Claude Code

No entries match these filters. Clear filters

1. v2.1.283 Sep 25, 2026 · release notes

   Claude Code v2.1.283 adds prompt-audit, model management settings, gateway hints, and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface./otel telemetry hooks.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.283 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.283

   └──▷ TRY IT
   Enable prompt grouping in your gateway to identify requests belonging to a single user prompt.
   copy

   $ export CLAUDE_CODE_GATEWAY_HINT_HEADERS=1

   Audit your local configuration files for prompting patterns written for older models that may no longer be effective.
   copy

   $ /doctor prompt-audit

   • Adds x-claude-code-prompt-id header grouping for LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. gateways, controlled by the CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 environment variable
   • Adds /doctor prompt-audit and /checkup prompt-audit subcommands to audit CLAUDE.md files, skills, agents, and commands for outdated prompting patterns
   • Adds availableModelsMatch managed setting (e.g., exact) and deniedModels managed setting to block specific models in managed environments
   • Adds MCP tool, WebFetch, and WebSearch output capture to OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. spans when OTEL_LOG_TOOL_CONTENT=1 is set
   • Adds opt-in load_test_mode block to the Claude apps gateway config for load testing without sending upstream requests

   +1 more show less

   • Adds mantle upstream provider to the Claude apps gateway for Amazon Bedrock's Mantle endpoint

   6 details in Everything

2. v2.1.282 Sep 24, 2026 · release notes

   New telemetry filtering and Chrome/MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. coexistence controls in Claude Code v2.1.282

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.282 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.282

   • Adds maxProseWidth config key to cap AI prose width while preserving table/code block formatting.
   • Adds /status and claude doctor commands to list ignored or disabled telemetry variables from settings files.
   • Adds allowClaudeInChromeWithManagedMcp managed setting to permit claude --chrome operation alongside exclusive MCP configurations.
   • Adds store.readiness_grace_seconds gateway configuration to maintain /readyz health during brief Postgres outages.

   4 details in Everything

3. v2.1.281 Sep 23, 2026 · release notes

   Claude Code v2.1.281 adds Claude apps gateway configs (assume_role, guardrail, telemetry.resource_attributes), MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. URL-mode elicitation, and plugin validation checks.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.281 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.281

   • Adds blockReadsOutsideWorkingDirectories and disableBypassPermissionsMode to desktop policy blocks for newer Claude Desktop keys in the Claude apps gateway.
   • Adds assume_role support on Claude apps gateway Bedrock upstreams to call Bedrock as an assumed IAMIAMIdentity and Access Management — a framework of policies and services for controlling who can authenticate to systems and what resources they are authorized to use, letting cyber tools enforce least-privilege and audit access. role via STSSTSAWS Security Token Service, an Amazon Web Services API that issues short-lived credentials for IAM roles, letting cyber tools authenticate to AWS without storing long-term secrets. across AWS accounts.
   • Adds guardrail: {id, version} configuration on Claude apps gateway Bedrock upstreams to apply Amazon Bedrock guardrails to every request.
   • Adds telemetry.resource_attributes to the Claude apps gateway config to attach fixed labels to telemetry for Claude Desktop and /login sessions.
   • Adds "attribution": false support in settings.json to hide all commit and PR attribution across CLI versions.

   +2 more show less

   • Enables MCP URL-mode elicitation on 2026-07-28 protocol connections, allowing servers to trigger browser-based flows without leaving a waiting dialog.
   • Adds validation checks for MCP servers in claude plugin validate, reporting silently dropped .mcp.json entries, undeclared ${user_config.*} references, and insecure URLs.

   7 details in Everything

4. v2.1.280 Sep 22, 2026 · release notes

   v2.1.280 adds Claude Opus 5.5 support, mouse controls in fullscreen lists, and a configurable MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. description length limit

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.280 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.280

   └──▷ USE IT
   Configure MCP servers with descriptions longer than the default limit.
   shell copy

   export CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH=4096

   Interact with skills using a mouse in fullscreen mode.
   📍In fullscreen mode, use the mouse wheel to scroll the /skills list or click skill state options in /plugin.
   Use the new Opus 5.5 model for a session.
   shell copy

   claude -m claude-opus-5-5

   • Adds claude-opus-5-5 as the new default Opus model with 1M context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it.
   • Enables mouse wheel scrolling for /skills list and click interaction for skill state options in /plugin within fullscreen mode
   • Introduces CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH environment variable to override the 2,048-character cap on MCP tool descriptions and server instructions

   3 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !Reverts ctrl+l / cmd+k in fullscreen mode clearing the transcript view (behavior added in 2.1.260)
   • !Changes default model on Pro and Team Standard plans from Sonnet to Opus
   • !Restricts effort levels saved before /effort became per-model so they no longer apply to newly released models such as Opus 5.5
5. v2.1.277 Sep 18, 2026 · release notes

   Adds AGENTS.md project instruction fallback, CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY proxy env var, and optional headers: map for gateway upstreams.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.277 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.277

   └──▷ USE IT
   Use AGENTS.md as the fallback project instruction file when no CLAUDE.md exists.
   shell copy

   echo '# Agent Instructions' > AGENTS.md
   claude

   • AGENTS.md support: in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead; change it under Project instructions in /config (not yet on Bedrock, Vertex or Foundry)
   • CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1 for Claude apps gateways whose only egress is a forward proxy: every outbound request hands the proxy the hostname instead of resolving it locally
   • Optional headers: map on Claude apps gateway upstreams, to send static headers to a proxy you run in front of a provider

   3 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !The deprecated TaskOutput tool is removed; Claude reads a background task's output file with Read instead, and the taskOutputMaxChars setting and TASK_MAX_OUTPUT_LENGTH no longer have any effect
6. v2.1.275 Sep 17, 2026 · release notes

   Adds syncClaudeAiSkills/plugins config, --forward-subagent-text flag, /plugin install marketplace, and VSCode accept/reject diff buttons

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.275 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.275

   • Adds syncClaudeAiSkills: false and syncClaudeAiPlugins: false config keys to disable syncing of claude.ai account skills and plugins to terminal sessions
   • Adds /plugin install <plugin> --marketplace <source> command to install plugins directly from a named marketplace source
   • Adds claudeCode.scrollToBottomOnSend setting in VS Code to disable automatic scroll-to-bottom after sending a message
   • Adds accept and reject buttons in the VS Code proposed-change diff tab for reviewing edits change-by-change

   4 details in Everything

7. v2.1.274 Sep 17, 2026 · release notes

   New env var CLAUDE_CODE_MCP_STARTUP_WAIT_MS, store.connect_timeout_seconds config, and claude_code.managed_settings_resolved OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. event in v2.1.274

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.274 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.274

   • Adds CLAUDE_CODE_MCP_STARTUP_WAIT_MS environment variable to control how long the first non-interactive turn waits for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server connections.
   • Adds store.connect_timeout_seconds setting in the Claude apps gateway config to adjust the Postgres connection timeout.
   • Adds claude_code.managed_settings_resolved OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. event exposing managed-settings sources and policy helper state.

   3 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !Changes /code-review to use leaner inline review prompts instead of spawning many review subagents, altering the output format and execution path.
8. v2.1.273 Sep 15, 2026 · release notes

   Claude Code v2.1.273 adds LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. gateway hint headers, session forking via remote control, and several improved MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface./AWS integration capabilities.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.273 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.273

   • Adds x-claude-code-request-class, x-claude-code-agent-type, x-claude-code-prev-tool-durations, x-claude-code-compaction and x-claude-code-context-compacted request headers for LLM gateways; opt in with CLAUDE_CODE_GATEWAY_HINT_HEADERS=1
   • Adds forking a session started with claude --remote-control or /remote-control from the Claude app, running it as a background session on your computer
   • Changes auto mode on Bedrock, Vertex and Foundry to use the local classifier by default; set CLAUDE_CODE_AUTO_MODE_SERVER=1 to use the platform's server-side classifier
   • Improves OAuth client-credentials and JWTJWTAn open standard (RFC 7519) for encoding claims as a signed, optionally encrypted JSON object passed between parties. Cyber tools use it to verify identity and permissions without querying a central session store.-bearer connections with providers that return a lowercase token type: requests now send the standard Bearer scheme

   4 details in Everything

9. v2.1.271 Sep 14, 2026 · release notes

   Claude Code v2.1.271 adds --drain-marker-file, per-command allowed_domains, --accept-command for plugins, and modelPricing multiplier.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v2.1.271 https://github.com/anthropics/claude-code.git
   # already have the repo? check out this version:
   $ git checkout v2.1.271

   └──▷ TRY IT
   Configure a self-hosted runner to gracefully drain and report its exit via a marker file when receiving SIGTERM.
   copy

   $ claude self-hosted-runner --drain-marker-file /tmp/drain.marker

   Install a plugin by securely accepting only the specific command hash verified in a previous JSON run.
   copy

   $ claude plugin install --accept-command <sha256>

   Set up agent subagents to bypass user, project, and local CLAUDE.md files using the new omitClaudeMd flag in JSON input.
   copy

   $ claude --agents '{"omitClaudeMd": true}' "Run analysis"

   • Adds claude self-hosted-runner --drain-marker-file <path> to signal host drain on SIGTERM
   • Adds per-command allowed_domains configuration for Bash, PowerShell and Monitor in auto mode with sandboxingsandboxingRunning code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls.
   • Adds omitClaudeMd field to agent frontmatter and --agents JSON for subagent runs without standard CLAUDE.md files
   • Adds --accept-command <sha256> flag to claude plugin install and claude plugin update for safe plugin installation from JSON output
   • Adds support for a multiplier above 1, up to 10, in the modelPricing managed setting and gateway pricing block for chargeback rates

   5 details in Everything

10. v2.1.269 Sep 11, 2026 · release notes

    Adds claude plugin eval subcommand, /output-style slash command, new env vars for telemetry and concurrency, and VSCode permission rules dialog.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.269 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.269

    └──▷ USE IT
    Evaluate a plugin's performance against Claude Code using its built-in suite.
    shell copy

    claude plugin eval --help

    Extend the LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. gateway model discovery timeout to 10 seconds.
    shell copy

    export CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS=10000

    Set the maximum concurrent agents for a workflow run to 64.
    shell copy

    export CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS=64

    • Adds the claude plugin eval subcommand to run a plugin's eval suite with JSON + HTML reports
    • Adds the /output-style [name] slash command to list and switch output styles in headless sessions
    • Adds the CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS environment variable to configure the LLM gateway discovery timeout (default 3s)
    • Adds the OTEL_METRICS_INCLUDE_REPOSITORY environment variable to tag OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. metrics with vcs.* repository attributes
    • Adds the CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS environment variable (1–256) to raise the Workflow tool's per-run concurrent agent limit

    +5 more show less

    • Adds a diff of files changed by Bash commands to the Bash tool result when bashEditDiffEnabled is set
    • Adds an agent map in VSCode that opens from an N agents footer pill to show sub-agents with Stop and transcript controls
    • Adds a Permission rules dialog in VSCode for managing permission rules in user, project, and local settings files
    • Adds a Hooks dialog in VSCode for viewing and editing hooks in user, project, and local settings files
    • Adds the ability to take back a queued message in a cloud session via Esc o before Claude reads it

    10 details in Everything

11. v2.1.268 Sep 10, 2026 · release notes

    v2.1.268 adds gateway network allowlisting, session state cleanup CLI, JSON plugin commands, and configurable WebFetch timeout

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.268 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.268

    └──▷ TRY IT
    Run the self-hosted runner with automatic session state cleanup enabled.
    copy

    $ claude self-hosted-runner --remove-session-state

    Set a custom 60-second timeout for WebFetch requests via environment variable.
    copy

    $ export CLAUDE_CODE_WEBFETCH_DEADLINE_MS=60000 && claude 'fetch https://example.com'

    • Adds gatewayInternalNetworks managed setting to allow /login from a specific public IPv4 block on a Claude apps gateway
    • Adds claude self-hosted-runner --remove-session-state subcommand to delete per-session directories under <base-dir>/_sessions/ when the session ends
    • Adds CLAUDE_CODE_WEBFETCH_DEADLINE_MS environment variable to configure the WebFetch timeout (default 300 seconds)
    • Adds --json output to claude plugin install, uninstall, update, enable, and disable commands
    • Adds errorDetails and noteDetails fields to each row when running claude plugin list --json

    5 details in Everything

12. v2.1.267 Sep 9, 2026 · release notes

    Claude Code v2.1.267 adds maxEffortLevel config and --system-prompt-snapshot CLI flag for prompt iteration.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.267 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.267

    └──▷ USE IT
    Iterate on prompt text without the model caching the initial system prompt.
    shell copy

    claude --system-prompt-snapshot off

    • Adds maxEffortLevel setting (top-level or per model under modelSettings) to cap the effort level on every provider, including Bedrock, Vertex and Foundry
    • Adds --system-prompt-snapshot off CLI argument to render the system prompt fresh on every request instead of reusing the conversation's recorded prompt

    2 details in Everything

13. v2.1.265 Sep 8, 2026 · release notes

    v2.1.265 adds dynamic plugin-dir scanning, OTLPOTLPOpenTelemetry Protocol, a standard way to send traces, metrics and logs between applications, collectors and monitoring systems. It lets telemetry move across compatible tools. telemetry export, and VSCode session archiving

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.265 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.265

    • --plugin-dir now loads a folder of plugins (each child folder with a manifest) and hot-adds/hot-removes them while running
    • OTEL_EXPORTER_OTLP_ENDPOINT is used by Claude apps gateway sessions to export OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. directly to a collector, bypassing the relay
    • [VSCode] New Archive inactive sessions setting (default 14 days) adds automatic session archiving

    3 details in Everything

14. v2.1.261 Sep 4, 2026 · release notes

    Claude Code v2.1.261 adds --append-subagent-system-prompt-file, new /skill-doctor command, and output size limits.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.261 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.261

    • Adds bashOutputMaxChars and taskOutputMaxChars config keys to control inline output limits up to 128K characters
    • Adds --append-subagent-system-prompt-file CLI flag to read subagent system prompts from a file instead of the command line
    • Adds /skill-doctor slash command to identify unused loaded skills and their context cost for pruning
    • Adds VS Code walkthrough to build custom output styles and an Add/Remove form for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers in the IDE

    4 details in Everything

15. v2.1.260 Sep 3, 2026 · release notes

    Claude Code v2.1.260 adds /diff panel, /reload-plugins for headless, OIDCOIDCOpenID Connect, a sign-in protocol built on OAuth 2.0. An application uses identity information from a trusted provider instead of maintaining a separate password for each user. scope_on_refresh, and agent team fixes.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.260 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.260

    └──▷ TRY IT
    Monitor uncommitted code changes in real-time during an agent session by toggling the new diff view.
    copy

    $ /diff

    • Adds a /diff slash command to toggle a fullscreen diff panel showing uncommitted changes as Claude edits files
    • Adds /reload-plugins subcommand to headless sessions (SDK and Remote Control), appearing in standard command lists
    • Adds a text form of the /advisor command for desktop, remote control, and -p agent SDK sessions (/advisor, /advisor <model>, /advisor off)
    • Adds oidc.scope_on_refresh support in the Claude apps gateway for IdPs requiring openid scope on refresh to return an id_token
    • Adds support for newer Claude Desktop policy fields userPluginMarketplacesEnabled and userPluginUploadsEnabled in desktop policy blocks via the Claude apps gateway

    5 details in Everything

16. v2.1.259 Sep 2, 2026 · release notes

    Adds managedMcpServers org setting, --permission-prompts none, and --json output to claude plugin validate

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.259 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.259

    └──▷ USE IT
    Run Claude Code in headless mode where any permission prompt is automatically denied.
    shell copy

    claude --permission-prompts none 'Summarize this file'

    Get a JSON validation report from the plugin system for CI integration.
    shell copy

    claude plugin validate --json

    • Adds managedMcpServers managed setting allowing organizations to provide HTTP/SSESSEServer-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers to every user.
    • Adds --permission-prompts none flag for unattended headless hosts to automatically deny prompts.
    • Adds --json output format to claude plugin validate for machine-readable validation reports.

    3 details in Everything

17. v2.1.257 Sep 1, 2026 · release notes

    Adds Claude Fable 5.1 as default, CLAUDE_CODE_SUBAGENT_MODEL_FORCE, time settings, auto-mode containment rules, and /effort session scope.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.257 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.257

    • claude-fable-5-1 is now the default Fable model with 1M context and new pricing tiers ($10/$50 per Mtok, $0.25/Mtok cache reads).
    • Added timeFormat and timeZone settings to control the turn-end clock and transcript-view timestamps.
    • Added a Containment Escape rule in auto mode that requires explicit approval for cloud metadata-credential fetches, egress evasion, and cross-tenant reach.
    • CLAUDE_CODE_SUBAGENT_MODEL_FORCE environment variable now applies the subagent model to every subagent, ignoring per-spawn and agent-definition overrides.
    • Added s modifier to /effort command to change effort for the current session only.

    +3 more show less

    • Added /doctor warning for stale sandbox mask files left by killed sessions.
    • Added a one-time prompt in auto mode before the first file read outside working directories, with permissions.blockReadsOutsideWorkingDirectories config key to block such reads.
    • Support for gateway-supplied description on discovered /model picker entries via CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY.

