
- Fixed Bash tool producing no output when CLAUDE_ENV_FILE (e.g. ~/.zprofile) ends with a # comment line(v2.1.108)
- Fixed claude --resume <session-id> losing the session's custom name and color set via /rename(v2.1.108)
- Fixed session titles showing placeholder example text when the first message is a short greeting(v2.1.108)
- Fixed terminal escape codes appearing as garbage text in the prompt input after --teleport(v2.1.108)
- Fixed /feedback retry: pressing Enter to resubmit after a failure now works without first editing the description(v2.1.108)
- Fixed --teleport and --resume <id> precondition errors (e.g. dirty git tree, session not found) exiting silently instead of showing the error message(v2.1.108)
- Fixed Remote Control session titles set in the web UI being overwritten by auto-generated titles after the third message(v2.1.108)
- Fixed --resume truncating sessions when the transcript contained a self-referencing message(v2.1.108)
- Fixed transcript write failures (e.g. disk full) being silently dropped instead of being logged(v2.1.108)
- Fixed diacritical marks (accents, umlauts, cedillas) being dropped from responses when the language setting is configured(v2.1.108)
- Fixed policy-managed plugins never auto-updating when running from a different project than where they were first installed(v2.1.108)
- Fixed images attached to queued messages (sent while Claude is working) being dropped.(v2.1.105)
- Fixed the screen going blank when the prompt input wraps to a second line in long conversations.(v2.1.105)
- Fixed leading whitespace being copied when selecting multi-line assistant responses in fullscreen mode.(v2.1.105)
- Fixed leading whitespace being trimmed from assistant messages, which broke ASCII art and indented diagrams.(v2.1.105)
- Fixed garbled bash output when commands print clickable file links (e.g. Python `rich`/`loguru` logging).(v2.1.105)
- Fixed alt+enter not inserting a newline in terminals using ESC-prefix alt encoding, and Ctrl+J not inserting a newline (regression in 2.1.100).(v2.1.105)
- Fixed duplicate "Creating worktree" text in EnterWorktree/ExitWorktree tool display.(v2.1.105)
- Fixed queued user prompts disappearing from focus mode.(v2.1.105)
- Fixed one-shot scheduled tasks re-firing repeatedly when the file watcher missed the post-fire cleanup.(v2.1.105)
- Fixed inbound channel notifications being silently dropped after the first message for Team/Enterprise users.(v2.1.105)
- Fixed marketplace plugins with `package.json` and lockfile not having dependencies installed automatically after install/update.(v2.1.105)
- Fixed marketplace auto-update leaving the official marketplace in a broken state when a plugin process holds files open during the update.(v2.1.105)
- Fixed "Resume this session with..." hint not printing on exit after `/resume`, `--worktree`, or `/branch`.(v2.1.105)
- Fixed feedback survey shortcut keys firing when typed at the end of a longer prompt.(v2.1.105)
- Fixed stdio MCP server emitting malformed (non-JSON) output hanging the session instead of failing fast with "Connection closed".(v2.1.105)
- Fixed MCP tools missing on the first turn of headless/remote-trigger sessions when MCP servers connect asynchronously.(v2.1.105)
- Fixed `/model` picker on AWS Bedrock in non-US regions persisting invalid `us.*` model IDs to `settings.json` when inference profile discovery is still in-flight.(v2.1.105)
- Fixed 429 rate-limit errors showing a raw JSON dump instead of a clean message for API-key, Bedrock, and Vertex users.(v2.1.105)
- Fixed crash on resume when session contains malformed text blocks.(v2.1.105)
- Fixed `/help` dropping the tab bar, Shortcuts heading, and footer at short terminal heights.(v2.1.105)
- Fixed malformed keybinding entry values in `keybindings.json` being silently loaded instead of rejected with a clear error.(v2.1.105)
- Fixed `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` in one project's settings permanently disabling usage metrics for all projects on the machine.(v2.1.105)
- Fixed washed-out 16-color palette when using Ghostty, Kitty, Alacritty, WezTerm, foot, rio, or Contour over SSH/mosh.(v2.1.105)
- Fixed Bash tool suggesting `acceptEdits` permission mode when exiting plan mode would downgrade from a higher permission level.(v2.1.105)
- Fixed a command injection vulnerability in the POSIX which fallback used by LSP binary detection.(v2.1.101)
- Fixed a memory leak where long sessions retained dozens of historical copies of the message list in the virtual scroller.(v2.1.101)
- Fixed resume/continue losing conversation context on large sessions when the loader anchored on a dead-end branch.(v2.1.101)
- Fixed resume chain recovery bridging into an unrelated subagent conversation when a subagent message landed near a main-chain write gap.(v2.1.101)
- Fixed a crash on resume when a persisted Edit\/Write tool result was missing its file_path.(v2.1.101)
- Fixed a hardcoded 5-minute request timeout that aborted slow backends regardless of API_TIMEOUT_MS.(v2.1.101)
- Fixed permissions.deny rules not overriding a PreToolUse hook's permissionDecision: "ask", preventing hooks from downgrading denies to prompts.(v2.1.101)
- Fixed setting sources without user causing background cleanup to ignore cleanupPeriodDays and delete history older than 30 days.(v2.1.101)
- Fixed Bedrock SigV4 authentication failing with 403 when Authorization headers were set via environment variables or helpers.(v2.1.101)
- Fixed claude -w <name> failing with "already exists" after stale worktree cleanup from a previous session.(v2.1.101)
- Fixed subagents not inheriting MCP tools from dynamically-injected servers.(v2.1.101)
- Fixed sub-agents running in isolated worktrees being denied Read\/Edit access to files inside their own worktree.(v2.1.101)
- Fixed sandboxed Bash commands failing with mktemp: No such file or directory after a fresh boot.(v2.1.101)
- Fixed claude mcp serve tool calls failing with "Tool execution failed" in MCP clients that validate outputSchema.(v2.1.101)
- Fixed RemoteTrigger tool's run action sending an empty body and being rejected by the server.(v2.1.101)
- Fixed several resume picker issues including narrow default view, unreachable preview on Windows Terminal, incorrect cwd, missing session-not-found errors, and terminal title not being set.(v2.1.101)
- Fixed Grep tool ENOENT when the embedded ripgrep binary path becomes stale; now falls back to system rg and self-heals.(v2.1.101)
- Fixed \/btw writing a copy of the entire conversation to disk on every use.(v2.1.101)