    8 details in Everything

18. v2.1.251 Aug 28, 2026 · release notes

    Claude Code v2.1.251 adds model switch hooks, subagent streaming, and session context to status lines.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.251 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.251

    └──▷ USE IT
    Attach to a specific background session ID to view its output or interact with it
    shell copy

    claude attach <id>

    • Adds PreModelSwitch and PostModelSwitch hook events for blocking, confirming, or annotating model changes
    • Enables live streaming of foreground subagent tool calls and results to Remote Control clients via new event hooks
    • Adds rate_limits.spend_limit status line field for tracking spend caps behind Claude apps gateways
    • Adds prompt_cache object to status line scripts with hit ratio, misses, re-cache count, and warm/cold metrics

    4 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !CLAUDE_CODE_SUBAGENT_MODEL no longer overrides all models; it now sets the default for unconfigured spawns, yielding to agent definition or per-spawn model settings
19. v2.1.248 Aug 27, 2026 · release notes

    Claude Code v2.1.248 adds --restricted mode, agent frontmatter cache TTLTTLTime to live, a limit on how long data remains valid or how far a packet can travel. Caches use a time limit; IP packets use a hop limit., cross-session messaging, and self-hosted-runner client labeling.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.248 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.248

    └──▷ TRY IT
    Run Claude Code in a locked-down environment where code execution and external API calls are forbidden.
    copy

    $ claude --restricted

    Configure a specific agent to cache prompts for 1 hour instead of the default duration.
    text copy

    ---
    name: my-agent
    model: sonnet
    experimental:
      cacheTtl: "1h"
    ---

    Identify a specific self-hosted runner with a custom label for routing or logging purposes.
    copy

    $ claude self-hosted-runner --client-label prod-worker-01

    • Adds --restricted flag (or CLAUDE_CODE_RESTRICTED=1 env var) to disable command/code execution tools, limit file access to the working directory, refuse bypassPermissions, and ignore user/project/local settings files
    • Adds experimental.cacheTtl key to agent frontmatter (accepting values like "5m" or "1h") to control per-agent prompt cache TTL when no subagent setting is configured
    • Adds claude self-hosted-runner --client-label <label> argument (or SELF_HOSTED_RUNNER_CLIENT_LABEL env var) to override the label the runner registers with (default: hostname)
    • Adds /doctor and /status slash commands in server-managed settings diagnostics to explain load failures for Bedrock/Vertex/third-party providers or custom ANTHROPIC_BASE_URL
    • Adds /usage-credits slash command for Enterprise organizations billed through AWS Marketplace, self-serve Enterprise, and Enterprise trials to request higher usage limits from admin

    +1 more show less

    • Adds cross-session messaging capabilities via SendMessage and ListAgents between sessions on the same machine on Bedrock, Vertex, Foundry, and when telemetry is disabled

    6 details in Everything

20. v2.1.247 Aug 26, 2026 · release notes

    New /claude-api cost-optimize subcommand, SendFeedback tool with feedbackDrafts config, and spinnerTipsOverride structure expansion

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.247 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.247

    └──▷ TRY IT
    Profile an existing project's Claude API spend to identify cost optimization levers like caching and token hygiene.
    copy

    $ claude /claude-api cost-optimize

    • Added /claude-api cost-optimize to profile an existing project's Claude API spend and work through cost levers (caching, token hygiene, batch, effort, model choice) one measured change at a time
    • Added the SendFeedback tool: when something goes wrong in a session, Claude can draft a feedback report for you to review and send from /feedback (turn off with the feedbackDrafts setting)
    • Added {id, text, cooldownSessions, priority} entries, tipsFile, and label to spinnerTipsOverride, so organizations can rotate their own tips alongside the built-in ones
    • Updated the /claude-api skill with Admin APIAdmin APIA dedicated HTTP interface that exposes management and configuration operations separately from the data plane, letting cyber tools automate provisioning, policy changes, or monitoring without touching the main user-facing service. coverage (organization members, invites, workspaces, API keys, rate limit reports, workload identityworkload identityAn identity assigned to software, such as a service or container, rather than a person. It lets the workload authenticate and receive permissions without sharing a user's credentials. federation, CMEK)

    4 details in Everything

21. v2.1.246 Aug 25, 2026 · release notes

    Claude Code v2.1.246 adds Auto mode permissions tab, subagent output tracking, and improved latency.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.246 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.246

    • Adds /permissions auto mode tab for viewing and editing classifier rules
    • Returns partial output marker from subagents that hit maxTurns limit
    • Improves Bash tool latency by replaying snapshot functions instead of base64 subshells
    • Automatically continues non-interactive responses cut off mid-stream
    • Enables /code-review on Bedrock, Vertex AI, Foundry via Claude apps gateway

    5 details in Everything

22. v2.1.239 Aug 21, 2026 · release notes

    New /claude-api upgrade command for Anthropic SDK migration, synced cloud plugins, and new CLAUDE_CODE_RETRY_WATCHDOG behavior

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.239 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.239

    └──▷ TRY IT
    Migrate an existing Python project to use the new Anthropic SDK 1.x API structure.
    copy

    $ /claude-api upgrade

    Configure persistent retry behavior that fails fast on budget limits rather than waiting indefinitely.
    copy

    $ export CLAUDE_CODE_RETRY_WATCHDOG=1

    Enable a plugin synced from the cloud in your local session using the synced naming convention.
    copy

    $ claude plugin enable my-synced-plugin@synced

    • Adds the /claude-api upgrade subcommand to migrate Python projects from anthropic 0.x to 1.x.
    • New cloud session plugins sync from claude.ai as name@synced, supporting claude plugin enable/disable <name>@synced.
    • New persistent retry mode via CLAUDE_CODE_RETRY_WATCHDOG environment variable for immediate failure on organization spend-limit errors.
    • Added /cost, status line, and --max-budget-usd support for the 1.1× US-only-inference premium in data-residency workspaces.

    4 details in Everything

23. v2.1.238 Aug 20, 2026 · release notes

    Claude Code v2.1.238 adds readline keybinding flavor, self-hosted runner shutdown delays, proxy auth commands, and plugin headersHelper

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.238 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.238

    └──▷ TRY IT
    Keep the runner running for 5 minutes after SIGTERM to finish active sessions before shutting down
    copy

    $ claude self-hosted-runner --defer-shutdown-max-min 5

    Pass a script that outputs a fresh Proxy-Authorization header on every connection via the runner's CLI flag
    copy

    $ claude self-hosted-runner --proxy-authorization-command "./fetch-proxy-token.sh"

    • Adds keybindingFlavor setting to .claude/config.yaml to select readline style editing
    • Adds --defer-shutdown-max-min <minutes> flag to claude self-hosted-runner for controlled session teardown
    • Adds --proxy-authorization-command and --proxy-authorization-file flags to claude self-hosted-runner for dynamic proxy auth
    • Introduces headersHelper capability in plugin marketplaces and catalog entries for minting HTTP headers during installation/update
    • Enables crossSessionInbound: "refuse" setting to explicitly reject inbound cross-session messages with a status report

    5 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Removed the double-press /clear shortcut; Ctrl+L and Cmd+K in fullscreen now only repaint without clearing history
    • !MCP headersHelper from project .mcp.json, plugin, or agent file no longer inherits credential environment variables
24. v2.1.237 Aug 20, 2026 · release notes

    Adds a Concise output style to /config that suppresses preamble and narration

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.237 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.237

    └──▷ HOW TO FIND IT
    Reduce output noise when using Claude Code in a terminal or CI pipeline where only the final result matters.
    📍/config › Output style › Concise
25. v2.1.236 Aug 19, 2026 · release notes

    Adds ANTHROPIC_DEFAULT_MODEL env var, cross-session notify_when_idle, and macOS wildcard sandbox rules

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.236 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.236

    • Added ANTHROPIC_DEFAULT_MODEL environment variable: sets the model new sessions start on, while a /model pick still overrides it and persists across restarts (unlike ANTHROPIC_MODEL)
    • Added notify_when_idle to cross-session SendMessage: ask another Claude Code session on this machine to send one notice when it next goes idle — opt-in, one-shot, no polling (macOS and Linux)
    • Sandbox: on macOS, wildcard read-deny rules (e.g. **/.env) now take precedence inside allowed read regions, cover matched directories' contents, and can't be bypassed by renaming the denied file
    • [VSCode] Added screen reader support for the transcript: live announcements for replies, permission requests, errors, and status changes, plus per-turn heading navigation

    4 details in Everything

26. v2.1.235 Aug 18, 2026 · release notes

    Claude Code v2.1.235 adds optional spellcheck setting with aspell/hunspell/ispell support and hardens SendMessage size limits.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.235 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.235

    └──▷ USE IT
    Enable live spellchecking in the prompt input so typos are underlined as you type — requires aspell, hunspell, or ispell to be installed.
    yaml copy

    spellcheck: true

    • Adds optional spellcheck setting that underlines misspelled words in the prompt input as you type, using your installed aspell, hunspell, or ispell.
    • SendMessage now rejects messages too large for cross-session delivery up front instead of silently dropping them.
    • Remote Control: claude rc now applies the same enterprise-gateway availability check as interactive startup.
    • Improved memory and CPUCPUThe main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. usage while cloud sessions such as /ultrareview or /autofix-pr run in the background — their event streams are no longer re-scanned and re-rendered on every update.
    • Context-limit error now indicates when auto-compact is off and points to /config to re-enable it.

    +1 more show less

    • Vim mode now preserves NORMAL mode and cursor position when toggling the detailed transcript (ctrl+o) or closing a panel.

    6 details in Everything

27. v2.1.234 Aug 17, 2026 · issue 049

    Claude Code v2.1.234 adds CLAUDE_CODE_PROJECT_DIR_NAME, selection:clear keybinding, GitLab MR badges, auto-resume at usage limits, and mid-turn /permissions edits.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.234 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.234

    └──▷ TRY IT
    Give each CI job its own short transcript directory name instead of a long auto-generated one.
    copy

    $ export CLAUDE_CODE_PROJECT_DIR_NAME=my-feature-branch
    claude

    Opt out of the 30-minute background-task check-in so /goal never interrupts long-running agents.
    copy

    $ export CLAUDE_CODE_GOAL_CHECKIN_MINUTES=0
    claude

    • Adds optional CLAUDE_CODE_PROJECT_DIR_NAME environment variable so hosts that give each session its own config directory can choose a short name for the per-project transcript directory.
    • Adds CLAUDE_CODE_GOAL_CHECKIN_MINUTES environment variable (set to 0 to opt out) to control how long /goal waits before checking in on background tasks that have been running 30+ minutes.
    • Adds selection:clear keybinding action, bindable to any key, that clears an in-app text selection in both the main view and the agents view.
    • Adds a GitLab merge request badge to the footer and statusline: repos with a GitLab remote and an authenticated glab CLI show MR !N with draft/pending/green states.
    • Adds automatic session continuation when a claude.ai usage limit resets; toggle off via /config (Continue automatically at usage limit).

    +10 more show less

    • /permissions can now be opened while Claude is actively working, and rule changes apply immediately to the rest of the current turn.
    • /add-dir <path> can now be invoked while Claude is working; /add-dir, /autocompact, /theme, /help, /config, and /advisor dialogs all open mid-turn in the fullscreen TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency..
    • /goal now clears itself with a notice when a turn ends on an unrecoverable error (revoked auth, exhausted credit balance, or context overflow) instead of staying armed.
    • claude setup-token now rejects unexpected extra arguments instead of silently ignoring them.
    • Reduces context cost of loading the built-in claude-api skill from ~200k+ tokens to ~25k by loading reference docs on demand.
    • Remote Control: effort picks made on a phone or on claude.ai/code now apply to terminal- and Desktop/VS Code-hosted sessions, and the session publishes its effort level to connected clients.
    • Remote Control sessions from Claude Code Desktop or VS Code now keep phones and claude.ai/code updated on the session's permission mode and model as they change.
    • SendMessage and ListAgents now report when your account's session list was too long to check completely, instead of treating unseen sessions as absent.
    • Improves auto-generated session titles to read as short, specific names (e.g. Login button bug) rather than sentences restating the request.
    • Removes the Default teammate model setting from /config; agent-team teammates now use the leader's model unless the spawn command names one.

    15 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The Default teammate model setting is removed from /config; agent-team teammates now inherit the leader's model unless the spawn command explicitly names one.
28. v2.1.233 Aug 14, 2026 · issue 046

    Claude Code v2.1.233 adds GitLab MR support, per-user identity forwarding, Linux memory limits for Bash tools, and a WebFetch cache TTLTTLTime to live, a limit on how long data remains valid or how far a packet can travel. Caches use a time limit; IP packets use a hop limit. knob.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.233 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.233

    └──▷ TRY IT
    Cap memory for Bash tool commands on Linux to prevent a runaway build from consuming all RAMRAMWorking memory that holds programs and data while a computer is running. Its contents usually disappear when power is removed. and stalling the session.
    copy

    $ CLAUDE_CODE_TOOL_MEMORY_LIMIT=2G claude

    Shorten the WebFetch URL cache lifetime so repeatedly-fetched pages stay fresh during a long session.
    copy

    $ CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS=300000 claude

    Re-enable todo/task-tracking tools on newer models that disable them by default.
    copy

    $ CLAUDE_CODE_ENABLE_TODO_TOOLS=1 claude

    • Adds CLAUDE_CODE_TOOL_MEMORY_LIMIT environment variable (Linux) to cap memory for Bash tool commands via cgroup support, preventing runaway builds from stalling a session.
    • Adds CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS environment variable to control the WebFetch session URL cache TTL (default: 15 minutes).
    • Adds opt-in forward_user_identity apps gateway setting on Anthropic upstreams, forwarding the signed-in user's identity as headers so a proxy can attribute spend per user.
    • Extends the --worktree flag and the claude agents view to accept GitLab merge request URLs, displaying MRs as !N.
    • Improves claude plugin validate to inspect a bare .claude/skills directory and report SKILL.md files with unparseable frontmatter.

    +5 more show less

    • Adds [claude-code:unrecognized_model] diagnostic line to stderr in print mode when a request is sent for an unrecognized model ID; suppress with modelOverrides.
    • Improves claude self-hosted-runner session start time by creating the session branch without rewriting the working tree and eliminating two blocking server round trips.
    • Improves apps gateway error forwarding: 400/413 responses from Vertex, Foundry, and Claude Platform on AWS upstreams now surface the upstream's own error message.
    • Improves screen reader mode: the /effort selector renders as a numbered list with a typed-number prompt, and hint and dialog text is no longer clipped.
    • Suppresses the GitHub app setup tip in repositories whose origin remote is on gitlab.com or bitbucket.org; the enterprise marketplace tip now covers non-GitHub internal git hosts.

    10 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Todo/task-tracking tools (TaskCreate, TaskGet, TaskUpdate, TaskList, TodoWrite) are no longer available on Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models; set CLAUDE_CODE_ENABLE_TODO_TOOLS=1 to restore them.
29. v2.1.232 Aug 13, 2026 · issue 045

    Claude Code v2.1.232 adds cross-session @-mentions, subagent forking by default, GitLab secret redaction, and GitLab plugin marketplace support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.232 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.232

    └──▷ HOW TO FIND IT
    Configure cross-session messaging policy so inbound messages from your other sessions are held for review rather than accepted automatically.
    📍/config
    Block a private GitLab plugin marketplace from being used, even when the CLI identifies the URL as a git clone.
    json copy

    {
      "blockedMarketplaces": [
        { "type": "url", "value": "gitlab.com/my-org/internal-plugins" }
      ]
    }

    Use the friendlier allowedMarketplaces alias in settings instead of the legacy strictKnownMarketplaces key to restrict which marketplaces Claude Code may use.
    json copy

    {
      "allowedMarketplaces": [
        "gitlab.com/my-org/approved-plugins"
      ]
    }

    • Subagent forking is now on by default: a subagent_type: "fork" subagent inherits the full conversation and prompt cache, and non-teammate agent spawns in interactive sessions now run in the background by default.
    • Type @ in the prompt to mention another Claude session by name; Claude uses SendMessage to reach that session directly — and SendMessage now delivers to a bare name that exactly matches one live session without a confirmation step.
    • Adds /config rows for Dialog expiry and Messages from your other sessions, enabling per-session control over cross-session inbound messaging (accept/hold/refuse).
    • Adds secret redaction for GitLab token families glrt-, gloas-, glptt-, glagent-, glimt-, glsoat-, glcbt-, glft-, and glffct-; fully redacts routable glpat- and gldt- tokens; extends the same sandbox and credential-path protection to the glab CLI config store.
    • Adds GitLab support to plugin marketplaces: bare gitlab.com repo URLs (including nested subgroups) now clone like github.com URLs.

    +13 more show less

    • Settings now accept additionalMarketplaces and allowedMarketplaces as friendlier aliases for extraKnownMarketplaces and strictKnownMarketplaces.
    • Enterprise policy: a url-typed blockedMarketplaces entry for a bare repo URL continues blocking that URL even when the CLI classifies it as a git clone.
    • Gateway: the desktop: overlay now accepts every released Desktop setting (previously limited to 11 hand-listed keys), validated at boot against Desktop's own schema; unknown or invalid keys now fail boot.
    • Gateway: empty managed.policies[].match.groups / admin.admin_groups entries and malformed email_domain values (empty, or containing @, whitespace, or commas) now fail at boot instead of silently matching no one or granting admin access.
    • Fable 5 is offered as an advisor in /advisor again for organizations with Fable access, with usage-credits consent set up through /model fable.
    • Interactive sessions on one machine now keep unique names: renaming a session to a name already in use generates a name-word-word variant and notifies the user.
    • Bash input redirections (< file) are now permission-checked like their argument spellings on all platforms.
    • Sandbox: sandbox.ripgrep is now honored only from user, managed, and --settings settings; project settings can no longer override the sandbox's ripgrep binary.
    • /feedback and /bug now open immediately when invoked while Claude is responding, instead of waiting for the turn to finish.
    • /plugin install plugin@marketplace now refreshes the marketplace first, so newly published plugins install without a manual marketplace update.
    • /code-review at high, xhigh, and max effort now runs in a background agent, consistent with other effort levels.
    • Remote Control now reconnects for approximately 30 minutes after a network blip and no longer drops after a few blips spread across an hour.
    • Managed settings approval dialog now shows endpoint URLs, uses clearer wording for telemetry-only changes, skips routine OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. options, and requires explicit approval for server-managed sandbox binary overrides (sandbox.bwrapPath, sandbox.socatPath, sandbox.ripgrep).