- Fixed \/context Free space and Messages breakdown disagreeing with the header percentage.(v2.1.101)
- Fixed several plugin issues including slash commands resolving incorrectly, \/plugin update failing with ENAMETOOLONG, Discover showing installed plugins, directory-source plugins loading stale cache, and skills not honoring context and agent frontmatter.(v2.1.101)
- Fixed the \/mcp menu offering OAuth-specific actions for servers configured with headersHelper; Reconnect is now offered instead.(v2.1.101)
- Fixed ctrl+], ctrl+\, and ctrl+^ keybindings not firing in terminals that send raw C0 control bytes.(v2.1.101)
- Fixed \/login OAuth URL rendering with padding that prevented clean mouse selection.(v2.1.101)
- Fixed rendering issues including flicker in non-fullscreen mode, scrollback wiping, and mouse-scroll escape sequences leaking into the prompt.(v2.1.101)
- Fixed crash when settings.json env values are numbers instead of strings.(v2.1.101)
- Fixed in-app settings writes not refreshing the in-memory snapshot, preventing removed directories from being revoked mid-session.(v2.1.101)
- Fixed custom keybindings not loading on Bedrock, Vertex, and other third-party providers.(v2.1.101)
- Fixed claude --continue -p not correctly continuing sessions created by -p or the SDK.(v2.1.101)
- Fixed several Remote Control issues including worktrees removed on session crash, connection failures not persisting, spurious Disconnected indicator in brief mode, and \/remote-control failing over SSH when only Organization UUID is set.(v2.1.101)
- Fixed \/insights sometimes omitting the report file link from its response.(v2.1.101)
- Fixed the file attachment below the chat input not clearing when the last editor tab is closed in VSCode.(v2.1.101)
- Fixed a Bash tool permission bypass where an escaped flag could be auto-allowed as read-only, potentially leading to arbitrary code execution.(v2.1.98)
- Fixed compound Bash commands bypassing forced permission prompts for safety checks and explicit ask rules in auto and bypass-permissions modes.(v2.1.98)
- Fixed read-only commands with environment variable prefixes not prompting unless the variable is known-safe (e.g., `LANG`, `TZ`).(v2.1.98)
- Fixed redirects to `/dev/tcp/...` or `/dev/udp/...` not prompting instead of auto-allowing.(v2.1.98)
- Fixed stalled streaming responses timing out instead of falling back to non-streaming mode.(v2.1.98)
- Fixed 429 retries burning all attempts too quickly by ensuring exponential backoff applies as a minimum.(v2.1.98)
- Fixed MCP OAuth `oauth.authServerMetadataUrl` configuration override not being honored on token refresh after restart.(v2.1.98)
- Fixed capital letters being dropped to lowercase in xterm and VS Code integrated terminals when the kitty keyboard protocol is active.(v2.1.98)
- Fixed macOS text replacements deleting the trigger word instead of inserting the substitution.(v2.1.98)
- Fixed `--dangerously-skip-permissions` being silently downgraded to accept-edits mode after approving a write to a protected path via Bash.(v2.1.98)
- Fixed managed-settings allow rules remaining active after an admin removed them until process restart.(v2.1.98)
- Fixed `permissions.additionalDirectories` changes not applying mid-session; removed directories now lose access immediately.(v2.1.98)
- Fixed removing a directory from `additionalDirectories` revoking access to the same directory passed via `--add-dir`.(v2.1.98)
- Fixed `Bash(cmd:*)` and `Bash(git commit *)` wildcard permission rules failing to match commands with extra spaces or tabs.(v2.1.98)
- Fixed `Bash(...)` deny rules being downgraded to a prompt for piped commands that mix `cd` with other segments.(v2.1.98)
- Fixed false Bash permission prompts for commands like `cut -d /` and filenames containing `%`.(v2.1.98)
- Fixed permission rules with names matching JavaScript prototype properties (e.g., `toString`) causing `settings.json` to be silently ignored.(v2.1.98)
- Fixed agent team members not inheriting the leader's permission mode when using `--dangerously-skip-permissions`.(v2.1.98)
- Fixed a crash in fullscreen mode when hovering over MCP tool results.(v2.1.98)
- Fixed copying wrapped URLs in fullscreen mode inserting spaces at line breaks.(v2.1.98)
- Fixed file-edit diffs disappearing from the UI on `--resume` when the edited file was larger than 10KB.(v2.1.98)
- Fixed several `/resume` picker issues including uneditable resumes, search state wiping, and transient status text.(v2.1.98)
- Fixed `/export` not honoring absolute paths and `~`, and silently rewriting user-supplied extensions to `.txt`.(v2.1.98)
- Fixed `/effort max` being denied for unknown or future model IDs.(v2.1.98)
- Fixed slash command picker breaking when a plugin's frontmatter `name` is a YAML boolean keyword.(v2.1.98)
- Fixed rate-limit upsell text being hidden after message remounts.(v2.1.98)
- Fixed MCP tools with `_meta["anthropic/maxResultSizeChars"]` not bypassing the token-based persist layer.(v2.1.98)
- Fixed voice mode leaking space characters into input when re-holding the push-to-talk key during transcript processing.(v2.1.98)
- Fixed `DISABLE_AUTOUPDATER` not fully suppressing the npm registry version check and symlink modification.(v2.1.98)
- Fixed a memory leak where Remote Control permission handler entries were retained for the entire session.(v2.1.98)
- Fixed background subagents that fail with an error not reporting partial progress to the parent agent.(v2.1.98)
- Fixed prompt-type Stop/SubagentStop hooks failing on long sessions.(v2.1.98)
- Fixed feedback survey rendering when dismissed.(v2.1.98)
- Fixed Bash `grep -f FILE` / `rg -f FILE` not prompting when reading a pattern file outside the working directory.(v2.1.98)
- Fixed stale subagent worktree cleanup removing worktrees that contain untracked files.(v2.1.98)
- Fixed `sandbox.network.allowMachLookup` not taking effect on macOS.(v2.1.98)
- Fixed `CLAUDE_CODE_MAX_CONTEXT_TOKENS` not honoring `DISABLE_COMPACT` when set.(v2.1.98)
- Fixed VSCode reporting a false-positive "requires git-bash" error on Windows when the path is correctly configured.(v2.1.98)
- Fixed --dangerously-skip-permissions being silently downgraded to accept-edits mode after approving a write to a protected path(v2.1.97)
- Fixed and hardened Bash tool permissions, tightening checks around env-var prefixes and network redirects, and reducing false prompts on common commands(v2.1.97)
- Fixed permission rules with names matching JavaScript prototype properties (e.g. toString) causing settings.json to be silently ignored(v2.1.97)