    18 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !sandbox.ripgrep set in project settings is no longer honored; it must now be placed in user, managed, or --settings settings.
    • !Gateway: unknown or invalid keys under the desktop: overlay now fail boot (previously accepted silently).
    • !Gateway: empty managed.policies[].match.groups / admin.admin_groups entries and malformed email_domain values now fail at boot instead of silently matching no one or granting admin access.
30. v2.1.229 Aug 12, 2026 · issue 044

    Claude Code v2.1.229 adds plugin command sources, Remote Control session status, SSESSEServer-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. keepalives, and VSCode session groups.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.229 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.229

    └──▷ TRY IT
    Resume the most recent Remote Control session without looking up its session ID — useful when reconnecting after a network drop.
    copy

    $ claude remote-control --continue

    Disable workflow fan-out staggering in a latency-sensitive pipeline where you do not want agents to wait before starting.
    copy

    $ CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS=0 claude

    • Documents claude remote-control --continue to resume the most recent Remote Control session without specifying a session ID.
    • Adds plugin marketplace command sources: a local command (e.g. an IDE) prints the plugin directory, re-resolved each session without a restart; mode: "link" uses it in place.
    • Sets CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS=0 to disable the new workflow fan-out staggering that lets same-prefix sibling agents share a cached prompt prefix.
    • ListAgents now marks disconnected Remote Control sessions as offline and labels cloud sessions as cloud, giving clearer session inventory.
    • Adds SSE keepalive pings to gateway streaming responses during long thinking pauses, preventing idle-timeout disconnects on Vertex and Bedrock upstreams.

    +5 more show less

    • Adds server-supplied Claude Code hook support for self-hosted runner sessions, matching managed-environment behavior.
    • Self-hosted runner Windows startup now requires an explicit --base-dir; there is no default checkout directory on Windows.
    • Removes auto-approval of git/gh commands with dangerous flags (--force, --amend, --no-verify, etc.) in /commit-push-pr.
    • [VSCode] Adds session groups in the sidebar — right-click to create, rename, or delete; Cmd/Ctrl- or Shift-click to move multiple sessions at once.
    • [VSCode] Makes the /btw side-question panel resizable by dragging its boundary, in both side-docked and stacked layouts.

    10 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Self-hosted runner Windows startup now requires an explicit --base-dir; launches that previously relied on a default checkout directory will fail without it.
    • !/commit-push-pr no longer auto-approves git/gh commands that include dangerous flags (--force, --amend, --no-verify, etc.); workflows depending on silent approval of these flags will now prompt for confirmation.
31. v2.1.228 Aug 11, 2026 · issue 043

    Claude Code v2.1.228 hardens synced skills, improves cross-session messaging display, and lets newer models overwrite unread files.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.228 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.228

    • Hardens skills synced from claude.ai: they no longer shadow local commands or MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. prompts, descriptions are sanitized and labeled, and skill bodies cannot run ! commands or expand @ files on your machine.
    • Changes the Write tool so newer models can overwrite an existing file they have not read in the current session, matching the Edit tool's rules; older models still require a prior read.
    • Improves cross-session messages: sender and body now display inline instead of as a collapsed line, and messages to Remote Control sessions on other machines show your Remote Control session name as the sender.
    • Improves Vertex AI credential handling: expired or missing Google Cloud credentials now fail within seconds instead of retrying for minutes.
    • Improves compaction progress: the retry countdown and stall hint now appear during compaction alongside the progress bar.

    5 details in Everything

32. v2.1.225 Aug 8, 2026 · issue 039

    v2.1.225 adds gateway spend-limit warnings, workspace trust in claude agents, and SendMessage can now initiate Remote Control sessions by name.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.225 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.225

    • Adds gateway spend-limit support to Claude Code's usage warning — the limit-reached message now names the cap, its reset time, and the operator's message (requires gateway on 2.1.225).
    • Adds a workspace trust prompt to claude agents for untrusted directories, matching the existing behavior of claude.
    • SendMessage can now initiate a conversation with Remote Control sessions on other machines by name — ListAgents surfaces them as name [ref] — instead of only being able to reply after they message you first.
    • SendMessage now locks in a Remote Control recipient you have already confirmed, preventing it from being silently swapped for a same-named session on the local machine when the remote list could not be checked.
    • Remote Control: photos attached from the Claude app are now passed directly to Claude rather than being read from disk via a separate tool call.

    5 details in Everything

33. v2.1.224 Aug 7, 2026 · issue 038

    Claude Code v2.1.224 adds self-hosted runners, cross-session messaging, archive plugin sources, and new sandbox credential-masking options.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.224 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.224

    └──▷ TRY IT
    Route Bedrock API calls through a specific cross-region inference profile instead of the one derived from AWS_REGION.
    copy

    $ export ANTHROPIC_BEDROCK_REGION_PREFIX=us-west-2
    claude

    Enable JWTJWTAn open standard (RFC 7519) for encoding claims as a signed, optionally encrypted JSON object passed between parties. Cyber tools use it to verify identity and permissions without querying a central session store.-aware credential masking in the sandbox to redact specific claims from bearer tokensbearer tokensA credential that grants access to whoever presents it, without a separate proof of possession. Protecting the token matters because a stolen copy can be used by someone else. before they leave the environment.
    yaml copy

    decode: "jwt"
    maskClaims:
      - sub
      - email
      - aud

    • New claude self-hosted-runner subcommand turns your own machines or containers into execution environments for Claude Code web, mobile, and desktop sessions (Team and Enterprise plans).
    • New ANTHROPIC_BEDROCK_REGION_PREFIX environment variable lets Bedrock users prefer a specific cross-region inference profile over the AWS_REGION-derived default.
    • New crossSessionInbound and dialogExpiry settings: cross-session messages sent to a bypassed-permission session are held for approval; messages to other sessions auto-deliver.
    • New sandbox credential-masking options — extract and onExtractNoMatch for structured env values, decode: "jwt" with maskClaims for JWT-aware masking, and awsPairs/sigv4 for AWS SigV4SigV4AWS's request-signing protocol that authenticates HTTP calls by hashing the request body, headers, and credentials into a signature; cyber tools use it to talk to AWS APIs without embedding long-lived secrets. re-signing — honored only from user, managed, or --settings settings and require network.tlsTerminate.
    • New archive plugin source: install plugins from a zip over HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. without git or npm, with optional SHA-256SHA-256A cryptographic hash function that turns data into a fixed 256-bit fingerprint. It is used to check integrity; hashing does not encrypt the original data. pinning.

    +5 more show less

    • New cross-session SendMessage API lets Claude Code sessions message each other across any of your machines; ListAgents discovers available sessions (macOS and Linux).
    • Removes the 200-subagent-per-session spawn cap, allowing long-running sessions to continue spawning agents without hitting a hard limit (concurrency and depth limits still apply).
    • Remote Control now shows a persistent failure indicator with details and a reconnect shortcut on connection failure, instead of only an 8-second toast.
    • Remote Control improvements: attached web and mobile clients now see compaction progress and the post-compaction boundary; /clear resets propagate to attached clients.
    • Fullscreen mode now keeps the full pre-compaction history in scrollback across repeated compactions instead of only the most recent interval.

    10 details in Everything

34. v2.1.223 Aug 6, 2026 · issue 037

    Claude Code v2.1.223 adds owner wildcard marketplace controls, /teleport for cloud-to-local handoff, and upgraded /code-review with effort levels.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.223 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.223

    └──▷ USE IT
    Block all marketplace repos from a specific GitHub org in managed settings with a single wildcard entry instead of listing each repo.
    json copy

    {
      "blockedMarketplaces": ["acme-corp/*"]
    }

    Pick up a cloud session on your local machine when you need to switch environments mid-task.
    copy

    $ claude --teleport <session id>

    Run a deep review of a pull request straight from the terminal using the new effort-level syntax.
    copy

    $ /code-review ultra 4321

    • Adds "owner/*" wildcard entries to strictKnownMarketplaces and blockedMarketplaces managed settings to allow or block all marketplace repos under a GitHub org in one entry.
    • Adds claude --teleport <session id> flag and a /teleport hint in cloud sessions to continue a cloud session locally.
    • Adds CLAUDE_CODE_DISABLE_UNKNOWN_MODEL_WINDOW_ENFORCEMENT=1 env var to opt out of the new behavior that keeps sessions on unrecognized model IDs within the assumed context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it..
    • Expands CLAUDE_CODE_DISABLE_1M_CONTEXT to cap every Claude model with a native 1M context window to 200K via auto-compaction (not just a fixed model list); adds a startup warning when auto-compaction isn't holding the session to 200K.
    • Makes /review an alias of /code-review; /code-review now accepts an effort level and PR number (/code-review <level> <pr#>), with /code-review ultra triggering a deep cloud review.

    +2 more show less

    • Makes /code-review remember the last effort level used when invoked without one, so repeat reviews don't require retyping the level.
    • Adds a warning when a workflow agent, forked skill, slash command, or resumed background agent's requested subagent model is restricted and the parent model runs instead.

    7 details in Everything

35. v2.1.222 Aug 4, 2026 · issue 036

    Claude Code v2.1.222 tightens multi-agent permission enforcement and improves Remote Control scoping

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.222 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.222

    • Remote Control auto-start can no longer be enabled via repo-local .claude/settings.json or .claude/settings.local.json; enable it at user scope via /config instead (repo-local settings can still turn it off)
    • Improves auto mode safety: messages dispatched to other agent sessions via SendMessage are now evaluated by the permission classifier before delivery
    • Improves disable-model-invocation refusal behavior: Claude now prompts the user to run the skill directly rather than attempting to replicate its workflow
    • The /diff view, Remote Control workspace diff, and file-edit diffs in web sessions now use raw git blob content, ignoring workspace-configured diff drivers and textconv
    • Removes the ultraplan feature

    5 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Remote Control auto-start can no longer be enabled from .claude/settings.json or .claude/settings.local.json; configurations relying on repo-local settings to turn it on will stop working — use /config at user scope instead.
    • !The ultraplan feature has been removed; any configuration or workflows depending on it will break.
36. v2.1.221 Aug 4, 2026 · issue 035

    Claude Code v2.1.221 adds Focus view, sandbox credential masking, prompt-audit subcommand, and smarter plugin/session management.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.221 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.221

    └──▷ TRY IT
    Audit your prompts and tool descriptions for patterns written for older models before migrating to Claude 4.5-generation.
    copy

    $ claude-api prompt-audit

    Validate a plugin and catch names that would be rejected by Claude Desktop's managed marketplace sync.
    copy

    $ claude plugin validate

    Configure a sandbox credential file to use masking on Linux/WSLWSLWindows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. so sandboxed commands see a sentinel value while the proxy substitutes the real credential on egress.
    yaml copy

    mode: "mask"
    extract: "(sk-[A-Za-z0-9]+)"

    • Adds mode: "mask" for sandbox credential files on Linux and WSL — sandboxed commands read a sentinel copy (whole file or spans captured by an extract regex) while the sandbox proxy substitutes the real value on egress; on macOS falls back to deny.
    • Adds prompt-audit subcommand to the claude-api skill for auditing prompts and tool descriptions for patterns written for older models.
    • Adds warnings to claude plugin validate when a marketplace or plugin name would be rejected by Claude Desktop's managed marketplace sync.
    • [VSCode] Adds Focus view: a chat-menu toggle that hides tool activity behind an expandable per-turn summary with a live running-tool indicator, toggled with Ctrl+Alt+F or the Claude Code: Toggle Focus view command.
    • Changes /status to show the session kind: interactive, or a background job that is attached or unattended.

    +12 more show less

    • Changes sessions forked with /fork to create a new worktree of their own instead of working in the original session's checkout.
    • Changes plugins to accept "." as a skills path, with the root-level SKILL.md validation error now suggesting using the plugin root.
    • Changes /plugin install to refresh a stale marketplace catalog and retry before reporting a plugin not found.
    • Changes plugins installed from /plugin to activate immediately when safe, instead of always requiring /reload-plugins.
    • Changes background sessions to commit and push to preserve work, open a draft PR only when the task calls for one, follow CLAUDE.md git instructions, and always report where the work lives.
    • Improves Stats panel to count cache tokens in its token totals, with a breakdown by input, output, cache read, and cache write.
    • Re-enables tool search on Google Vertex AI for Claude 4.5-generation and newer models.
    • Improves Windows startup by reading process creation times via a native kernel32 call instead of spawning PowerShell, so endpoint security tools that gate powershell.exe no longer prompt.
    • Changes emoji autocomplete to accept common alternate shortcodes like :thumbsup:, :thumbsdown:, and :love:.
    • Changes fast mode to report on the stream when usage credits run out mid-session, instead of failing silently.
    • Changes Monitor: a watch that exits without producing any output now says so instead of reporting stream ended.
    • Changes Gateway model field validation to reject non-string values with a 400 instead of forwarding them.

    17 details in Everything

37. v2.1.219 Jul 24, 2026 · issue 025

    Claude Opus 5 lands as the new default Opus model, with 1M context, strict sandbox networking, and depth-3 nested subagents.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.219 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.219

    └──▷ USE IT
    Lock sandbox network access to only approved hosts so any unexpected outbound call is silently denied without an interactive prompt.
    json copy

    # In your Claude Code settings file (e.g. ~/.claude/settings.json)
    {
      "sandbox": {
        "network": {
          "strictAllowlist": true
        }
      }
    }

    Cap the depth of nested subagent spawning at 1 (flat) to reproduce pre-v2.1.219 behaviour in automation pipelines that shouldn't recurse.
    copy

    $ CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 claude -p "refactor the auth module"

    Pin the Dynamic workflow size guideline org-wide via settings so all users get it without having to touch /config.
    json copy

    # In a shared settings file
    {
      "workflowSizeGuideline": "medium"
    }

    • Adds Claude Opus 5 (claude-opus-5) as the default Opus model — 1M context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it., fast mode at $10/$50 per Mtok.
    • Adds sandbox.network.strictAllowlist setting to silently deny any sandboxed-command network request to a non-allowlisted host (no prompt).
    • Adds DirectoryAdded hook that fires whenever /add-dir or the SDK register_repo_root control request registers a new working directory mid-session.
    • Adds mcp_server_errors field to the headless stream-json init event, listing --mcp-config entries skipped by config validation; terminal runs surface a startup warning.
    • Adds workflowSizeGuideline settings key so the Dynamic workflow size advisory can be set from any settings file (hides the /config row while active).

    +4 more show less

    • Adds nested subagent forwarding in stream-json: depth-2+ subagents now appear in the stream when --forward-subagent-text is set, keyed by their spawning Agent tool_use id.
    • Subagents can now spawn nested subagents up to depth 3 (was 1); set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 to revert to flat spawning.
    • Adds current default workflow size to the running-workflow status line with a pointer to /config for changes.
    • Removes Opus 4.7 from fast mode; /fast now applies to Opus 5 and Opus 4.8 only.

    9 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Opus 4.7 is removed from fast mode; /fast now applies only to Opus 5 and Opus 4.8.
    • !Dynamic workflows now default to a medium size guideline (fewer than 15 agents); workflows relying on the previous unrestricted default will be subject to this new cap unless overridden in /config.
    • !Managed MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. allowlist/denylist ${VAR} entries now resolve from the startup environment and managed-settings env instead of the settings-file env, which may change which values are substituted.
38. v2.1.218 Jul 22, 2026 · issue 023

    Claude Code v2.1.218 runs /code-review as a background subagent, adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. error details, and expands auto-mode intelligence.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.218 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.218

    └──▷ USE IT
    Opt a forked skill back into foreground execution when it must block the conversation before continuing.
    yaml copy

    # In your skill's frontmatter:
    ---
    context: fork
    background: false
    ---

    • Runs /code-review as a background subagent so review output no longer floods the conversation and stacked slash commands stay on target.
    • Adds HTTP status codes and error text to claude mcp list and /mcp when a server fails to connect, plus a warning for MCP config values with hidden leading or trailing whitespace.
    • Adds screen-reader announcements of deleted text for word and line deletions (Option+Delete, Ctrl+W, Cmd+Backspace, Ctrl+U, Ctrl+K) in --ax-screen-reader mode.
    • Announces fast-mode changes when switching models via /config model=<x> or Remote Control.
    • Changes /deep-research to start only on explicit invocation — Claude no longer launches it autonomously.

    +6 more show less

    • Auto mode now routes dangerous-rm, background-&, and suspicious-Windows-path checks through the auto-mode classifier instead of opening permission dialogs.
    • Plan mode with auto no longer prompts for Bash commands the static analyzer can't prove read-only; the auto-mode classifier judges them instead.
    • Skills with context: fork now run in the background by default; individual skills can opt out with background: false in frontmatter.
    • Expands accepted frontmatter boolean values for skills and plugins to include yes/no/on/off/1/0 (case-insensitive) alongside true/false.
    • Agent markdown files now reject agent names containing :, reserving that character for plugin namespacing.
    • Server-managed settings changes for benign feature and cost toggles no longer trigger the settings-approval prompt.

    11 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Agent markdown files with names containing : are now rejected; any agent whose name includes : must be renamed before it will load.
    • !Skills with context: fork now run in the background by default; skills that must run in the foreground need background: false added to their frontmatter.
39. v2.1.217 Jul 21, 2026 · issue 022

    Claude Code v2.1.217 adds emoji autocomplete, subagent concurrency controls, and AWS gateway deployment assets.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.217 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.217

    └──▷ TRY IT
    Cap parallel subagents to a lower value in resource-constrained CI environments to prevent runaway agent fan-out.
    copy

    $ CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS=5 claude

    Allow nested subagent spawning up to two levels deep for multi-stage agentic workflows that require it.
    copy

    $ CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=2 claude

    • Adds emoji shortcode autocomplete in the prompt input (type :heart: to insert or :hea for suggestions); disable with the emojiCompletionEnabled setting.
    • Adds visible warnings when transcript writes are failing (e.g. disk full) or when session saving is disabled via an inherited environment variable, preventing silent transcript loss.
    • Adds a cap on concurrently-running subagents (default 20), overridable with CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS, preventing unbounded agent fan-out from a single message.
    • Adds CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH to control nested subagent depth; nested spawning is now disabled by default.
    • Makes footer PR badge links clickable hyperlinks in environments where terminal support can't be detected (e.g. over SSH/tmux); opt out with FORCE_HYPERLINK=0.

    +1 more show less

    • Adds AWS reference deployment assets for Claude apps gateway on Bedrock under examples/gateway/aws/, including setup.sh, a distroless Dockerfile, gateway.yaml.example (Bedrock/Okta template), and a Terraform module.

    6 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Subagents no longer spawn nested subagents by default; existing workflows that relied on nested subagent spawning will stop working unless CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH is set to permit deeper nesting.
40. v2.1.216 Jul 20, 2026 · issue 021

    Claude Code v2.1.216 adds sandbox filesystem-isolation bypass, smarter /context warnings, and improved /fork, /ultrareview, and /code-review ultra feedback.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.216 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.216

    └──▷ USE IT
    Allow a sandboxed session to read/write the filesystem freely while still blocking unexpected network egress — useful when your task requires broad file access but you still want outbound network control.
    json copy

    # In your Claude Code settings (e.g. .claude/settings.json)
    {
      "sandbox": {
        "filesystem": {
          "disabled": true
        }
      }
    }

    • Adds sandbox.filesystem.disabled setting to skip filesystem isolation while keeping network egress control active.
    • Adds explicit context-window overflow warning to /context and surfaces failed /compact as a visible error.
    • Improves /fork confirmation to include the new session name, claude attach id, and whether the copy shares the checkout.
    • Improves /ultrareview diff-too-large error to display configured limits, measured diff size, and the largest contributing files.
    • Improves /code-review ultra empty-diff message to name the exact base ref and suggest passing an explicit base.

    +3 more show less

    • Improves spend-limit rejection prompt to show the server's stated reason.
    • Background sessions: /mcp and /install-github-app now park a needs input request in the agent view when no client is attached.
    • Adds safety guardrail to /rewind: skips files reachable via symlinks or hard links and reports how many paths were skipped.

    8 details in Everything

41. v2.1.215 Jul 19, 2026 · issue 020

    Claude Code v2.1.215 makes /verify and /code-review user-invoked only — no more automatic runs.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.215 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.215

    └──▷ TRY IT
    Manually trigger a code review after making changes, instead of waiting for Claude to decide to run it automatically.
    copy

    $ /code-review

    Run verification on demand after a refactor to confirm correctness on your own schedule.
    copy

    $ /verify

    • The /verify and /code-review slash commands now run only on explicit user invocation, giving practitioners full control over when those checks execute.

    1 detail in Everything

    └──▷ BREAKING ON UPGRADE

    • !Claude no longer automatically runs the /verify and /code-review skills; any workflow that relied on Claude triggering these autonomously must now invoke /verify or /code-review explicitly.
42. docs update Jul 18, 2026 · issue 019

    Claude Code now runs in the browser and mobile app with --cloud flag for remote, parallel, and local-setup-free tasks.

    └──▷ TRY IT
    Start a cloud-hosted Claude Code session to work on a remote repo or run a long task without local setup.
    copy

    $ claude --cloud

    • New --cloud flag runs Claude Code sessions remotely in the browser with no local setup required.
    • Supports long-running async tasks — kick off work and check back when done.
    • Enables working on repos not cloned locally via cloud-hosted sessions.
    • Run multiple Claude Code tasks in parallel via the cloud mode.
    • Available on desktop browsers and the Claude mobile app (iOS and Android).

    5 details in Everything

43. docs update Jul 18, 2026 · issue 019

    Claude Code adds a cloud/browser mode for running long-running tasks without local setup, accessible on desktop and mobile.

    └──▷ TRY IT
    Start a cloud-based Claude Code session when you don't have a repo locally or want to hand off a long-running task and check back later.
    copy

    $ claude --cloud

    • New --cloud flag launches Claude Code in-browser with no local environment required, enabling long-running async tasks and parallel workstreams.
    • Supports working on repos you don't have checked out locally via the cloud runtime.
    • Cloud sessions are accessible from the Claude mobile app on iOS and Android, letting you kick off tasks on desktop and check back from mobile.

    3 details in Everything

44. v2.1.214 Jul 18, 2026 · issue 019

    Claude Code v2.1.214 adds EndConversation tool, OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. message correlation attributes, docker daemon-redirect permission prompts, and more.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.214 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.214

    └──▷ TRY IT
    Cap OTel content attribute size below the default 60 KB to stay within your observability backend's event-size limits.
    copy

    $ CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH=16384 claude

    • Adds the EndConversation tool so Claude can autonomously terminate sessions involving highly abusive users or jailbreak attempts.
    • Adds a periodic progress heartbeat for long-running tool calls that previously went silent.
    • Adds an ISOISOA widely adopted international standards body that publishes technical specifications; cyber tools cite ISO standards to signal compliance with recognized benchmarks for quality, security, or interoperability. modified timestamp to memory file frontmatter.
    • Adds message.uuid, client_request_id, and tool_source attributes to OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. log events for message-level correlation and tool provenanceprovenanceEvidence recording where a software artifact came from and how it was built, including its source and build process. It helps a consumer check that the artifact matches a trusted build..
    • Adds CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH environment variable to configure the 60 KB truncation limit on OpenTelemetry content attributes.

    +2 more show less

    • Adds reasoning effort to the subagentStatusLine payload so custom agent rows can render model and effort.
    • Adds permission prompts for docker commands (including the Podman docker shim) carrying daemon-redirect flags (--url, --connection, --identity, and Podman's remote mode).

    7 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Single-segment dir/** hook if: conditions now match only <cwd>/dir; write **/dir/** to match at any depth.
    • !file commands using -m/--magic-file or -f/--files-from now require permission instead of being auto-allowed as read-only.
    • !SessionStart hooks now report source "fork" when a session begins as a fork instead of "resume".
45. v2.1.212 Jul 17, 2026 · issue 018

    Claude Code v2.1.212 adds background /fork, auto-mode reset, per-session WebSearch/subagent caps, and auto-backgrounding slow MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. calls.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.212 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.212

    └──▷ TRY IT
    Reset auto-mode to its defaults non-interactively in a script or CI context.
    copy

    $ claude auto-mode reset --yes

    Cap WebSearch calls and subagent spawns to tighter limits for a long-running autonomous session to prevent runaway loops.
    copy

    $ CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION=50 CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION=20 claude

    Lower the MCP auto-background threshold to 30 seconds so slow MCP calls yield the session sooner.
    copy

    $ CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS=30000 claude

    • Redesigns /fork to copy the current conversation into a new background session (visible in claude agents) rather than an in-session subagent; the old in-session subagent behavior is now /subtask.
    • Adds claude auto-mode reset command to restore default auto-mode configuration, with an optional --yes flag to skip the confirmation prompt.
    • Adds a per-session WebSearch call cap (default 200, tunable via CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION) to prevent runaway search loops.
    • Adds a per-session subagent spawn cap (default 200, override via CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION); /clear resets the budget.
    • MCP tool calls exceeding 2 minutes automatically move to the background to keep the session responsive; threshold is configurable or disableable via CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS.

    +8 more show less

    • Typing /resume in the agent view now opens a session picker — including previously deleted sessions — and resumes the chosen session as a background session.
    • Improves prompt caching: mid-conversation system block now works behind LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. gateways and custom base URLs (Bedrock, Vertex, 1P).
    • Reduces token usage in inter-agent messaging by no longer duplicating SendMessage bodies into replayed history and tool results.
    • Extends Enterprise forceLoginMethod enforcement to VS Code extension, SDK, setup-token, and install-github-app logins.
    • Headless/SDK sessions now apply a set_model control request mid-turn so the next LLM round-trip uses the new model immediately.
    • Agent view and claude agents --json now show sessions waiting on a sandbox, MCP-input, or managed-settings prompt as Needs input instead of Working.
    • Session transcripts now record the reasoning effort level on each assistant message.
    • Background agent cold-attach now instantly renders the formatted transcript while the session boots, eliminating the blank-wait screen.

    13 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The Task tool's mode parameter is now ignored; subagents inherit the parent session's permission mode by default.
46. v2.1.211 Jul 15, 2026 · issue 017

    Claude Code v2.1.211 adds subagent stream visibility, cross-worktree permission persistence, and scientific-notation env vars.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.211 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.211

    └──▷ TRY IT
    Capture subagent reasoning in a CI pipeline that parses stream-json to audit what sub-tasks were run.
    copy

    $ claude --output-format stream-json --forward-subagent-text -p 'Refactor the auth module and write tests' | tee agent-stream.ndjson

    Override a token-budget environment variable using scientific notation — useful in shell configs where large numbers are error-prone.
    copy

    $ CLAUDE_CODE_MAX_OUTPUT_TOKENS=1e6 claude -p 'Summarize the entire repo'

    • Adds --forward-subagent-text flag and CLAUDE_CODE_FORWARD_SUBAGENT_TEXT env var to surface subagent text and thinking inside --output-format stream-json output.
    • Changes always allow permission rules to save at the repository root, so approvals granted in one git worktree persist across all worktrees and sessions.
    • Accepts scientific notation and digit-separator spellings (e.g. 1e6, 64_000) for integer environment variables such as timeouts, token budgets, and retry counts.
    • Enables save_to_disk on Claude-in-Chrome screenshot actions to actually write the image to disk and return its path.
    • Improves background agent result reporting so Claude waits for real completion and reports still-running agents instead of fabricating results.

    +3 more show less

    • Adds Vim NORMAL-mode support for s (substitute char) and S (substitute line) in the editor input.
    • Improves the memory index over-limit warning to measure only loaded content, excluding frontmatter and HTML comments.
    • Changes /usage-credits to require confirmation before sending a request to organization admins.

    8 details in Everything

47. v2.1.210 Jul 14, 2026 · issue 015

    Claude Code v2.1.210 adds a live tool timer, memory overflow errors, screen-reader mode announcements, and agent input-waiting counts.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.210 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.210

    • Adds a live elapsed-time counter to the collapsed tool summary line so long-running tool calls visibly tick instead of appearing frozen.
    • Adds a startup warning when permission rules use Write(path), NotebookEdit(path), or Glob(path) — directing users to Edit(path) or Read(path) instead.
    • Memory writes that would push a MEMORY.md index over its read limit now surface an explicit error instead of silently truncating content.
    • Screen reader mode now announces permission mode changes aloud when cycling through modes with Shift+Tab.
    • The agents footer hint now displays how many background agents are waiting on user input, with brief color emphasis when the count changes.

    +3 more show less

    • Hardens the Agent tool against indirect prompt injectionindirect prompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. via content read by a subagent.
    • Improves auto mode: the permission classifier now defaults to Sonnet 5 for external sessions, validated on the first request and pinned for the session.
    • Improves Bash/PowerShell timeout messaging so the model can distinguish a hung command from an intentional background request.

    8 details in Everything

48. v2.1.208 Jul 14, 2026 · issue 015

    Claude Code v2.1.208 adds screen reader mode, vim insert remaps, corporate process wrapper, and fullscreen mouse support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.208 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.208

    └──▷ TRY IT
    Enable screen reader-compatible plain-text output for assistive technology users in a one-off session.
    copy

    $ claude --ax-screen-reader

    Persist screen reader mode across all sessions via settings instead of passing the flag every time.
    json copy

    # In your Claude Code settings (JSON)
    { "axScreenReader": true }

    Route all Claude Code subprocess spawns through a corporate launcher in a managed enterprise environment.
    copy

    $ export CLAUDE_CODE_PROCESS_WRAPPER=/usr/local/bin/corp-launcher
    claude

    • Adds --ax-screen-reader flag (also CLAUDE_AX_SCREEN_READER=1 env var or "axScreenReader": true in settings) for opt-in plain-text rendering compatible with screen readers.
    • Adds vimInsertModeRemaps setting to map two-key vim insert-mode sequences (e.g. jj → Escape) without leaving the editor.
    • Adds CLAUDE_CODE_PROCESS_WRAPPER env var so every Claude Code self-spawn routes through a corporate launcher executable, enabling enterprise wrapper enforcement.
    • Adds mouse-click support for multi-select menus and Other input rows in fullscreen mode.
    • Up to 7x faster tool rounds in print/SDK sessions with many MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools via cached tool-pool assembly.

    +3 more show less

    • Completed background agents now remain listed in /tasks until cleanup instead of disappearing immediately upon finishing.
    • Catastrophic removals (e.g. rm -rf ~) inside command substitutions ($(…), backticks, <(…)) now prompt for confirmation in --dangerously-skip-permissions and auto mode.
    • /usage now displays last-known usage bars with an as of timestamp when the usage endpoint is rate-limited, instead of an error screen.

    8 details in Everything

49. v2.1.207 Jul 11, 2026 · issue 012

    Auto mode goes GA on Bedrock/Vertex/Foundry, Opus 4.8 becomes the default, and plugin config scoping tightens.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.207 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.207

    └──▷ USE IT
    Opt out of auto mode on a Bedrock or Vertex deployment where you want to keep manual control over model selection.
    json copy

    # In ~/.claude/settings.json
    {
      "disableAutoMode": true
    }

    Migrate a per-repo auto mode override to the correct user-level settings file so it is still respected after this release.
    json copy

    # Move autoMode from .claude/settings.local.json to ~/.claude/settings.json
    {
      "autoMode": "auto"
    }

    Update a plugin hook that previously used shell-form ${user_config.*} to the new exec-form with $CLAUDE_PLUGIN_OPTION_<KEY> to avoid the shell-injection rejection.
    json copy

    # Before (now rejected):
    # command: "my-hook --token ${user_config.api_token}"
    
    # After — exec form with env var:
    {
      "args": ["my-hook", "--token"],
      "env": { "TOKEN": "$CLAUDE_PLUGIN_OPTION_API_TOKEN" }
    }

    • Auto mode is now generally available on Bedrock, Vertex AI, and Foundry without the CLAUDE_CODE_ENABLE_AUTO_MODE opt-in flag; disable it via disableAutoMode in settings.
    • Bedrock, Vertex, and Claude Platform on AWS now default to Claude Opus 4.8.
    • Auto mode config is now read exclusively from ~/.claude/settings.json; the repo-resident .claude/settings.local.json autoMode key is no longer honored (breaking change for teams using per-repo overrides).
    • Plugin option values (pluginConfigs) are no longer read from project-level .claude/settings.json; only user-level, --settings, and managed settings are honored (breaking change for project-scoped plugin configs).
    • Plugin hooks/monitors/MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. headersHelper: ${user_config.*} interpolation in shell-form commands is now rejected; hooks must use exec form (args array) or $CLAUDE_PLUGIN_OPTION_<KEY> env vars instead (breaking change).

    +2 more show less

    • /doctor now reports when the launcher at ~/.local/bin/claude is externally managed (e.g. a custom script or symlink).
    • Amounts over $1,000 in /usage-credits now require a typed confirmation, and malformed inputs are rejected with an error rather than silently stripped.

    7 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Auto mode no longer reads autoMode from .claude/settings.local.json (repo-resident); use ~/.claude/settings.json instead
    • !${user_config.*} in shell-form commands for plugin hooks/monitors/MCP headersHelper is now rejected; hooks must use exec form (args array) or $CLAUDE_PLUGIN_OPTION_<KEY>; monitors and headersHelper must read the value inside the script
    • !Plugin option values (pluginConfigs) are no longer read from project-level .claude/settings.json; only user, --settings, and managed settings are honored
50. v2.1.206 Jul 10, 2026 · issue 011

    Claude Code v2.1.206 adds smarter /cd suggestions, /doctor CLAUDE.md trimming, broader push-remote support, and faster background agent upgrades.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.206 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.206

    └──▷ TRY IT
    Identify and trim bloated CLAUDE.md files before committing them — useful when onboarding a repo where the memory file has grown to repeat what the codebase already shows.
    copy

    $ /doctor

    Push and open a PR from a repo whose push remote is set to something other than origin (e.g. a fork workflow with remote.pushDefault=upstream) without manually allowing the remote first.
    copy

    $ /commit-push-pr

    • Adds directory path tab-completion to /cd, matching the existing /add-dir behavior.
    • New /doctor check analyzes checked-in CLAUDE.md files and proposes trimming content Claude can already derive from the codebase.
    • Extends /commit-push-pr to auto-allow git push to the repo's configured remote.pushDefault (or the sole remote when only one exists), not just origin.
    • Enables /login to authenticate against Anthropic-operated public gateway endpoints.
    • EnterWorktree now prompts for confirmation before entering a git worktree that lives outside the project's .claude/worktrees/ directory.