- Fixed managed-settings allow rules remaining active after an admin removed them until process restart(v2.1.97)
- Fixed permissions.additionalDirectories changes in settings not applying mid-session(v2.1.97)
- Fixed removing a directory from settings.permissions.additionalDirectories revoking access to the same directory passed via --add-dir(v2.1.97)
- Fixed MCP HTTP/SSE connections accumulating ~50 MB/hr of unreleased buffers when servers reconnect(v2.1.97)
- Fixed MCP OAuth oauth.authServerMetadataUrl not being honored on token refresh after restart, fixing ADFS and similar IdPs(v2.1.97)
- Fixed 429 retries burning all attempts in ~13 seconds when the server returns a small Retry-After — exponential backoff now applies as a minimum(v2.1.97)
- Fixed rate-limit upgrade options disappearing after context compaction(v2.1.97)
- Fixed several /resume picker issues: --resume <name> opening uneditable, Ctrl+A reload wiping search, empty list swallowing navigation, task-status text replacing conversation summary, and cross-project staleness(v2.1.97)
- Fixed file-edit diffs disappearing on --resume when the edited file was larger than 10KB(v2.1.97)
- Fixed --resume cache misses and lost mid-turn input from attachment messages not being saved to the transcript(v2.1.97)
- Fixed messages typed while Claude is working not being persisted to the transcript(v2.1.97)
- Fixed prompt-type Stop/SubagentStop hooks failing on long sessions, and hook evaluator API errors displaying "JSON validation failed" instead of the actual message(v2.1.97)
- Fixed subagents with worktree isolation or cwd: override leaking their working directory back to the parent session's Bash tool(v2.1.97)
- Fixed compaction writing duplicate multi-MB subagent transcript files on prompt-too-long retries(v2.1.97)
- Fixed claude plugin update reporting "already at the latest version" for git-based marketplace plugins when the remote had newer commits(v2.1.97)
- Fixed slash command picker breaking when a plugin's frontmatter name is a YAML boolean keyword(v2.1.97)
- Fixed copying wrapped URLs in NO_FLICKER mode inserting spaces at line breaks(v2.1.97)
- Fixed scroll rendering artifacts in NO_FLICKER mode when running inside zellij(v2.1.97)
- Fixed a crash in NO_FLICKER mode when hovering over MCP tool results(v2.1.97)
- Fixed a NO_FLICKER mode memory leak where API retries left stale streaming state(v2.1.97)
- Fixed slow mouse-wheel scrolling in NO_FLICKER mode on Windows Terminal(v2.1.97)
- Fixed custom status line not displaying in NO_FLICKER mode on terminals shorter than 24 rows(v2.1.97)
- Fixed Shift+Enter and Alt/Cmd+arrow shortcuts not working in Warp with NO_FLICKER mode(v2.1.97)
- Fixed Korean/Japanese/Unicode text becoming garbled when copied in no-flicker mode on Windows(v2.1.97)
- Fixed Bedrock SigV4 authentication failing when AWS_BEARER_TOKEN_BEDROCK or ANTHROPIC_BEDROCK_BASE_URL are set to empty strings (as GitHub Actions does for unset inputs)(v2.1.97)
- Fixed Bedrock requests failing with a 403 "Authorization header is missing" error when using AWS_BEARER_TOKEN_BEDROCK or CLAUDE_CODE_SKIP_BEDROCK_AUTH.(v2.1.96)
- Fixed agents appearing stuck after a 429 rate-limit response with a long Retry-After header; the error now surfaces immediately.(v2.1.94)
- Fixed Console login on macOS silently failing when the login keychain is locked or out of sync; the error is now surfaced and `claude doctor` diagnoses the fix.(v2.1.94)
- Fixed plugin skill hooks defined in YAML frontmatter being silently ignored.(v2.1.94)
- Fixed plugin hooks failing with "No such file or directory" when `CLAUDE_PLUGIN_ROOT` was not set.(v2.1.94)
- Fixed `${CLAUDE_PLUGIN_ROOT}` resolving incorrectly for local-marketplace plugins on startup.(v2.1.94)
- Fixed scrollback showing repeated diffs and blank pages in long-running sessions.(v2.1.94)
- Fixed multiline user prompts in the transcript indenting wrapped lines incorrectly under the caret.(v2.1.94)
- Fixed Shift+Space inserting the literal word "space" instead of a space character in search inputs.(v2.1.94)
- Fixed hyperlinks opening two browser tabs when clicked inside tmux running in an xterm.js-based terminal.(v2.1.94)
- Fixed an alt-screen rendering bug that could leave compounding ghost lines during scrolling.(v2.1.94)
- Fixed the `FORCE_HYPERLINK` environment variable being ignored when set via `settings.json` `env`.(v2.1.94)
- Fixed native terminal cursor not tracking the selected tab in dialogs.(v2.1.94)
- Fixed Bedrock invocation of Sonnet 3.5 v2 by correctly using the `us.` inference profile ID.(v2.1.94)
- Fixed SDK/print mode not preserving the partial assistant response in conversation history when interrupted mid-stream.(v2.1.94)
- Fixed CJK and other multibyte text being corrupted with U+FFFD in stream-json input/output when chunk boundaries split a UTF-8 sequence.(v2.1.94)
- Fixed VSCode dropdown menus selecting the wrong item when the mouse was over the list while typing or using arrow keys.(v2.1.94)
- Fixed subagent spawning permanently failing with "Could not determine pane count" after tmux windows were killed or renumbered during a long-running session.(v2.1.92)
- Fixed prompt-type Stop hooks incorrectly failing when the small fast model returns ok:false, restoring preventContinuation:true semantics for non-Stop prompt-type hooks.(v2.1.92)
- Fixed tool input validation failures when streaming emits array/object fields as JSON-encoded strings.(v2.1.92)
- Fixed an API 400 error that could occur when extended thinking produced a whitespace-only text block alongside real content.(v2.1.92)
- Fixed accidental feedback survey submissions caused by auto-pilot keypresses and consecutive-prompt digit collisions.(v2.1.92)
- Fixed misleading "esc to interrupt" hint appearing alongside "esc to clear" when text selection exists in fullscreen mode during processing.(v2.1.92)
- Fixed Homebrew install update prompts to use the correct release channel (claude-code → stable, claude-code@latest → latest).(v2.1.92)
- Fixed ctrl+e jumping to the end of the next line when already at the end of the line in multiline prompts.(v2.1.92)
- Fixed an issue where the same message could appear at two positions when scrolling up in fullscreen mode (for terminals supporting DEC 2026).(v2.1.92)
- Fixed idle-return "/clear to save X tokens" hint showing cumulative session tokens instead of the current context size.(v2.1.92)
- Fixed plugin MCP servers getting stuck "connecting" on session start when they duplicate an unauthenticated claude.ai connector.(v2.1.92)