    +3 more show less

    • Background agents now upgrade to the latest Claude Code version immediately after an update, eliminating slow upgrade delays when you attach to a stale session.
    • Agents view status column now uses the full terminal width instead of truncating at 64 characters.
    • Adds Ctrl+X in the agents view to permanently remove a completed session; deleted background jobs now stay deleted across renders.

    8 details in Everything

51. v2.1.205 Jul 8, 2026 · issue 009

    Claude Code v2.1.205 upgrades /doctor into a full setup checkup, hardens auto mode safety rules, and improves the agents view.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.205 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.205

    └──▷ TRY IT
    Run a full environment health check and let Claude Code automatically fix any issues it finds.
    copy

    $ /doctor

    Review the enriched agent list — with colored state labels, classifier headlines, and PR links — for all background sessions.
    copy

    $ claude agents

    • Adds auto mode rule that blocks tampering with session transcript files.
    • Improves auto mode to prompt for confirmation before running rm -rf on an unresolvable variable, reducing accidental data loss.
    • Upgrades /doctor into a full setup checkup that can both diagnose and fix issues; adds /checkup as an alias.
    • Enhances claude agents view: PR-linked sessions (edit, merge, comment, push) now surface the PR link directly in the list.
    • Enhances claude agents view: rows show a colored state word and classifier-written headline instead of raw tool call text; peek view shows full status and the exact ask for blocked sessions.

    +3 more show less

    • Background task notifications now explicitly state no human input has occurred, preventing fabricated in-transcript approvals from being acted on.
    • Streams auto-update binary downloads to disk instead of buffering in memory, cutting peak updater memory usage by ~400 MB.
    • Reserves the Claude Browser MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server name (alongside Claude Preview); user-configured MCP servers can no longer register under either name.

    8 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !User-configured MCP servers can no longer register under the names Claude Browser or Claude Preview
52. v2.1.203 Jul 7, 2026 · issue 009

    Claude Code v2.1.203 adds login-expiry warnings, manual-mode visibility, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. roots updates, and a VSCode remote-control toggle.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.203 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.203

    • Adds a pre-expiry warning when your login session is about to expire, letting you re-authenticate before background sessions are interrupted.
    • Adds a grey ⏸ badge in the footer when manual permission mode is active, so the current permission mode is always visible.
    • Exposes the session's additional working directories in MCP roots/list, and sends notifications/roots/list_changed whenever that set changes — enabling MCP servers to stay in sync with multi-root workspaces.
    • Changes the Escape key (instead of left arrow) to close background tasks, diff, and workflow detail views — practitioners must update muscle memory.
    • Changes the empty claude agents view to always display organized sections (Needs Input / Working / Completed) with descriptions, even when no agents are running.

    +2 more show less

    • Moves startup 'claude command missing or broken' warnings out of the startup flow into /doctor and /status only.
    • [VSCode] Adds a Settings toggle for Enable Remote Control for all sessions.

    7 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The left arrow key no longer closes the background tasks, diff, and workflow detail views — press Esc instead
53. v2.1.202 Jul 6, 2026 · issue 008

    Claude Code v2.1.202 adds dynamic workflow sizing, OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. workflow attributes, and a new /code-review command for multi-agent PR review.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.202 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.202

    └──▷ TRY IT
    Run a thorough multi-agent review on a PR when you want deeper analysis than the default single-pass review.
    copy

    $ /code-review high 4821

    Tune agent parallelism for a complex workflow without hitting a hard cap — useful before kicking off a large refactor.
    📍/config
    • Adds a Dynamic workflow size setting in /config to guide how many agents Claude spawns for dynamic workflows (small/medium/large) — advisory, not an enforced cap.
    • Adds workflow.run_id and workflow.name OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. attributes to telemetry from workflow-spawned agents, enabling full reconstruction of a workflow run from OTel data.
    • Introduces /code-review <level> <pr#> for multi-agent PR review at a chosen effort level; /review <pr> reverts to a fast single-pass review.
    • Improves /workflows agent list layout with wider titles, a dedicated time column, and shorter model names.
    • Improves MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. error messages to suggest "type": "http" when a server config has url but no type.

    5 details in Everything

54. v2.1.200 Jul 3, 2026 · issue 004

    Claude Code v2.1.200 makes Manual the default permission mode and adds opt-in idle timeout for user-question dialogs.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.200 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.200

    └──▷ TRY IT
    Enforce Manual permission mode explicitly in automation or shared configs to match the new canonical name.
    copy

    $ claude --permission-mode manual

    • Changes AskUserQuestion dialogs to no longer auto-continue by default; opt into an idle timeout via /config.
    • Renames the default permission mode to Manual across CLI, --help, VS Code, and JetBrains; accepts --permission-mode manual and "defaultMode": "manual" as canonical values.

    2 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The AskUserQuestion dialogs no longer auto-continue by default; auto-continue must be explicitly opted into via /config
    • !The default permission mode is renamed from default to Manual across the CLI, --help, VS Code, and JetBrains (though --permission-mode default and "defaultMode": "default" remain accepted alongside manual)
55. v2.1.199 Jul 2, 2026 · issue 003

    Stacked slash-skills (up to 5), auto-retry on 429s, and a raised retry ceiling via CLAUDE_CODE_MAX_RETRIES

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.199 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.199

    └──▷ TRY IT
    Apply multiple skills together in one prompt to layer context (e.g. a security-review skill and a repo-conventions skill) before asking Claude to act.
    copy

    $ /skill-security-review /skill-repo-conventions find all places we call eval() and suggest safer alternatives

    On long-running agentic jobs in flaky network environments, raise the retry ceiling so transient 429s don't abort the session.
    copy

    $ CLAUDE_CODE_MAX_RETRIES=200 claude 'run the full test suite and summarise failures'

    • Supports stacking up to 5 slash-skill invocations in a single prompt (e.g. /skill-a /skill-b /skill-c do XYZ), loading all leading skills instead of just the first.
    • Auto-retries transient server rate-limit errors (HTTP 429s unrelated to usage limits) with backoff for subscribers, instead of failing the turn.
    • New CLAUDE_CODE_RETRY_WATCHDOG behavior raises the default retry count for non-capacity transient errors to 300 and removes the previous hard cap of 15 on CLAUDE_CODE_MAX_RETRIES.
    • Idle surplus subagents in the agent panel now collapse into an expandable summary row instead of vanishing.

    4 details in Everything

56. v2.1.198 Jul 1, 2026 · issue 002

    Claude Code v2.1.198 adds Chrome GA, automated draft PRs from background agents, /dataviz skill, and AWS gateway support.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.198 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.198

    └──▷ TRY IT
    Get notified in your hook handler when a background agent needs input or completes, so you can automate follow-up actions.
    copy

    $ # In your Notification hook handler, inspect the event type:
    # agent_needs_input — agent is waiting for a response
    # agent_completed  — agent finished its work
    claude agents

    Use the new /dataviz skill inside a session to get chart design guidance and validate color palettes for dashboards.
    copy

    $ /dataviz

    • Launches Claude in Chrome as generally available.
    • Adds agent_needs_input and agent_completed Notification hook events in claude agents so background sessions fire alerts when they need input or finish.
    • New /dataviz skill provides chart and dashboard design guidance with a runnable color-palette validator.
    • Adds Claude Platform on AWS (anthropicAws) as a gateway upstream provider; model-not-found responses now advance the failover chain instead of failing hard.
    • Background agents launched from claude agents now automatically commit, push, and open a draft PR when they finish code work in a worktree.

    +4 more show less

    • Built-in Explore agent now inherits the main session's model (capped at opus) instead of always running on haiku.
    • Subagents and context compactioncontext compactionCondensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost. now inherit the session's extended thinkingextended thinkingIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. configuration, improving output quality on delegated tasks.
    • Removes the /agents wizard; subagents are now created by asking Claude directly or editing .claude/agents/.
    • Subagents now treat messages from their launching agent as normal task direction (never as user approval), clarifying the trust model.

    9 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The --bg flag combined with --print/-p is now rejected up front instead of silently creating an unattachable session.
    • !The /agents wizard has been removed; subagents must now be created by asking Claude or by editing .claude/agents/ directly.
57. v2.1.197 Jun 30, 2026 · issue 002

    Claude Code now defaults to Claude Sonnet 5 with a native 1M-token context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. and promotional pricing through August 31.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.197 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.197

    • Switches default model to Claude Sonnet 5, featuring a native 1M-token context window for working across very large codebases in a single session.
    • Introduces promotional pricing of $2/$10 per million input/output tokens for Claude Sonnet 5 through August 31.

    2 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Claude Sonnet 5 is now the default model in Claude Code, replacing the previously default model
58. v2.1.196 Jun 29, 2026 · issue 002

    Claude Code v2.1.196 adds org-default models, clickable file attachments, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. security hardening, and an always-on stream watchdog.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.196 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.196

    └──▷ TRY IT
    Inspect which MCP servers are approved before trusting a cloned repo — untrusted servers now show as pending instead of silently starting.
    copy

    $ claude mcp list

    • Adds organization default model support — admins configure it in the org console and users see Org default or Role default in /model when no personal selection is made.
    • Adds readable auto-generated names for sessions at startup, making sessions easier to identify and reference.
    • Adds clickable file attachments in chat — Cmd/Ctrl-click a file to reveal it in Finder/Explorer.
    • Security hardening: claude mcp list/get no longer auto-spawn MCP servers that a repo self-approved via a committed .claude/settings.json; untrusted workspaces now show a ⏸ Pending approval indicator instead.
    • Enables the streaming idle watchdog by default for all providers — aborts and retries any response stream that produces no events for 5 minutes; set CLAUDE_ENABLE_STREAM_WATCHDOG=0 to opt out.

    +4 more show less

    • Disables Remote Control when ANTHROPIC_BASE_URL points at a non-Anthropic host, consistent with existing Bedrock/Vertex/Foundry behavior.
    • Simplifies agents view navigation from a foreground session — now requires a single ← press instead of two.
    • Improves /code-review workflow by consolidating five cleanup finders into one, cutting token usage by ~25%.
    • Improves background agent resilience — workers killed by a daemon restart auto-resume from their last state when the agents view next opens.

    9 details in Everything

59. v2.1.195 Jun 26, 2026 · issue 001

    Adds CLAUDE_CODE_DISABLE_MOUSE_CLICKS env var and remote session provisioning checklist in v2.1.195

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.195 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.195

    └──▷ TRY IT
    Prevent Claude Code's fullscreen UI from intercepting mouse clicks in a tmux or remote terminal session while still allowing scroll.
    copy

    $ CLAUDE_CODE_DISABLE_MOUSE_CLICKS=1 claude

    • Adds CLAUDE_CODE_DISABLE_MOUSE_CLICKS environment variable to suppress mouse click, drag, and hover events in fullscreen mode while preserving wheel scroll — useful when running Claude Code inside terminal multiplexers or remote sessions where mouse events conflict with the host terminal.
    • Adds a provisioning checklist during Remote session startup so practitioners can track container initialization progress in real time.
    • Hook matchers now require exact matching for hyphenated identifiers (e.g. mcp__brave-search); use mcp__brave-search__.* glob syntax to match all tools from a hyphenated MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server — a breaking change for any hooks relying on previous substring behavior.
    • Linux voice mode now distinguishes between no microphone detected and SoX not installed when SoX is present but no audio capture device exists, enabling clearer troubleshooting for automated or headless environments.

    4 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Hook matchers now exact-match hyphenated identifiers (e.g. code-reviewer, mcp__brave-search) instead of substring-matching; existing configurations that relied on substring matching must be updated to use explicit wildcards (e.g. mcp__brave-search__.*) to match all tools from a hyphenated MCP server.
60. v2.1.193 Jun 25, 2026 · issue 001

    Claude Code v2.1.193 adds shell classifier controls, OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. response logging, live path autocomplete, and memory-pressure reaping.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.193 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.193

    └──▷ TRY IT
    Enforce auto-mode classification on every shell command — useful in CI/enterprise deployments where you want consistent policy enforcement regardless of command pattern.
    copy

    $ claude config set autoMode.classifyAllShell true

    Capture full conversation transcripts (prompts + responses) in your OTel pipeline for audit or observability — opt in explicitly so response text isn't silently included on upgrade.
    copy

    $ OTEL_LOG_ASSISTANT_RESPONSES=1 claude

    Run Claude Code in memory-constrained environments without the automatic background-shell reaper interfering with long-running idle shells you want to keep alive.
    copy

    $ CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1 claude

    • New autoMode.classifyAllShell setting routes all Bash/PowerShell commands through the auto-mode classifier, not just arbitrary-code-execution patterns.
    • Surfaces auto-mode denial reasons in the transcript, denial toast, and /permissions recent-denials view.
    • New claude_code.assistant_response OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. log event captures model response text (redacted by default; enable with OTEL_LOG_ASSISTANT_RESPONSES=1).
    • Adds live file-path autocomplete when typing in bash mode (!).
    • Startup notice now appears when MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers require authentication, directing users to /mcp.

    +4 more show less

    • Automatic memory-pressure reaping terminates idle background shell commands to reclaim resources (disable with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1).
    • Background agents now continue working on additional tasks after launching a sub-agent instead of being instructed to end their response.
    • MCP headersHelper auth automatically re-runs and reconnects when a tool call returns a 401/403.
    • Plugin auto-rename now follows marketplace renames maps, automatically updating settings to the new plugin name.

    9 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Deployments that already log prompt content via OTEL_LOG_USER_PROMPTS will automatically start receiving assistant response content in claude_code.assistant_response OpenTelemetry log events on upgrade; set OTEL_LOG_ASSISTANT_RESPONSES=0 to keep prompts-only logging.
61. v2.1.191 Jun 24, 2026 · issue 001

    Claude Code v2.1.191 adds /rewind for post-clear conversation recovery and cuts streaming CPUCPUThe main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. usage by ~37%.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.191 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.191

    └──▷ TRY IT
    Recover context after accidentally running /clear mid-session — rewind to the last pre-clear state and continue where you left off.
    copy

    $ /rewind

    • Adds /rewind command to resume a conversation from the state before /clear was last run, enabling recovery of cleared context.
    • Sandbox network permission dialog now remembers hosts approved with Yes for the entire session, eliminating repeated prompts for the same host.
    • MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. capability discovery (tools/list, prompts/list, resources/list) now retries on transient network errors with backoff, improving server reliability.
    • MCP OAuthMCP OAuthAn authentication layer for the Model Context Protocol that lets MCP clients obtain delegated access tokens so cyber tools can call MCP servers on a user's behalf without handling raw credentials. in headless environments skips the browser popup and goes directly to a paste-the-URL prompt.
    • MCP HTTP 404 errors now surface the failing URL and link to the MCP config for faster troubleshooting.

    +2 more show less

    • Reduces streaming response CPU usage by ~37% via 100ms text-update coalescing, enabling longer uninterrupted agentic sessions.
    • Reduces memory growth in long sessions by optimizing the terminal output cache.

    7 details in Everything

62. v2.1.187 Jun 23, 2026 · issue 001

    Claude Code v2.1.187 adds credential sandboxingsandboxingRunning code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls., org model restrictions, mouse support in fullscreen, and UX improvements to /btw, /plugin, and /install-github-app.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.187 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.187

    └──▷ TRY IT
    Prevent sandboxed subcommands from exfiltrating credentials or secret env vars — useful when running untrusted or AI-generated shell commands.
    copy

    $ claude config set sandbox.credentials false

    • Adds sandbox.credentials setting to block sandboxed commands from reading credential files and secret environment variables.
    • Adds org-configured model restrictions enforced across the model picker, --model, /model, and ANTHROPIC_MODEL, with a clear 'restricted by your organization's settings' message.
    • Adds mouse click support for selecting menus (permission prompts, /model, /config, etc.) in fullscreen mode.
    • Improves /install-github-app so GitHub ActionsGitHub ActionsGitHub's built-in CI/CD system that runs automated workflows triggered by repository events, letting cyber tools automate testing, building, and releasing without managing separate pipeline infrastructure. workflow setup is now optional — install just the GitHub App without configuring the workflow or secrets.
    • Improves /btw with ←/→ arrow navigation to step through earlier answers.

    +1 more show less

    • Improves /plugin to surface recently unused plugins so practitioners can clean them up.

    6 details in Everything

63. v2.1.186 Jun 22, 2026 · issue 001

    Claude Code v2.1.186 adds CLI MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. auth commands, bash auto-response, AWS credential refresh, and subagent permission surfacing.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.186 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.186

    └──▷ TRY IT
    Authenticate an MCP server over an SSH session where a browser cannot open — complete the OAuth flow via stdin redirect instead.
    copy

    $ claude mcp login my-server --no-browser

    Use CLAUDE_CODE_RETRY_WATCHDOG instead of CLAUDE_CODE_MAX_RETRIES for robust unattended CI/CD pipeline runs now that the latter is capped at 15.
    copy

    $ CLAUDE_CODE_RETRY_WATCHDOG=1 claude --print 'run the full test suite and fix any failures'

    • Adds claude mcp login <name> and claude mcp logout <name> commands to authenticate MCP servers directly from the CLI, with --no-browser stdin redirect for SSH sessions.
    • Enables ! bash commands to automatically trigger Claude to respond to their output; disable with "respondToBashCommands": false in settings.json to restore context-only behavior.
    • Adds status filtering (press f) to the /workflows agent detail view for narrowing displayed agents by state.
    • Adds a Skills section to the /plugin Installed tab for visibility into installed skills.
    • Adds teammateMode: "iterm2" setting for iTerm2 pane-based teammate integration, with a warning when the it2 CLI cannot be found.

    +6 more show less

    • Adds a 'Claude Platform on AWS - refresh credentials' option to /login when awsAuthRefresh is configured.
    • Surfaces background subagent permission prompts in the main session instead of auto-denying them; the dialog identifies which agent is requesting, and Esc denies only that tool.
    • Changes /review <pr> to use the same engine as /code-review medium for consistent PR review quality.
    • Caps CLAUDE_CODE_MAX_RETRIES at 15; introduces CLAUDE_CODE_RETRY_WATCHDOG as the recommended alternative for unattended sessions.
    • Supports kebab-case, snake_case, and camelCase for skill frontmatter keys (display-name, default-enabled, fallback, metadata.*).
    • Loads malformed SKILL.md YAML frontmatter with empty metadata instead of silently failing, surfacing skills even when their headers are broken.

    11 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The ! bash commands now trigger Claude to respond to the output automatically by default; set "respondToBashCommands": false in settings.json to keep the previous context-only behavior
    • !CLAUDE_CODE_MAX_RETRIES is now capped at 15; unattended sessions relying on higher values must switch to CLAUDE_CODE_RETRY_WATCHDOG instead
64. v2.1.183 Jun 19, 2026 · issue 001

    Claude Code v2.1.183 adds auto-mode guardrails against destructive commands, attribution controls, and improved /config UX.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.183 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.183

    └──▷ TRY IT
    Discover all shorthand config keys you can set inline without navigating the interactive menu.
    copy

    $ /config --help

    • Adds auto-mode safety guardrails that block destructive git commands (git reset --hard, git checkout -- ., git clean -fd, git stash drop) unless you explicitly asked to discard local work, blocks git commit --amend on commits not made by the agent this session, and blocks terraform destroy/pulumi destroy/cdk destroy unless a specific stack was requested.
    • Adds deprecation warnings when a requested model is deprecated or auto-upgraded, surfaced on stderr in print mode (-p) and now also covering models set in agent frontmatter.
    • New attribution.sessionUrl setting lets you omit the claude.ai session link from commits and PRs in web and Remote Control sessions.
    • New /config --help subcommand lists all available shorthand keys for use with /config key=value.
    • Updated /config toggle behavior: Enter and Space both toggle the selected setting; Esc now saves and closes (previously reverted changes).

    +1 more show less

    • Removes the startup setup issues line from under the logo — configuration issues are now surfaced on demand via /doctor or --debug.

    6 details in Everything

65. v2.1.181 Jun 17, 2026 · issue 001

    Claude Code v2.1.181 adds inline /config setting, Apple Events sandbox opt-in, and presence-based push notification suppression.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.181 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.181

    └──▷ TRY IT
    Toggle extended thinkingextended thinkingIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. off mid-session without leaving the prompt — useful when you want faster, cheaper responses for a straightforward task.
    copy

    $ /config thinking=false

    Suppress mobile push notifications automatically while you're at your workstation by pointing Claude Code at a marker file your session manager creates on login.
    copy

    $ CLAUDE_CLIENT_PRESENCE_FILE=~/.at-desk claude

    • Adds /config key=value slash command to set any Claude Code setting mid-session from the prompt (e.g. /config thinking=false), usable in interactive mode, -p, and Remote Control.
    • Adds sandbox.allowAppleEvents opt-in setting so sandboxed commands can send Apple Events on macOS — enables osascript and similar automation inside the sandbox.
    • Adds CLAUDE_CLIENT_PRESENCE_FILE environment variable to suppress mobile push notifications when a specified marker file is present, signaling you're already at the machine.
    • Upgrades bundled Bun runtime to 1.4.
    • Enables auto-retry on API connection drops mid-thinking, eliminating manual intervention when the connection closes during extended reasoning.

    +3 more show less

    • Improves subagent panel UX: idle subagents auto-hide after 30 seconds, list caps at 5 rows with scroll hints, and keyboard hints appear in the footer.
    • Changes fullscreen mode URL opening to require Cmd+click (macOS) / Ctrl+click, aligning with native terminal behavior.
    • Improves long-paragraph streaming so text appears line-by-line rather than waiting for the first line break.

    8 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Fullscreen mode URL opening now requires Cmd+click (macOS) / Ctrl+click instead of a plain click, matching native terminal behavior
66. v2.1.178 Jun 15, 2026 · issue 001

    Claude Code v2.1.178 adds parameter-scoped permission rules, nested skills/config resolution, and streamlined agent teams.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.178 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.178

    └──▷ TRY IT
    Block subagents from spawning with a specific model (e.g. Opus) by adding a parameter-scoped deny rule to your permission config.
    copy

    $ claude --disallow-tool 'Agent(model:opus)' "Refactor the auth module"

    Spawn a named teammate directly without a team-setup step when agent teams are enabled.
    copy

    $ CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 claude "Use the Agent tool with name:reviewer to audit my PR diff"

    • Adds Tool(param:value) syntax for permission rules to match a tool's input parameters with * wildcard support — e.g. Agent(model:opus) to block Opus subagents.
    • Supports skills in nested .claude/skills directories, loading them automatically when working in that subtree; name clashes surface as <dir>:<name> so both remain accessible.
    • Nearest-wins resolution for nested .claude/ directories: the agent, workflow, and output-style closest to the working directory take precedence on name collisions; project-scope workflow saves target the closest existing .claude/workflows/.
    • Strengthens auto mode by evaluating subagent spawns through the classifier before launch, closing a gap where a subagent could request a blocked action without review.
    • Removes TeamCreate and TeamDelete tools for agent teams (experimental): with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, every session now has one implicit team — spawn teammates directly via the Agent tool's name parameter.

    +3 more show less

    • Enhances /doctor with a consistent flat tree layout, clearer section status icons, and highlighted command names.
    • Changes the workflow prompt keyword to use a purple shimmer highlight and trigger only on explicit phrases like run a workflow or "workflow:", not on any mention of the word.
    • Improves Remote Control error messages: connection failures show a persistent red /rc failed footer indicator, and the not yet enabled error now explains whether the cause is a gate, check failure, stale entitlement, or org policy.

    8 details in Everything

67. v2.1.176 Jun 12, 2026 · release notes

    Claude Code v2.1.176 adds language-aware session titles, footer link badges, and smarter Bedrock credential caching.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.176 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.176

    • Session titles are now generated in the language of the conversation; pin a specific language with the language setting.
    • New footerLinksRegexes setting lets you add regex-matched link badges to the footer row via user or managed settings.

    2 details in Everything

68. v2.1.175 Jun 12, 2026 · release notes

    Claude Code adds enforceAvailableModels managed setting to lock model selection within admin-defined allowlists.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.175 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.175

    └──▷ USE IT
    Lock an enterprise deployment so users can never select a model outside the admin-approved list, even via project or user settings.
    json copy

    {
      "enforceAvailableModels": true,
      "availableModels": ["claude-3-5-sonnet-20241022", "claude-3-haiku-20240307"]
    }

    • Adds enforceAvailableModels managed setting that forces the Default model to fall back to the first allowed model when it would otherwise resolve to a disallowed model, and prevents users or projects from widening a managed availableModels list.

    1 detail in Everything

69. v2.1.174 Jun 12, 2026 · release notes

    Claude Code v2.1.174 adds wheel-scroll acceleration control and detailed VS Code usage attribution breakdowns.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.174 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.174

    └──▷ HOW TO FIND IT
    Review where your token spend is going — broken down by subagents, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools, skills, and context size — without leaving VS Code.
    📍In VS Code, open the Claude Code panel, then run /usage to open the Account & usage dialog showing the last 24h or 7d breakdown.
    • Adds wheelScrollAccelerationEnabled setting to disable mouse-wheel scroll acceleration in fullscreen mode.
    • Adds usage attribution panel in the VS Code Account & usage dialog (/usage) showing cache misses, long context, subagents, and per-skill/agent/plugin/MCP breakdowns over the last 24h or 7d.

    2 details in Everything

70. v2.1.172 Jun 10, 2026 · release notes

    Claude Code v2.1.172 adds recursive sub-agents (5 levels deep), Bedrock region auto-detection, plugin search, and a new OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. model attribute.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.172 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.172

    └──▷ TRY IT
    Inspect which AWS region Bedrock is using and where it came from, without needing to set AWS_REGION explicitly.
    copy

    $ /status

    Filter a long marketplace plugin list without scrolling through every entry.
    📍/plugin
    • Enables sub-agents to spawn their own sub-agents, up to 5 levels of nesting.
    • Amazon Bedrock now auto-detects AWS region from ~/.aws config files following AWS SDK precedence when AWS_REGION is unset; /status shows the region source.
    • Adds a search bar inside the /plugin marketplace browser for filtering plugins.
    • Adds a model attribute to the claude_code.lines_of_code.count OTEL metric, enabling per-model telemetry breakdowns.

    4 details in Everything

71. v2.1.170 Jun 9, 2026 · release notes

    Claude Code v2.1.170 adds access to Claude Fable 5, Anthropic's most capable generally available model.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.170 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.170

72. v2.1.169 Jun 8, 2026 · release notes

    Claude Code v2.1.169 adds --safe-mode, /cd command, and disableBundledSkills for cleaner troubleshooting and session control.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.169 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.169

    └──▷ TRY IT
    Troubleshoot a misbehaving setup by launching Claude Code with all customizations stripped out — no CLAUDE.md, hooks, plugins, or MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers.
    copy

    $ claude --safe-mode

    Switch your active session to a different project directory without losing prompt cache continuity.
    copy

    $ /cd ~/projects/other-repo

    List all background agent sessions — including completed ones — with their IDs and states for scripting or monitoring pipelines.
    copy

    $ claude agents --json --all

    • Adds --safe-mode flag (and CLAUDE_CODE_SAFE_MODE env var) to launch with all customizations disabled — CLAUDE.md, plugins, skills, hooks, and MCP servers — for clean troubleshooting.
    • Adds /cd command to change the working directory mid-session without breaking the prompt cache.
    • Adds disableBundledSkills setting and CLAUDE_CODE_DISABLE_BUNDLED_SKILLS env var to hide bundled skills, workflows, and built-in slash commands from the model.
    • Adds --all flag to claude agents --json to include completed sessions; also surfaces new id and state fields on each session entry.
    • The claude agents --json command now correctly includes blocked and just-dispatched background sessions.

    +5 more show less

    • Background sessions now preserve --ide, --chrome, --bare, --remote-control, and other flags across retire→wake cycles.
    • Remote-managed settings with an invalid entry now apply remaining valid policies and surface the validation error instead of silently dropping the whole payload.
    • /workflows command now opens immediately even while a turn is in progress.
    • The CLAUDE.md is too long warning threshold now scales with the model's context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it..
    • Adds a tip suggesting claude agents when running multiple concurrent sessions.

    10 details in Everything

73. v2.1.166 Jun 6, 2026 · release notes

    Claude Code v2.1.166 adds fallback model chains, glob-based tool deny rules, and hardened cross-session security.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.166 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.166

    └──▷ TRY IT
    Run a model that thinks by default without extended thinkingextended thinkingIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. to reduce latency and token cost.
    copy

    $ MAX_THINKING_TOKENS=0 claude -p "Analyze this code for vulnerabilities" < main.go

    • Adds fallbackModel setting to configure up to three fallback models tried in order when the primary model is overloaded or unavailable; --fallback-model now also applies to interactive sessions.
    • Adds glob pattern support in deny rule tool-name position ("*" denies all tools); allow rules reject non-MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. globs, and unknown tool names in deny rules warn at startup.
    • Hardens cross-session messaging: messages relayed via SendMessage from other Claude sessions no longer carry user authority — receivers refuse relayed permission requests, and auto mode blocks them.
    • Enables disabling thinking on models that think by default via the Claude API using MAX_THINKING_TOKENS=0, --thinking disabled, or the per-model thinking toggle (3P providers unchanged).
    • Retries a turn once on the fallback model when the API rejects an unexpected non-retryable error; auth, rate-limit, request-size, and transport errors still surface immediately.

    +2 more show less

    • claude update now announces the target version before downloading.
    • In claude agents, typing a URL into the session list now filters to the session whose first prompt contained it.

    7 details in Everything

74. v2.1.163 Jun 4, 2026 · release notes

    Claude Code v2.1.163 adds version enforcement settings, /plugin list, hook feedback context, and stdio MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. session IDs.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.163 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.163

    └──▷ USE IT
    Enforce an approved version range across a managed fleet so users on out-of-range builds are blocked at startup.
    json copy

    {
      "requiredMinimumVersion": "2.1.163",
      "requiredMaximumVersion": "2.1.999"
    }

    Audit which plugins are active in the current session — useful after an update or policy change.
    copy

    $ /plugin list --enabled

    Surface stop-hook diagnostics back to Claude (e.g. a linter summary) so it can react without the run being marked as a hook error.
    json copy

    {
      "hookSpecificOutput": {
        "additionalContext": "Linter found 3 warnings in src/main.ts — please review before finishing."
      }
    }

    • Adds requiredMinimumVersion and requiredMaximumVersion managed settings to enforce a permitted Claude Code version range, refusing startup if the installed version falls outside it.
    • New /plugin list command lists installed plugins, with --enabled/--disabled filters to narrow results.
    • Adds a c to copy shortcut in /btw that copies the raw markdown answer to the clipboard, preserving formatting on paste.
    • Skills: new \$ escape syntax lets command bodies include a literal $ before a digit.
    • stdio MCP servers now receive the same CLAUDE_CODE_SESSION_ID environment variable as hooks and Bash when using --resume.

    +1 more show less

    • Background agent sessions now update to a new Claude Code version silently in the background, eliminating cold-restart wait times when reopening a session after an update.

    6 details in Everything

75. v2.1.162 Jun 3, 2026 · release notes

    claude agents gets richer status info, --tools Grep/Glob now works on native builds, and /effort persists as default.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.162 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.162

    └──▷ TRY IT
    Poll agent session status in CI/automation to detect which sessions are blocked waiting for a permission prompt.
    copy

    $ claude agents --json | jq '.[] | select(.waitingFor != null) | {id, waitingFor}'

    Restrict Claude Code to only built-in Grep and Glob search tools on a native build — previously listing these names was silently ignored.
    copy

    $ claude --tools Grep,Glob

    • Adds waitingFor field to claude agents --json output, showing what a waiting session is blocked on (e.g. a permission prompt).
    • Explicitly listing Grep or Glob in --tools now activates the dedicated embedded search tools on native builds (previously silently ignored).
    • The /effort command now confirms when the chosen level will persist as the default for new sessions.
    • Clicking a slash command in the autocomplete menu fills it into the prompt instead of running it immediately — press Enter to run.
    • Remote Control status is now shown as a persistent footer pill with a session link instead of a one-time startup message.

    +2 more show less

    • Launch-prompt warnings (deep link/pre-filled prompt) now stay pinned below the input until acted upon instead of scrolling away.
    • Background service startup and claude update now wait for endpoint-security scanning of new binaries instead of timing out after 5 seconds.

    7 details in Everything

76. v2.1.161 Jun 2, 2026 · release notes

    Claude Code v2.1.161 adds OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. label slicing, smarter agents UI, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. connector collapsing, and independent parallel tool results.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.161 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.161

    └──▷ TRY IT
    Slice Claude Code usage metrics by team or repo in your OTEL backend without any code changes — just set resource attributes before launching.
    copy

    $ OTEL_RESOURCE_ATTRIBUTES="team=security,repo=infra-scanner" claude <prompt>

    • Adds OTEL_RESOURCE_ATTRIBUTES values as labels on metric datapoints, enabling usage metrics to be sliced by custom dimensions like team or repo.
    • Shows done/total progress before detail text in claude agents rows when work is fanned out; peek view highlights the longest-running item.
    • Collapses never-signed-in claude.ai connectors in /mcp behind a Show unused connectors row to reduce noise.
    • Parallel tool calls now return independent results — a failed Bash command no longer cancels sibling calls in the same batch.

    4 details in Everything

77. v2.1.160 Jun 2, 2026 · release notes

    Claude Code v2.1.160 adds shell-startup write prompts, safer acceptEdits mode, and a grep-satisfies-read shortcut.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.160 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.160

    • Adds a confirmation prompt before writing to shell startup files (.zshenv, .zlogin, .bash_login) and ~/.config/git/ to prevent unintended command execution.
    • In acceptEdits mode, prompts before writing build-tool config files that grant code execution (.npmrc, .yarnrc*, bunfig.toml, .bazelrc, .pre-commit-config.yaml, .devcontainer/, etc.).
    • Single-file grep/egrep/fgrep commands now satisfy the read-before-edit check, eliminating the need for a separate Read step after grepping a file.
    • Renames the dynamic-workflow trigger keyword from workflow to ultracode; the keyword is highlighted in violet in the prompt input.