- Restored unix-socket blocking for sandboxed commands by ensuring the apply-seccomp helper ships in both npm and native builds for the Linux sandbox.(v2.1.92)
- Fixed transcript chain breaks occurring with `--resume` that previously caused silent failures in async transcript writes, leading to lost conversation history.(v2.1.91)
- Fixed `cmd+delete` functionality not deleting text to the start of the line across iTerm2, kitty, WezTerm, Ghostty, and Windows Terminal.(v2.1.91)
- Fixed plan mode in remote sessions losing track of the plan file after a container restart, which previously caused permission prompts on plan edits and an empty plan-approval modal.(v2.1.91)
- Fixed JSON schema validation errors when setting `permissions.defaultMode: "auto"` in settings.json.(v2.1.91)
- Fixed an issue where Windows version cleanup was incorrectly protecting the rollback copy of the active version.(v2.1.91)
- Fixed an infinite loop where the rate-limit options dialog would repeatedly auto-open after hitting the usage limit, preventing session crashes(v2.1.90)
- Fixed regression where `--resume` caused a full prompt-cache miss on the first request for users with deferred tools, MCP servers, or custom agents(v2.1.90)
- Fixed Edit/Write operations failing with "File content has changed" when a PostToolUse format-on-save hook rewrites the file between consecutive edits(v2.1.90)
- Fixed PreToolUse hooks emitting JSON to stdout and exiting with code 2 not correctly blocking the tool call(v2.1.90)
- Fixed collapsed search/read summary badge appearing multiple times in fullscreen scrollback when a CLAUDE.md file auto-loads during a tool call(v2.1.90)
- Fixed auto mode not respecting explicit user boundaries (e.g., "don't push", "wait for X before Y")(v2.1.90)
- Fixed UI crash that occurred when malformed tool input reached the permission dialog(v2.1.90)
- Hardened PowerShell tool permission checks, fixing trailing & background job bypass, -ErrorAction Break debugger hang, archive-extraction TOCTOU, and parse-fail fallback deny-rule degradation(v2.1.90)
- Fixed Edit(//path/**) and Read(//path/**) allow rules to correctly check the resolved symlink target instead of just the requested path.(v2.1.89)
- Fixed voice push-to-talk activation issues with modifier-combo bindings and resolved WebSocket upgrade rejections in voice mode on Windows.(v2.1.89)
- Fixed Edit/Write tools incorrectly doubling CRLF and stripping Markdown hard line breaks (two trailing spaces) on Windows.(v2.1.89)
- Fixed StructuredOutput schema cache bug that caused approximately 50% failure rate when using multiple schemas.(v2.1.89)
- Fixed memory leak where large JSON inputs were retained as LRU cache keys during long-running sessions.(v2.1.89)
- Fixed crash occurring when removing a message from very large session files (over 50MB).(v2.1.89)
- Fixed LSP server zombie state after a crash; the server now restarts on the next request instead of failing until the session restarts.(v2.1.89)
- Fixed prompt history entries containing CJK or emoji being silently dropped when they fell on a 4KB boundary in ~/.claude/history.jsonl.(v2.1.89)
- Fixed /stats undercounting tokens by excluding subagent usage and losing historical data beyond 30 days when the stats cache format changed.(v2.1.89)
- Fixed -p --resume hangs when deferred tool input exceeds 64KB or lacks a deferred marker, and fixed -p --continue not resuming deferred tools.(v2.1.89)
- Fixed claude-cli:// deep links failing to open on macOS.(v2.1.89)
- Fixed MCP tool errors truncating to only the first content block when the server returned multi-element error content.(v2.1.89)
- Fixed skill reminders and other system context being dropped when sending messages with images via the SDK.(v2.1.89)
- Fixed PreToolUse/PostToolUse hooks receiving file_path as a relative path instead of the documented absolute path for Write/Edit/Read tools.(v2.1.89)
- Fixed autocompact thrash loop by detecting when context refills immediately after compacting three times and stopping with an actionable error.(v2.1.89)
- Fixed prompt cache misses in long sessions caused by tool schema bytes changing mid-session.(v2.1.89)
- Fixed nested CLAUDE.md files being re-injected dozens of times in long sessions that read many files.(v2.1.89)
- Fixed --resume crash when the transcript contained a tool result from an older CLI version or an interrupted write.(v2.1.89)
- Fixed misleading "Rate limit reached" message when the API returned an entitlement error; now shows the actual error.(v2.1.89)
- Fixed hooks if condition filtering not matching compound commands (e.g., ls && git push) or commands with env-var prefixes.(v2.1.89)
- Fixed collapsed search/read group badges duplicating in terminal scrollback during heavy parallel tool use.(v2.1.89)
- Fixed notification invalidates not clearing the currently-displayed notification immediately.(v2.1.89)
- Fixed prompt briefly disappearing after submit when background messages arrived during processing.(v2.1.89)
- Fixed truncation of Devanagari and other combining-mark text in assistant output.(v2.1.89)
- Fixed rendering artifacts on main-screen terminals after layout shifts.(v2.1.89)
- Fixed voice mode failing to request microphone permission on macOS Apple Silicon.(v2.1.89)
- Fixed Shift+Enter submitting instead of inserting a newline on Windows Terminal Preview 1.25.(v2.1.89)
- Fixed periodic UI jitter during streaming in iTerm2 when running inside tmux.(v2.1.89)
- Fixed PowerShell tool incorrectly reporting failures when commands like git push wrote progress to stderr on Windows PowerShell 5.1.(v2.1.89)
- Fixed potential out-of-memory crash when the Edit tool was used on very large files (>1 GiB).(v2.1.89)
Improvements in April 2026
- OpenTelemetry numeric attributes on `api_request`/`api_error` log events are now correctly emitted as numbers instead of strings.(v2.1.122)
- Clarified the message shown in `/mcp` when an MCP server remains unauthorized after the browser sign-in flow.(v2.1.122)
- Fullscreen mode now prevents the prompt from jumping scroll back to the bottom after scrolling up to review earlier output.(v2.1.121)
- Dialogs that overflow the terminal are now scrollable using arrow keys, PgUp/PgDn, home/end, and mouse wheel in all modes.(v2.1.121)
- Clicking any line of a long, wrapped URL in fullscreen mode now correctly opens the full URL.(v2.1.121)
- MCP servers experiencing a transient error during startup will now auto-retry up to 3 times instead of remaining disconnected.(v2.1.121)