    4 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE environment variable is removed and is now a no-op.
    • !The dynamic-workflow trigger keyword is renamed from workflow to ultracode; the word workflow no longer triggers a dynamic workflow run.
78. v2.1.157 May 29, 2026 · release notes

    Claude Code v2.1.157 adds auto-loaded plugins, plugin scaffolding, richer telemetry, and smarter agents dispatch.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.157 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.157

    └──▷ TRY IT
    Bootstrap a new plugin without touching any marketplace — scaffold it locally and it loads automatically on next launch.
    copy

    $ claude plugin init my-recon-tool

    Capture detailed tool parameters (bash commands, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface./skill names) in OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. traces for audit or pipeline debugging.
    copy

    $ OTEL_LOG_TOOL_DETAILS=1 claude --print 'Run nmap on 10.0.0.0/24 and summarize open ports'

    • Adds automatic loading of plugins from .claude/skills directories — no marketplace required.
    • New claude plugin init <name> command scaffolds a new plugin directly into .claude/skills.
    • Adds autocomplete for /plugin arguments: subcommands, installed plugin names, and plugins from known marketplaces.
    • Honors the agent field in settings.json for dispatched sessions, with --agent <name> to override per-run.
    • EnterWorktree can now switch between Claude-managed worktrees mid-session.

    +5 more show less

    • tool_decision telemetry events now include tool_parameters (bash commands, MCP/skill names) when OTEL_LOG_TOOL_DETAILS=1 is set.
    • Claude-managed worktrees are left unlocked after an agent finishes, enabling clean git worktree remove/prune workflows.
    • Adds a Workflow keyword trigger setting in /config to prevent the word workflow in a prompt from triggering a dynamic workflow.
    • Slash-command autocomplete in the claude agents dispatch input now matches substrings.
    • Pressing backspace immediately after a workflow trigger keyword dismisses the workflow request instead of deleting a character.

    10 details in Everything

79. v2.1.154 May 28, 2026 · release notes

    Claude Code v2.1.154 adds Opus 4.8 with fast mode, dynamic multi-agent workflows, shell background sessions, and smarter plugin discovery.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.154 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.154

    └──▷ TRY IT
    Hand off a large, multi-step research or coding task to a fleet of background agents and check their progress.
    copy

    $ claude
    # Inside the session:
    > Create a workflow to audit every Go package in this repo for security issues
    > /workflows

    Run a long-lived shell process as a detachable background session inside the agents view.
    copy

    $ claude agents
    # Inside agents view:
    ! npm run build:watch

    Run a background shell command directly from the CLI without entering the interactive UI.
    copy

    $ claude --bg --exec 'pytest tests/ -x -q'

    • Adds Claude Opus 4.8 as the new default model with /effort xhigh for maximum effort on hard tasks.
    • Introduces dynamic workflows: ask Claude to orchestrate work across tens to hundreds of background agents; view runs with /workflows.
    • Fast mode on Opus 4.8 now available at 2× the standard rate for 2.5× the speed.
    • The lean system prompt is now the default for all models except Haiku, Sonnet, and Opus 4.7 and earlier.
    • Reduces unsolicited multiple-choice prompts: Claude now reserves them for decisions it genuinely cannot make itself.

    +10 more show less

    • claude agents: type ! <command> to run a shell command as a background session you can attach to and detach from; also available as claude --bg --exec '<command>'.
    • claude agents: /logout now signs you out instead of routing to a background session.
    • The ←← agents view shortcut now works on Bedrock, Vertex, Foundry, and with telemetry disabled.
    • Claude in Chrome: choose which connected browser to use via /chrome → Select browser… or inline during a browser action with multiple browsers connected.
    • Plugins can declare defaultEnabled: false in plugin.json or a marketplace entry; enable them with /plugin or claude plugin enable. Dependencies of enabled plugins are still auto-enabled.
    • The /plugin Discover tab now pins contextually relevant plugins with a "suggested for this directory" annotation.
    • Streaming tool execution is now always on, including when telemetry is disabled or on Bedrock/Vertex/Foundry (previously behind a feature flag).
    • Stdio MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server subprocesses now receive CLAUDE_CODE_SESSION_ID and CLAUDECODE=1 environment variables.
    • claude mcp list/get now show unapproved .mcp.json servers as ⏸ Pending approval when output is piped, instead of auto-approving and connecting.
    • Improves auto-mode classifier detection of data exfiltrationdata exfiltrationMoving data out of a system or organization without authorization. It can happen through network transfers, removable media or other channels., particularly bulk transfers of repository contents.

    15 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE is deprecated and will be removed on 06/01; switch to /model claude-opus-4-6[1m] then /fast on to use fast mode on Opus 4.6.
    • !The lean system prompt is now the default for all models except Haiku, Sonnet, and Opus 4.7 and earlier, which may change Claude's behavior on upgrade for affected model users.
    • !/simplify no longer runs the full /code-review --fix bug-hunting review; it now performs a cleanup-only pass.
80. v2.1.153 May 28, 2026 · release notes

    Claude Code v2.1.153 adds LFS-skip for marketplace sources, richer agent autocomplete, terminal-aware status scripts, and persistent macOS permissions.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.153 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.153

    └──▷ USE IT
    Skip downloading large LFS objects when adding a heavy model or dataset repo as a marketplace source.
    yaml copy

    # In your Claude Code marketplace config:
    github:
      repo: owner/heavy-dataset-repo
      skipLfs: true

    Size a custom status-line script to the actual terminal width without hardcoding column counts.
    copy

    $ # status_line.sh — now receives COLUMNS and LINES automatically
    printf '%*s' "$COLUMNS" '' | tr ' ' '-'

    • Adds skipLfs option to github/git plugin marketplace sources to skip Git LFS downloads during clone and update.
    • Adds COLUMNS and LINES environment variables to status line commands so scripts can size output to the terminal width.
    • Expands claude agents dispatch autocomplete to suggest native slash commands and bundled skills, not just project skills.
    • Updates claude agents PR column to display PR #N for a single PR or N PRs for multiple.
    • Merges separate needs authentication startup notices for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers and connectors into a single message.

    +3 more show less

    • macOS background agents now appear as Claude Code in Privacy & Security and retain permission grants across upgrades.
    • New /model picker saves your model selection as the default for new sessions; press s to switch for the current session only.
    • Moving /bg while Claude is responding now continues the response in the background session instead of dropping it.

    8 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The modelPicker:setAsDefault keybinding is renamed to modelPicker:thisSessionOnly in keybindings.json; the previous d action is replaced by s.
81. v2.1.152 May 27, 2026 · release notes

    Claude Code v2.1.152 adds auto-fix code review, hook-driven skill reloading, a MessageDisplay hook, and live thinking indicators.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.152 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.152

    └──▷ TRY IT
    Run a code review and immediately apply all suggestions to your working tree in one step.
    copy

    $ claude /code-review --fix

    Restrict which tools are available when a skill is active — useful for locked-down review or planning skills.
    yaml copy

    # In a skill's frontmatter (e.g., .claude/skills/review.md)
    ---
    disallowed-tools:
      - Bash
      - Edit
    ---

    Reload skills installed by a SessionStart hook without restarting the session.
    copy

    $ claude /reload-skills

    • Adds --fix flag to /code-review: automatically applies review findings (reuse, simplification, efficiency suggestions) to the working tree after review; /simplify now invokes /code-review --fix.
    • Skills and slash commands can now set disallowed-tools in frontmatter to restrict which tools the model can use while that skill is active.
    • Adds /reload-skills command to re-scan skill directories mid-session without restarting.
    • SessionStart hooks can now return reloadSkills: true to make newly installed skills available in the same session.
    • SessionStart hooks can now set the session title via hookSpecificOutput.sessionTitle on startup and resume.

    +11 more show less

    • Adds MessageDisplay hook event, enabling hooks to transform or hide assistant message text as it is rendered.
    • Adds pluginSuggestionMarketplaces managed setting, allowing admins to allowlist org marketplaces whose plugins may be suggested via context-aware tips.
    • claude plugin marketplace remove now accepts --scope user|project|local for symmetry with marketplace add, install, and uninstall.
    • Automatically switches to the configured --fallback-model for the rest of the session when the primary model is not found, instead of failing every request.
    • Auto mode no longer requires opt-in consent.
    • Vim mode: / in NORMAL mode now opens reverse history search (like Ctrl+R), matching bash/zsh vi-mode behavior.
    • The /usage breakdown now includes large session files, scanned with a streaming read to keep memory usage flat.
    • Thinking summaries in collapsed groups now render as markdown, stay readable for at least 3 seconds, and cap at 10 lines; Ctrl+O shows the full thinking.
    • In fullscreen mode, the Thinking for Ns indicator now counts up live while the model thinks and retains its value if interrupted mid-thought.
    • Post-response timer now shows 'Waiting for N background agents/workflows to finish' and reports cumulative time once results are processed.
    • Adds session entrypoint as an OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. metric attribute (app.entrypoint), opt-in via OTEL_METRICS_INCLUDE_ENTRYPOINT=true.

    16 details in Everything

82. v2.1.149 May 22, 2026 · release notes

    Claude Code v2.1.149 adds per-category usage breakdowns, keyboard-scrollable diffs, GFM task list rendering, and enterprise MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. connector controls.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.149 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.149

    └──▷ TRY IT
    Check which categories (skills, subagents, plugins, MCP servers) are eating into your usage limits before hitting a cap.
    copy

    $ /usage

    Allow all claude.ai cloud MCP connectors in an enterprise managed deployment, alongside your existing managed-mcp.json config.
    json copy

    { "allowAllClaudeAiMcps": true }

    • Adds per-category breakdown in /usage showing what's consuming your limits — skills, subagents, plugins, and per-MCP-serverMCP-serverA process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface. cost.
    • Enables keyboard scrolling in /diff detail view via arrows, j/k, PgUp/PgDn, Space, and Home/End.
    • Renders GFM task list checkboxes (- [ ] todo / - [x] done) in Markdown output instead of plain bullets.
    • New enterprise managed setting allowAllClaudeAiMcps loads claude.ai cloud MCP connectors alongside managed-mcp.json.
    • Improves /feedback reports to include conversation context from before compaction, aiding triage of issues in long sessions.

    5 details in Everything

83. v2.1.147 May 21, 2026 · release notes

    Claude Code v2.1.147 adds a new /code-review command with inline GitHub PR comments and pinned background session persistence.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.147 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.147

    └──▷ TRY IT
    Run a high-effort correctness review on your current code and post findings as inline comments on an open GitHub PR.
    copy

    $ /code-review high --comment

    • New /code-review command (replaces /simplify) reports correctness bugs at a chosen effort level and can post findings as inline GitHub PR comments via --comment.
    • Pinned background sessions (Ctrl+T in claude agents) now stay alive when idle, restart in place to apply updates, and are only shed under memory pressure after non-pinned sessions.

    2 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The /simplify command is renamed to /code-review and its cleanup-and-fix behavior has been removed; scripts or workflows invoking /simplify will no longer work.
84. v2.1.146 May 21, 2026 · release notes

    Claude Code v2.1.146 renames /simplify to /code-review with optional effort levels.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.146 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.146

    └──▷ TRY IT
    Run a high-effort code review on your current working changes to catch deeper issues before merging.
    copy

    $ /code-review high

    • New /code-review command (replaces /simplify) with an optional effort level parameter (e.g. /code-review high).
    • Auto mode now respects AskUserQuestion when the user or a skill explicitly relies on it, enabling interactive flows in automated contexts.

    2 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The /simplify command has been renamed to /code-review; any scripts, docs, or habits referencing /simplify will need to be updated.
85. v2.1.145 May 19, 2026 · release notes

    Claude Code v2.1.145 adds JSON agent listing, richer OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. spans, plugin previews, and mouse support in fullscreen.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.145 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.145

    └──▷ TRY IT
    Pipe live agent session data into a status bar or session picker script.
    copy

    $ claude agents --json | jq '.[] | select(.status == "awaiting_input") | .agent_id'

    • New claude agents --json command lists live Claude sessions as JSON for scripting (tmux-resurrect, status bars, session pickers).
    • Adds agent_id and parent_agent_id attributes to claude_code.tool OTEL spans, with corrected trace parenting so subagent spans nest under their dispatching Agent tool span.
    • Status line JSON input now includes GitHub repo and PR information when detected.
    • /plugin Discover and Browse screens now preview a plugin's commands, agents, skills, hooks, and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface./LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code. servers before installation.
    • claude agents terminal tab title now shows the awaiting-input count so an alt-tabbed window indicates when an agent needs attention.

    +2 more show less

    • Slash command and @-mention suggestion list now supports mouse hover and click in fullscreen mode.
    • Read tool now returns a truncated first page with a PARTIAL view notice instead of a hard error when a whole-file read exceeds the token limit.

    7 details in Everything

86. v2.1.144 May 19, 2026 · release notes

    Claude Code v2.1.144 adds /resume for background sessions, elapsed-time notifications, plugin last-updated stamps, and per-session model switching.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.144 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.144

    └──▷ TRY IT
    Switch to a different model mid-session without changing your default, so tomorrow's new sessions still open with your preferred model.
    copy

    $ /model   # then select the desired model; press 'd' to set a new default for future sessions only if wanted

    • Adds /resume support for background sessions — sessions started via claude --bg or agent view now appear alongside interactive ones, marked with bg.
    • Adds elapsed duration to background subagent completion notifications (e.g. Agent completed · 3h 2m 5s).
    • Plugin browse and discover panes now display when each plugin was last updated.
    • /model now changes the model for the current session only; press d in the model picker to set a default for all new sessions.
    • Adds claude --bg --name <label> echoing the session name in the post-spawn confirmation message.

    +8 more show less

    • Adds claude agents Ctrl+R session rename that immediately updates the attached session's banner.
    • /doctor now shows an exec-form example when a command hook is missing the command field.
    • Skill-listing truncation moved out of startup notifications — full breakdown now available via /doctor.
    • Improved SDK/headless MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. startup: pre-wait now overlaps with startup instead of blocking before the first turn (up to 2 s faster with slow MCP servers).
    • Improved recovery from rare pre-response stream stalls — now retries streaming once instead of falling back to a slower non-streaming request.
    • claude agents / --bg rejection messages now name the specific gate (non-TTY, env var, or setting) instead of a generic message.
    • Plugin marketplace add/update now respects the CLAUDE_CODE_PLUGIN_PREFER_HTTPS environment variable.
    • /plugin now returns to the Installed list after enabling, disabling, or uninstalling a plugin.

    13 details in Everything

87. v2.1.143 May 15, 2026 · release notes

    Claude Code v2.1.143 adds plugin dependency enforcement, projected context costs, and richer claude agents session control.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.143 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.143

    └──▷ TRY IT
    Launch the agents dashboard scoped to a specific model, effort level, and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. config so every dispatched session inherits those defaults.
    copy

    $ claude agents --model claude-opus-4-5 --effort high --mcp-config ./mcp.json

    Allow background sessions to edit the working copy directly when git worktrees are impractical in your repo.
    json copy

    # In settings.json
    { "worktree": { "bgIsolation": "none" } }

    • Adds plugin dependency enforcement: claude plugin disable refuses when another enabled plugin depends on the target and shows a copy-pasteable disable-chain hint; claude plugin enable force-enables transitive dependencies.
    • Adds projected context cost (per-turn and per-invocation token estimates) to the /plugin marketplace browse pane.
    • New worktree.bgIsolation: "none" setting lets background sessions edit the working copy directly without EnterWorktree, for repos where worktrees are impractical.
    • PowerShell tool now passes -ExecutionPolicy Bypass by default; opt out with CLAUDE_CODE_POWERSHELL_RESPECT_EXECUTION_POLICY=1.
    • PowerShell tool is now enabled by default on Windows for Bedrock, Vertex, and Foundry users; opt out with CLAUDE_CODE_USE_POWERSHELL_TOOL=0.

    +8 more show less

    • claude agents now accepts --add-dir, --settings, --mcp-config, --plugin-dir, --permission-mode, --model, --effort, and --dangerously-skip-permissions to configure the dashboard and sessions dispatched from it.
    • claude --bg --dangerously-skip-permissions now persists across retire→wake cycles.
    • /bg now preserves --mcp-config, --settings, --add-dir, --plugin-dir, and --strict-mcp-config, keeping MCP servers and settings across respawn.
    • /bg and ←-detach now preserve --fallback-model, so backgrounded workers degrade gracefully on overload.
    • /bg and ←-detach now preserve --allow-dangerously-skip-permissions, keeping bypass-permissions available in the Shift+Tab cycle.
    • Background sessions now preserve the model and effort level set after waking from idle.
    • Shift+Tab in attached agent sessions now includes auto mode in the permission cycle.
    • Background sessions launched from claude agents now honor permissions.defaultMode from settings.json.

    13 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Worktree cleanup no longer falls back to rm -rf when git worktree remove fails, preventing accidental deletion of gitignored or in-progress files — workflows that relied on that fallback cleanup behavior will need an alternative removal step.
88. v2.1.142 May 14, 2026 · release notes

    Claude Code v2.1.142 expands claude agents configuration flags and upgrades Fast mode to Opus 4.7 by default.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.142 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.142

    └──▷ TRY IT
    Pin fast mode to Opus 4.6 to keep consistent behavior across a team before adopting the new default.
    copy

    $ CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1 claude --fast

    • Adds --add-dir, --settings, --mcp-config, --plugin-dir, --permission-mode, --model, --effort, and --dangerously-skip-permissions flags to claude agents for fine-grained control over dispatched background sessions.
    • Fast mode now defaults to Opus 4.7 (up from 4.6); set CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1 to pin back to 4.6.
    • Plugins with a root-level SKILL.md and no skills/ subdirectory are now automatically surfaced as a skill.
    • The /plugin details pane and claude plugin details command now display LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code. servers provided by a plugin.
    • /web-setup now warns before overwriting an existing GitHub App connection.