- Duplicate Claude.ai connectors with the same upstream URL are now deduplicated.(v2.1.121)
- Startup time is faster after upgrading due to the removal of the Recent Activity panel from the release-notes splash screen.(v2.1.121)
- LSP diagnostic summaries now expand on click or Ctrl+O and display the expand hint.(v2.1.121)
- Reduced peak file descriptor usage during `find` operations within the Bash tool on large directory trees.(v2.1.121)
- On Windows, Claude Code now defaults to using PowerShell as the shell tool if Git for Windows (Git Bash) is not installed.(v2.1.120)
- Hidden spinner tips recommending desktop app installation or skill/agent creation if the user already possesses them.(v2.1.120)
- Added a hint indicating "use PgUp/PgDn to scroll" when the terminal receives arrow keys instead of scroll events.(v2.1.120)
- Improved startup speed when many unauthorized claude.ai connectors are configured.(v2.1.120)
- The auto mode denial message now includes a direct link to the configuration documentation.(v2.1.120)
- Updated `claude plugin validate` to accept `$schema`, `version`, and `description` at the top level of `marketplace.json` and `$schema` in `plugin.json`.(v2.1.120)
- Auto-compact in auto mode now displays `auto` (lowercase, no token count) instead of a misleading token value.(v2.1.120)
- Set the `AI_AGENT` environment variable for subprocesses to correctly attribute traffic to Claude Code.(v2.1.120)
- In VSCode, `/usage` now opens the native Account & Usage dialog instead of returning plain-text session cost.(v2.1.120)
- In VSCode, voice dictation now respects the `language` setting in `~/.claude/settings.json`.(v2.1.120)
- Configuration settings like theme, editor mode, and verbosity now persist to `~/.claude/settings.json` and respect project/local/policy override precedence.(v2.1.119)
- Subagent and SDK MCP server reconfiguration now connects servers in parallel instead of serially, improving startup time.(v2.1.119)
- Slash command suggestions now highlight the characters that matched your query.(v2.1.119)
- Slash command picker now wraps long descriptions onto a second line instead of truncating them.(v2.1.119)
- Security enforcement for `blockedMarketplaces` now correctly respects `hostPattern` and `pathPattern` entries.(v2.1.119)
- Tool search is now disabled by default on Vertex AI to prevent errors related to an unsupported beta header; users can opt-in via `ENABLE_TOOL_SEARCH`.(v2.1.119)
- Status line stdin JSON now includes `effort.level` and `thinking.enabled` for richer status reporting.(v2.1.119)
- When auto-update skips a plugin due to another plugin's version constraint, the skip now appears in `/doctor` and the `/plugin` Errors tab(v2.1.118)
- Model selections in /model now persist across restarts even if the project pins a different model, and the startup header indicates if the active model is from a project or managed-settings pin.(v2.1.117)
- Startup time is faster when both local and claude.ai MCP servers are configured due to concurrent connection defaulting.(v2.1.117)
- Managed-settings blockedMarketplaces and strictKnownMarketplaces are now enforced during plugin install, update, refresh, and autoupdate operations.(v2.1.117)
- The Advisor Tool (experimental) now displays an "experimental" label, a learn-more link, and a startup notification when enabled; sessions no longer get stuck with processing errors.(v2.1.117)
- The cleanupPeriodDays retention sweep now also covers ~/.claude/tasks/, ~/.claude/shell-snapshots/, and ~/.claude/backups/.(v2.1.117)
- Native builds on macOS and Linux now use embedded bfs and ugrep via the Bash tool for faster searches, replacing external Glob and Grep tools.(v2.1.117)
- Windows now caches where.exe executable lookups per process for faster subprocess launches.(v2.1.117)
- Default effort for Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 is increased to high (was medium).(v2.1.117)
- Fixed an idle re-render loop when background tasks are present, reducing memory growth on Linux.(v2.1.117)
- [VSCode] Fixed the "Manage Plugins" panel breaking when multiple large marketplaces are configured.(v2.1.117)
- Significantly improved `/resume` speed (up to 67% faster on 40MB+ sessions) and efficiency for sessions with many dead-fork entries(v2.1.116)
- Accelerated MCP startup when multiple stdio servers are configured by deferring `resources/templates/list` to the first @-mention(v2.1.116)
- Smoother fullscreen scrolling in VS Code, Cursor, and Windsurf terminals by updating scroll sensitivity configuration via `/terminal-setup`(v2.1.116)
- The Usage tab in Settings now shows 5-hour and weekly usage immediately and handles rate-limiting on the usage endpoint gracefully(v2.1.116)
- The slash command menu now displays "No commands match" when zero results are found, instead of disappearing(v2.1.116)
- Changed the CLI to spawn a native Claude Code binary (via a per-platform optional dependency) instead of bundled JavaScript(v2.1.113)
- Enabled viewport scrolling in Fullscreen mode using Shift+↑/↓ when extending a selection past the visible edge(v2.1.113)
- Enabled `Ctrl+A` and `Ctrl+E` to move to the start/end of the current logical line in multiline input, matching readline behavior(v2.1.113)
- Enabled `Ctrl+Backspace` to delete the previous word on Windows(v2.1.113)
- Ensured long URLs in responses and bash output remain clickable when they wrap across lines (in terminals with OSC 8 hyperlinks)(v2.1.113)
- Improved `/loop` functionality: pressing Esc now cancels pending wakeups, and wakeups display as "Claude resuming /loop wakeup" for clarity(v2.1.113)
- Enabled `/extra-usage` to work from Remote Control (mobile/web) clients(v2.1.113)
- Improved handling of subagents that stall mid-stream by failing them with a clear error after 10 minutes instead of hanging silently(v2.1.113)
- Improved Bash tool transcript display so multi-line commands whose first line is a comment now show the full command, closing a UI-spoofing vector(v2.1.113)
- Prevented permission prompts when running `cd <current-directory> && git …` where the `cd` command is a no-op(v2.1.113)
- Enhanced security by treating `/private/{etc,var,tmp,home}` paths as dangerous removal targets under `Bash(rm:*)` allow rules on macOS(v2.1.113)
- Enhanced security by making Bash deny rules match commands wrapped in `env`, `sudo`, `watch`, `ionice`, `setsid`, and similar exec wrappers(v2.1.113)
- Enhanced security by preventing `Bash(find:*)` allow rules from auto-approving `find -exec` or `-delete`(v2.1.113)
- Plan files are now named after the prompt (e.g., `fix-auth-race-snug-otter.md`) instead of using purely random words.(v2.1.111)
- The `OTEL_LOG_RAW_API_BODIES` environment variable was added to emit full API request and response bodies as OpenTelemetry log events for debugging.(v2.1.111)
- Suppressed spurious decompression, network, and transient error messages that could appear in the TUI during normal operation.(v2.1.111)
- Reverted the v2.1.110 cap on non-streaming fallback retries, which caused long waits or outright failures during API overload.(v2.1.111)
- `Ctrl+U` now clears the entire input buffer (previously deleted only to the start of the line).(v2.1.111)
- Changed `Ctrl+O` to toggle between normal and verbose transcript only; focus view is now toggled separately with the new /focus command.(v2.1.110)
- Improved /plugin Installed tab: items needing attention and favorites appear at the top, disabled items are hidden behind a fold, and `f` favorites the selected item.(v2.1.110)
- Improved /doctor to warn when an MCP server is defined in multiple config scopes with different endpoints.(v2.1.110)
- `--resume`/--continue` now resurrects unexpired scheduled tasks.(v2.1.110)
- /autocompact, /context, /exit, and /reload-plugins now work from Remote Control (mobile/web) clients.(v2.1.110)
- Bash tool now enforces the documented maximum timeout instead of accepting arbitrarily large values.(v2.1.110)
- SDK/headless sessions now read `TRACEPARENT`/`TRACESTATE` from the environment for distributed trace linking.(v2.1.110)
- Improved the extended-thinking indicator with a rotating progress hint(v2.1.109)
- Added ENABLE_PROMPT_CACHING_1H env var to opt into 1-hour prompt cache TTL on API key, Bedrock, Vertex, and Foundry (ENABLE_PROMPT_CACHING_1H_BEDROCK is deprecated but still honored), and FORCE_PROMPT_CACHING_5M to force 5-minute TTL(v2.1.108)
- /undo is now an alias for /rewind(v2.1.108)
- Improved /model to warn before switching models mid-conversation, since the next response re-reads the full history uncached(v2.1.108)
- Improved /resume picker to default to sessions from the current directory; press Ctrl+A to show all projects(v2.1.108)
- Improved error messages: server rate limits are now distinguished from plan usage limits; 5xx/529 errors show a link to status.claude.com; unknown slash commands suggest the closest match(v2.1.108)
- Reduced memory footprint for file reads, edits, and syntax highlighting by loading language grammars on demand(v2.1.108)
- Show thinking hints sooner during long operations(v2.1.107)
- Improved handling of stalled API streams: streams now abort after 5 minutes of no data and retry non-streaming instead of hanging indefinitely.(v2.1.105)
- Improved network error messages: connection errors now show a retry message immediately instead of a silent spinner.(v2.1.105)
- Improved file write display: long single-line writes (e.g. minified JSON) are now truncated in the UI instead of paginating across many screens.(v2.1.105)
- Improved `/doctor` layout with status icons; users can now press `f` to have Claude fix reported issues.(v2.1.105)
- Improved `/config` labels and descriptions for clarity.(v2.1.105)
- Improved skill description handling: raised the listing cap from 250 to 1,536 characters and added a startup warning when descriptions are truncated.(v2.1.105)
- Improved `WebFetch` to strip `<style>` and `<script>` contents from fetched pages so CSS-heavy pages no longer exhaust the content budget before reaching actual text.(v2.1.105)
- Improved stale agent worktree cleanup to remove worktrees whose PR was squash-merged instead of keeping them indefinitely.(v2.1.105)
- Improved MCP large-output truncation prompt to give format-specific recipes (e.g. `jq` for JSON, computed Read chunk sizes for text).(v2.1.105)
- Improved brief mode to retry once when Claude responds with plain text instead of a structured message.(v2.1.101)
- Improved focus mode: Claude now writes more self-contained summaries since it knows only the final message is seen.(v2.1.101)
- Improved tool-not-available errors to explain why and how to proceed when a model calls an unavailable tool.(v2.1.101)
- Improved rate-limit retry messages to show which limit was hit and when it resets instead of an opaque countdown.(v2.1.101)
- Improved refusal error messages to include the API-provided explanation when available.(v2.1.101)
- Improved claude -p --resume <name> to accept session titles set via \/rename or --name.(v2.1.101)
- Improved settings resilience: unrecognized hook event names in settings.json no longer cause the entire file to be ignored.(v2.1.101)
- Improved plugin hooks from plugins force-enabled by managed settings to run when allowManagedHooksOnly is set.(v2.1.101)
- Improved \/plugin and claude plugin update to show a warning when the marketplace could not be refreshed.(v2.1.101)
- Improved plan mode to hide the "Refine with Ultraplan" option when the user's org or auth setup cannot reach Claude Code on the web.(v2.1.101)
- Improved beta tracing to honor OTEL_LOG_USER_PROMPTS, OTEL_LOG_TOOL_DETAILS, and OTEL_LOG_TOOL_CONTENT; sensitive span attributes are no longer emitted unless opted in.(v2.1.101)
- Improved SDK query() to clean up subprocess and temp files when consumers break from for await or use await using.(v2.1.101)
- Improved `/resume` filter hint labels and added project/worktree/branch names to the filter indicator.(v2.1.98)
- Improved footer indicators (Focus, notifications) to remain on the mode-indicator row instead of wrapping at narrow terminal widths.(v2.1.98)
- Improved `/agents` view with a tabbed layout: Running tab for live subagents and Library tab with actions to run or view instances.(v2.1.98)
- Improved `/reload-plugins` to pick up plugin-provided skills without requiring a full restart.(v2.1.98)
- Improved Vim mode: `j`/`k` in NORMAL mode now navigate history and select the footer pill at the input boundary.(v2.1.98)
- Improved hook errors in the transcript to include the first line of stderr for self-diagnosis without needing `--debug`.(v2.1.98)
- Improved OTEL tracing: interaction spans now correctly wrap full turns under concurrent SDK calls.(v2.1.98)
- Improved transcript entries to carry final token usage instead of streaming placeholders.(v2.1.98)
- Updated the `/claude-api` skill to cover Managed Agents alongside Claude API.(v2.1.98)
- Dropped `/compact` hints when `DISABLE_COMPACT` is set.(v2.1.98)
- Improved Accept Edits mode to auto-approve filesystem commands prefixed with safe env vars or process wrappers (e.g. LANG=C rm foo, timeout 5 mkdir out)(v2.1.97)
- Improved auto mode and bypass-permissions mode to auto-approve sandbox network access prompts(v2.1.97)
- Improved sandbox: sandbox.network.allowMachLookup now takes effect on macOS(v2.1.97)
- Improved image handling: pasted and attached images are now compressed to the same token budget as images read via the Read tool(v2.1.97)
- Improved slash command and @-mention completion to trigger after CJK sentence punctuation, so Japanese/Chinese input no longer requires a space before / or @(v2.1.97)
- Improved Bridge sessions to show the local git repo, branch, and working directory on the claude.ai session card(v2.1.97)
- Improved footer layout: indicators (Focus, notifications) now stay on the mode-indicator row instead of wrapping below(v2.1.97)
- Improved context-low warning to show as a transient footer notification instead of a persistent row(v2.1.97)
- Improved markdown blockquotes to show a continuous left bar across wrapped lines(v2.1.97)
- Improved session transcript size by skipping empty hook entries and capping stored pre-edit file copies(v2.1.97)
- Improved transcript accuracy: per-block entries now carry the final token usage instead of the streaming placeholder(v2.1.97)
- Improved Bash tool OTEL tracing: subprocesses now inherit a W3C TRACEPARENT env var when tracing is enabled(v2.1.97)
- Changed default effort level from medium to high for API-key, Bedrock/Vertex/Foundry, Team, and Enterprise users (controllable via `/effort`).(v2.1.94)
- Improved `--resume` to resume sessions directly from other worktrees of the same repo instead of printing a `cd` command.(v2.1.94)
- Reduced cold-open subprocess work when starting a session in VSCode.(v2.1.94)
- Added a warning banner in VSCode when `settings.json` files fail to parse, indicating permission rules may not be applied.(v2.1.94)
- Added forceRemoteSettingsRefresh policy setting to block startup until remote managed settings are freshly fetched, failing closed if fetch fails.(v2.1.92)
- /release-notes command is now an interactive version picker.(v2.1.92)
- Removed the /tag command.(v2.1.92)
- Removed the /vim command (Vim mode is now toggled via /config → Editor mode).(v2.1.92)
- Improved guidance for the `/claude-api` skill, specifically detailing agent design patterns like tool surface decisions, context management, and caching strategy.(v2.1.91)
- Improved performance by routing `stripAnsi` through `Bun.stripANSI` for faster processing when using Bun.(v2.1.91)
- The Edit tool now uses shorter `old_string` anchors, resulting in reduced output tokens.(v2.1.91)
- The `/feedback` command now displays an explanation when it is unavailable instead of simply disappearing from the slash menu.(v2.1.91)
- Added .husky to protected directories when in acceptEdits mode(v2.1.90)
- Fixed click-to-expand hover text being nearly invisible on light terminal themes(v2.1.90)
- Fixed headers disappearing when scrolling /model, /config, and other selection screens(v2.1.90)
- Improved performance by eliminating per-turn stringify of MCP tool schemas on cache-key lookup(v2.1.90)
- Improved performance: SSE transport now handles large streamed frames in linear time (was quadratic)(v2.1.90)
- Improved performance: SDK sessions with long conversations no longer slow down quadratically on transcript writes(v2.1.90)
- Improved /resume all-projects view to load project sessions in parallel, speeding up load times for users with many projects(v2.1.90)
- Changed the --resume picker to no longer show sessions created by claude -p or SDK invocations(v2.1.90)
- Removed Get-DnsClientCache and ipconfig /displaydns from auto-allow list due to DNS cache privacy concerns(v2.1.90)
- Improved collapsed tool summary to show "Listed N directories" for ls/tree/du instead of "Read N files".(v2.1.89)
- Improved Bash tool to warn when a formatter/linter command modifies files previously read, preventing stale-edit errors.(v2.1.89)
- Improved @-mention typeahead to rank source files above MCP resources with similar names.(v2.1.89)
- Improved PowerShell tool prompt with version-appropriate syntax guidance (5.1 vs 7+).(v2.1.89)
- Changed Edit tool to work on files viewed via Bash with sed -n or cat, without requiring a separate Read call first.(v2.1.89)
- Changed hook output over 50K characters to be saved to disk with a file path + preview instead of being injected directly into context.(v2.1.89)
- Changed cleanupPeriodDays: 0 in settings.json to be rejected with a validation error instead of silently disabling transcript persistence.(v2.1.89)
- Changed thinking summaries to no longer be generated by default in interactive sessions; set showThinkingSummaries: true to restore.(v2.1.89)
- Preserved task notifications when backgrounding a running command with Ctrl+B.(v2.1.89)
- PowerShell tool on Windows now prompts for external-command arguments containing both a double-quote and whitespace due to PS 5.1 hardening.(v2.1.89)
- /env now applies to PowerShell tool commands (previously only affected Bash).(v2.1.89)
- /usage now hides redundant "Current week (Sonnet only)" bar for Pro and Enterprise plans.(v2.1.89)
- Image paste no longer inserts a trailing space.(v2.1.89)
- Pasting !command into an empty prompt now enters bash mode, matching typed ! behavior.(v2.1.89)
All Releases in April 2026
v2.1.1231 fixThis release addresses a specific authentication issue related to the CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS environment variable. Users will no longer encounter a 401 retry loop during OAuth authentication when this setting is active.
v2.1.1224 features12 fixesThis release introduces granular control over Bedrock service tiers via an environment variable and improves session recovery by allowing PR URLs to be used in the `/resume` command. Several critical bugs were fixed, including issues with tool usage in branched sessions, structured output errors with Vertex AI/Bedrock, and incorrect image resizing for newer models.
v2.1.12113 features18 fixesThis release introduces significant quality-of-life improvements, including a type-to-filter search for skills and better handling of tool output replacement via PostToolUse hooks. Numerous critical bugs have been resolved, focusing heavily on fixing memory leaks related to image processing and transcript history, and improving session stability during restarts and network transients. Users will also notice better scrolling behavior in fullscreen mode and enhanced clipboard support in certain terminal setups.
v2.1.1202 features10 fixesThis release introduces non-interactive UltraReview via a new subcommand and allows skills to dynamically reference the current effort level. Numerous bugs have been resolved, including fixes for server connection stability during Esc presses and improved handling of telemetry suppression and terminal scrolling issues. Additionally, Windows users no longer strictly require Git Bash, as PowerShell is now used as a fallback shell.
v2.1.11912 features33 fixesThis release introduces significant configuration persistence, allowing settings like theme and editor mode to be saved locally and respected across overrides. Key updates include expanded support for various Git provider URLs in the `--from-pr` flag and improved handling of tool execution context in print mode and hooks. Numerous bugs were fixed, addressing issues with multi-line pasting, configuration overrides, and various command behaviors across different operating systems.
v2.1.11812 features21 fixesThis release introduces significant enhancements to Vim editing support, including visual and visual-line modes, and overhauls theme management by allowing named custom themes and plugin-shipped themes. Additionally, it brings numerous stability fixes, particularly around MCP server authentication, OAuth token handling, and session management across different operating systems.
v2.1.1175 features12 fixesThis release introduces significant improvements to plugin management, including better dependency handling and enforcement of marketplace restrictions. Users will also notice faster startup times when multiple MCP servers are configured and improved session resumption behavior. Several critical bugs related to token expiration, proxy handling, and context window calculation for Opus 4.7 have also been resolved.
v2.1.1166 features13 fixesThis release focuses heavily on performance and stability, bringing significant speedups to resuming large sessions and improving terminal interaction, especially for Indic scripts and Kitty keyboard protocol users. New features include inline thinking progress indicators and enhanced search capabilities within the `/config` command. Several bugs related to terminal scrolling, session loading, and plugin management have also been resolved.
v2.1.1141 fixThis release focuses on stability, specifically addressing a crash that occurred within the permission dialog when a teammate requested tool access. No new features were introduced in this update.
v2.1.1133 features22 fixesThis release introduces new network sandboxing controls via `sandbox.network.deniedDomains` and enables Remote Control clients to query file autocomplete suggestions. Numerous stability fixes address issues with session archiving, transcript display, markdown pasting, and various error conditions across CLI and UI interactions.
v2.1.1121 fixThis release focuses on stability by resolving an availability error encountered when using auto mode with the claude-opus-4-7 model. No new features or general improvements were introduced in this update.
v2.1.11113 features15 fixesThis release introduces the powerful Claude Opus 4.7 xhigh model, allowing fine-grained control over speed and intelligence via the updated `/effort` command. Key new capabilities include the comprehensive `/ultrareview` skill for multi-agent code analysis and an automatic theme matching your terminal settings. Several critical bugs related to file suggestions, permission prompts, and terminal rendering have also been resolved.
v2.1.1106 features20 fixesThis release introduces significant new interaction methods, including a flicker-free TUI mode and a mobile push notification tool for remote control. Several critical bugs related to hanging API calls, session state management, and UI rendering have been resolved. Additionally, usability is improved with better plugin management and enhanced external editor integration.
v2.1.109This release focuses on enhancing the user experience by refining the extended-thinking indicator. Users will now see a rotating progress hint during extended thinking processes.
v2.1.1084 features14 fixesThis release introduces a new session recap feature to restore context when returning to ongoing work, and enables the model to invoke built-in slash commands directly via the Skill tool. Several quality-of-life improvements include better error messaging, reduced memory usage for file operations, and configuration options for prompt caching TTLs.
v2.1.107This release focuses on improving the user experience during long operations by providing thinking hints sooner. There are no major new features or bug fixes noted in this update.
v2.1.1054 features24 fixesThis release introduces significant enhancements to tool capabilities, including adding a `path` parameter to `EnterWorktree` and enabling background monitor support for plugins. Numerous fixes address issues related to UI rendering, queued message handling, and network error reporting, alongside several quality-of-life improvements for configuration and debugging.
v2.1.104v2.1.1013 features31 fixesThis release introduces a new \/team-onboarding command for generating ramp-up guides and enables enterprise TLS proxy support by trusting the OS CA certificate store by default. Numerous stability and usability improvements were made, particularly around session resumption, error messaging for tool usage, and handling of settings and plugins.
v2.1.100v2.1.989 features38 fixesThis release introduces significant security and configuration enhancements, including an interactive Vertex AI setup wizard and new subprocess sandboxing features for Linux. Numerous bug fixes address critical issues related to Bash command permissions, streaming response fallbacks, and configuration overrides across various modes like fullscreen and resume.
v2.1.976 features28 fixesThis release introduces significant enhancements to the NO_FLICKER mode, including a new focus view toggle and numerous stability fixes for terminal rendering and input handling. Several critical bugs related to permissions, configuration persistence, and API retries have been resolved, alongside improvements to image token usage and CJK input handling for slash commands.
v2.1.961 fixThis release addresses a critical regression affecting Bedrock requests. Specifically, it resolves an issue where authorization headers were missing, causing 403 errors when using specific Bedrock authentication methods.
v2.1.945 features16 fixesThis release introduces support for Amazon Bedrock powered by Mantle and enhances plugin functionality by stabilizing invocation names and improving hook reliability. Several critical bugs were fixed, including silent login failures on macOS and issues with rate-limiting handling, ensuring a more robust user experience.
v2.1.924 features12 fixesThis release introduces an interactive Bedrock setup wizard for easier AWS integration and adds detailed cost breakdowns for subscription users. Several critical bugs have been resolved, including issues with subagent spawning in tmux sessions and incorrect hook behavior. Additionally, the /release-notes command is now interactive, and the /vim command has been replaced by a configuration toggle.
v2.1.914 features5 fixesThis release introduces significant new capabilities, including support for larger MCP tool results (up to 500K chars) and the ability for plugins to ship and execute their own binaries. Several critical bugs were resolved, such as fixing conversation history loss during resume operations and correcting `cmd+delete` behavior across major terminals.
v2.1.902 features8 fixesThis release introduces interactive lessons via /powerup to help users learn Claude Code features, alongside a new environment variable for better offline marketplace caching. Several critical bugs were fixed, including an infinite loop crash related to usage limits and regressions affecting `--resume` functionality. Performance has also been significantly improved for SDK sessions and large streamed data handling.
v2.1.897 features30 fixesThis release introduces significant new capabilities for headless sessions with deferred permission handling and enhanced tool usage hooks. Numerous bugs have been resolved across file operations, history persistence, and tool execution, particularly for Windows and large file handling. Quality of life improvements include better tool summaries, updated PowerShell guidance, and disabling default thinking summaries.