    5 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Fast mode now uses Opus 4.7 by default instead of Opus 4.6; workloads sensitive to model changes must set CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1 to preserve previous behavior.
89. v2.1.141 May 13, 2026 · release notes

    Claude Code v2.1.141 adds terminalSequence hooks, HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. plugin cloning, workspace identity scoping, and richer agent/rewind controls.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.141 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.141

    └──▷ TRY IT
    Trigger a desktop notification from a hook after a long-running task completes, without needing a controlling terminal.
    copy

    $ # In your hook script, emit a bell via terminalSequence:
    echo '{"terminalSequence": "\a"}'

    Install a Claude Code plugin in a CI/CD environment that lacks a GitHub SSH key.
    copy

    $ CLAUDE_CODE_PLUGIN_PREFER_HTTPS=1 claude plugin install <plugin-name>

    List only agents scoped to the current project directory, filtering out unrelated sessions.
    copy

    $ claude agents --cwd /path/to/project

    • Adds terminalSequence field to hook JSON output, enabling hooks to emit desktop notifications, window titles, and bells without a controlling terminal.
    • Adds CLAUDE_CODE_PLUGIN_PREFER_HTTPS env var to clone GitHub plugin sources over HTTPS instead of SSH, for environments without a GitHub SSH key.
    • Adds ANTHROPIC_WORKSPACE_ID env var for workload identityworkload identityAn identity assigned to software, such as a service or container, rather than a person. It lets the workload authenticate and receive permissions without sharing a user's credentials. federation, scoping minted tokens to a specific workspace when a federation rule covers more than one.
    • Adds claude agents --cwd <path> to scope the session list to a specific directory.
    • Extends /feedback to optionally include sessions from the last 24 hours or 7 days, covering issues that span multiple sessions.

    +5 more show less

    • Adds Summarize up to here option in the Rewind menu to compress earlier context while keeping recent turns intact.
    • Background agents launched via /bg or ←← now preserve the current permission mode instead of reverting to default.
    • Auto mode permission dialog now explains when a permissions.ask rule triggered the prompt.
    • Spinner now warms to amber after 10 seconds of thinking to signal Claude is still working.
    • Improves plugin menu navigation: →/Tab switch tabs, ↑ moves to the tab strip, and tab headers and search box are clickable in fullscreen mode.

    10 details in Everything

90. v2.1.139 May 11, 2026 · release notes

    Claude Code v2.1.139 adds agent view, /goal command, hook exec form, and subagent OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. tracing.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.139 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.139

    └──▷ TRY IT
    Review all active and completed Claude Code sessions across projects from a single pane.
    copy

    $ claude agents

    Let Claude autonomously iterate on a task until a specific condition is met, tracking progress live.
    copy

    $ /goal All unit tests pass and coverage is above 80%

    • Adds claude agents command (Research Preview): a unified list of all Claude Code sessions — running, blocked, or complete.
    • Adds /goal command: set a completion condition and Claude works autonomously across turns until it's met, with a live elapsed/turns/tokens overlay panel.
    • Adds /scroll-speed command to tune mouse wheel scroll speed with a live preview.
    • Adds claude plugin details <name> to inspect a plugin's component inventory and projected per-session token cost.
    • Adds transcript view navigation shortcuts: ? for keyboard shortcuts, {/} to jump between user prompts, v to toggle the shortcut panel.

    +12 more show less

    • Adds args: string[] field (exec form) to hooks, spawning the command directly without a shell so path placeholders never need quoting.
    • Adds continueOnBlock config option for PostToolUse hooks — when true, feeds the hook's rejection reason back to Claude and continues the turn instead of blocking.
    • MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. stdio servers now receive CLAUDE_PROJECT_DIR in their environment; plugin configs can reference ${CLAUDE_PROJECT_DIR} in commands.
    • Compaction prompt now instructs the model to preserve sensitive user instructions across compaction.
    • /mcp Reconnect now picks up .mcp.json edits without a full restart and surfaces the HTTP status and URL on reconnect failure.
    • API requests from subagents now carry x-claude-code-agent-id and x-claude-code-parent-agent-id headers; claude_code.llm_request OTEL spans include agent_id and parent_agent_id attributes.
    • Remote MCP server reconnect retry on transient failures is now enabled for all users.
    • /context all per-skill token estimates now account for the model's tokenizer and display rounded values.
    • claude plugin install <name>@<marketplace> now auto-refreshes the marketplace and retries before reporting a plugin as not found.
    • /context now shows the providing plugin's name for plugin-sourced skills.
    • Remote Control, /schedule, claude.ai MCP connectors, and notification preferences are now disabled when ANTHROPIC_API_KEY / apiKeyHelper / ANTHROPIC_AUTH_TOKEN is set, even if a Claude.ai login also exists.
    • [VSCode] Press Cmd/Ctrl+Shift+T to reopen the most recently closed session tab, configurable via claudeCode.enableReopenClosedSessionShortcut.

    17 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Remote Control, /schedule, claude.ai MCP connectors, and notification preferences are disabled when ANTHROPIC_API_KEY, apiKeyHelper, or ANTHROPIC_AUTH_TOKEN is set — even if a Claude.ai login also exists. Unset the API key to re-enable these features.
91. v2.1.136 May 8, 2026 · release notes

    Claude Code v2.1.136 adds hard-deny auto mode classifier rules and an OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. survey toggle for enterprises.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.136 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.136

    └──▷ USE IT
    Enforce a hard block on a category of auto mode actions that no allow rule or user intent can override.
    json copy

    {
      "autoMode": {
        "hard_deny": ["delete_files", "network_requests"]
      }
    }

    Re-enable the session quality survey when your enterprise pipelines session telemetry through OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor..
    copy

    $ CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL=1 claude

    • Adds CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL env var to re-enable the session quality survey for enterprises capturing responses via OpenTelemetry.
    • Adds settings.autoMode.hard_deny for auto mode classifier rules that block requests unconditionally, regardless of user intent or allow exceptions.
    • WSL2WSL2A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM.: image paste from Windows clipboard now works via a PowerShell fallback when xclip/wl-paste cannot read image data.

    3 details in Everything

92. v2.1.133 May 7, 2026 · release notes

    Claude Code v2.1.133 adds worktree base-ref control, custom sandbox binary paths, admin policy merge, and effort-level hook inputs.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.133 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.133

    └──▷ TRY IT
    Read the active effort level inside a hook script to adjust external tooling (e.g. skip heavy analysis on low-effort runs).
    copy

    $ #!/bin/bash
    # In a Claude hook script
    echo "Current effort: $CLAUDE_EFFORT"
    if [ "$CLAUDE_EFFORT" = "low" ]; then
      exit 0  # skip expensive checks
    fi

    • Adds worktree.baseRef setting (fresh | head) to control whether new worktrees branch from origin/<default> or local HEAD.
    • Adds sandbox.bwrapPath and sandbox.socatPath managed settings (Linux/WSLWSLWindows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host.) to specify custom bubblewrap and socat binary locations.
    • Adds parentSettingsBehavior admin-tier key ('first-wins' | 'merge') to opt SDK managedSettings into the policy merge chain.
    • Hooks now receive the active effort level via the effort.level JSON input field and the $CLAUDE_EFFORT environment variable, readable by Bash tool commands.

    4 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The default worktree.baseRef is fresh, which changes EnterWorktree's base back to origin/<default>; since v2.1.128 it has been local HEAD. Set worktree.baseRef: "head" to preserve unpushed commits in new worktrees.
93. v2.1.132 May 6, 2026 · release notes

    Claude Code v2.1.132 adds session ID env var in subprocesses and an opt-out for the fullscreen alternate-screen renderer.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.132 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.132

    └──▷ TRY IT
    Keep Claude Code output in your terminal's native scrollback instead of the fullscreen renderer — useful in tmux, screen, or any workflow where you rely on scroll history.
    copy

    $ CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN=1 claude

    Reference the session ID inside a Bash tool script to tag logs or correlate subprocess events back to the originating Claude Code session.
    copy

    $ echo "Running inside Claude Code session: $CLAUDE_CODE_SESSION_ID" >> /tmp/session-audit.log

    • Adds CLAUDE_CODE_SESSION_ID environment variable to Bash tool subprocess environments, matching the session_id passed to hooks — enabling scripts and hooks to correlate subprocess activity back to the parent session.
    • Adds CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN=1 env var to opt out of the fullscreen alternate-screen renderer, keeping conversation output in the terminal's native scrollback.
    • Adds a Pasting… footer hint while a Ctrl+V image paste is being read from the clipboard.

    3 details in Everything

94. v2.1.129 May 6, 2026 · release notes

    Claude Code v2.1.129 adds URL-based plugin loading, auto-update on Homebrew/WinGet, and cross-project history search by default.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.129 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.129

    └──▷ TRY IT
    Load a community or internal plugin directly from a URL without installing it permanently.
    copy

    $ claude --plugin-url https://example.com/my-plugin.zip

    Enable background auto-updates on a Homebrew install so Claude Code silently upgrades and prompts to restart.
    copy

    $ export CLAUDE_CODE_PACKAGE_MANAGER_AUTO_UPDATE=1
    claude

    Force synchronized output when running Claude Code inside Emacs eat or another terminal where auto-detection fails.
    copy

    $ CLAUDE_CODE_FORCE_SYNC_OUTPUT=1 claude

    • Adds --plugin-url <url> flag to fetch and load a plugin .zip archive from a URL for the current session.
    • Adds CLAUDE_CODE_FORCE_SYNC_OUTPUT=1 env var to force synchronized output on terminals where auto-detection fails (e.g. Emacs eat).
    • Adds CLAUDE_CODE_PACKAGE_MANAGER_AUTO_UPDATE env var: when set on Homebrew or WinGet installs, Claude Code upgrades itself in the background and prompts to restart.
    • Adds CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1 env var to opt in to Gateway /v1/models discovery for the /model picker (now opt-in instead of automatic).
    • Ctrl+R history picker now searches all prompts across all projects by default; press Ctrl+S to narrow to the current project or session.

    +4 more show less

    • skillOverrides setting is now functional: off hides a skill from the model and /, user-invocable-only hides from the model only, name-only collapses the description.
    • The claude_code.pull_request.count OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. metric now counts PRs/MRs created via MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools in addition to shell commands.
    • Policy refusal error messages now include the API Request ID to aid support debugging.
    • Plugin manifests: themes and monitors should now be declared under "experimental": { ... }; claude plugin validate warns on top-level declarations.

    9 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Gateway /v1/models model discovery (automatic in v2.1.126–2.1.128) is now opt-in via CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1; setups relying on automatic gateway model discovery will no longer see those models without setting this variable.
95. v2.1.128 May 4, 2026 · release notes

    Claude Code v2.1.128 adds ZIP plugin support, channels auth expansion, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server diagnostics, and smarter /mcp and /color UX.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.128 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.128

    └──▷ TRY IT
    Install a plugin distributed as a ZIP archive without unpacking it first.
    copy

    $ claude --plugin-dir ./my-plugin.zip

    Audit connected MCP servers to spot any that registered zero tools (possible misconfiguration).
    copy

    $ /mcp

    • Bare /color with no arguments now picks a random session color.
    • Adds tool-count display to /mcp output; servers that connect with zero tools are flagged.
    • --plugin-dir now accepts .zip plugin archives in addition to directories.
    • --channels now works with console (API key) authentication; console orgs with managed settings must set channelsEnabled: true to enable.
    • Updated /model picker collapses duplicate Opus 4.7 entries and labels the current Opus model simply as Opus.

    +5 more show less

    • Subprocesses (Bash, hooks, MCP, LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code.) no longer inherit OTEL_* environment variables, isolating OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor.-instrumented apps from the CLI's own OTLPOTLPOpenTelemetry Protocol, a standard way to send traces, metrics and logs between applications, collectors and monitoring systems. It lets telemetry move across compatible tools. endpoint.
    • Reconnecting MCP servers now summarize re-announced tools by server prefix instead of flooding the conversation with full tool-name lists.
    • SDK hosts now receive a persistent localSettings suggestion for Bash permission prompts, so Always allow writes to .claude/settings.local.json.
    • Auto mode classifier errors now include a hint (retry, /compact, or run with --debug) when an action cannot be evaluated.
    • Headless --output-format stream-json: init.plugin_errors now includes --plugin-dir load failures in addition to dependency demotions.

    10 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !workspace is now a reserved MCP server name — existing servers configured with that name will be skipped with a warning.
    • !Subprocesses (Bash, hooks, MCP, LSP) no longer inherit OTEL_* environment variables; any setup relying on subprocess inheritance of the CLI's OTEL config will stop working.
    • !EnterWorktree now creates the new branch from local HEAD instead of origin/<default-branch>; workflows that depended on the previous behavior of branching from the remote default may behave differently.
96. v2.1.126 May 1, 2026 · release notes

    Claude Code v2.1.126 adds project purge, gateway model picker, headless OAuth, and richer OTelOTelAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. events.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.126 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.126

    └──▷ TRY IT
    Clean up all Claude Code state for a stale or sensitive project before archiving or handing off a repo — preview what will be deleted first.
    copy

    $ claude project purge /path/to/project --dry-run

    Non-interactively purge all project state (transcripts, tasks, file history, config entry) in a CI or scripted cleanup flow.
    copy

    $ claude project purge /path/to/project -y

    • Adds claude project purge [path] command to delete all Claude Code state for a project (transcripts, tasks, file history, config entry) with --dry-run, -y/--yes, -i/--interactive, and --all flags.
    • The /model picker now lists models from a gateway's /v1/models endpoint when ANTHROPIC_BASE_URL points at an Anthropic-compatible gateway.
    • --dangerously-skip-permissions now bypasses write prompts for .claude/, .git/, .vscode/, shell config files, and other previously-protected paths (catastrophic removal commands still prompt).
    • claude auth login now accepts an OAuth code pasted directly into the terminal when the browser callback can't reach localhost (WSL2WSL2A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM., SSH, containers).
    • The claude_code.skill_activated OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. event now fires for user-typed slash commands and carries a new invocation_trigger attribute ("user-slash", "claude-proactive", or "nested-skill").

    +2 more show less

    • Windows: Claude now treats PowerShell as the primary shell (instead of Bash) when the PowerShell tool is enabled.
    • Windows: PowerShell 7 installed via the Microsoft Store, MSI without PATH, or .NET global tool is now detected automatically.

    7 details in Everything

97. v2.1.122 Apr 28, 2026 · release notes

    Claude Code v2.1.122 adds Bedrock service tier selection, PR-URL session lookup, and new OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. events.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.122 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.122

    └──▷ TRY IT
    Route Bedrock API calls to a specific service tier (e.g., priority) for latency-sensitive workloads.
    copy

    $ ANTHROPIC_BEDROCK_SERVICE_TIER=priority claude --model anthropic.claude-opus-4-5

    • Adds ANTHROPIC_BEDROCK_SERVICE_TIER env var to select a Bedrock service tier (default, flex, or priority), sent as the X-Amzn-Bedrock-Service-Tier header.
    • Pasting a GitHub, GitHub Enterprise, GitLab, or Bitbucket PR URL into the /resume search box now finds the session that created that PR.
    • Adds claude_code.at_mention OpenTelemetry log event for @-mention resolution.
    • OpenTelemetry numeric attributes on api_request/api_error log events are now emitted as numbers instead of strings.
    • The /mcp view now surfaces claude.ai connectors hidden by a manually-added server with the same URL, with a hint to remove the duplicate.

    5 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !OpenTelemetry consumers parsing numeric attributes on api_request/api_error log events as strings will receive numbers instead — pipelines that treat those fields as strings may need to be updated.
98. v2.1.121 Apr 28, 2026 · release notes

    Claude Code v2.1.121 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. alwaysLoad, plugin pruning, PostToolUse output replacement, mTLSmTLSMutual TLS, an encrypted connection where both client and server present certificates to prove their identity. It helps services verify who they are communicating with. Workload Identity, and expanded OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. spans.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.121 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.121

    └──▷ USE IT
    Pin a high-priority MCP server's tools so they are always available without waiting for tool-search deferral — useful for servers you rely on in every session.
    json copy

    # In your MCP server config (e.g. .claude/mcp_settings.json)
    {
      "mcpServers": {
        "my-server": {
          "command": "my-mcp-server",
          "alwaysLoad": true
        }
      }
    }

    Clean up orphaned plugin dependencies after removing plugins, or cascade-remove a plugin and all its unique dependencies in one step.
    copy

    $ # Prune all orphaned dependencies
    claude plugin prune
    
    # Or remove a plugin and cascade its dependencies
    claude plugin uninstall my-plugin --prune

    • Adds alwaysLoad option to MCP server config — when true, all tools from that server bypass tool-search deferral and are always available.
    • Adds claude plugin prune command to remove orphaned auto-installed plugin dependencies; plugin uninstall --prune cascades removals.
    • PostToolUse hooks can now replace tool output for ALL tools via hookSpecificOutput.updatedToolOutput, not just MCP tools.
    • Adds type-to-filter search box to /skills for quickly finding skills in long lists.
    • Vertex AI: supports X.509X.509A standard defining the format of public-key certificates, maintained by the ITU-T, used by cyber tools to authenticate identities and establish encrypted connections over TLS and similar protocols. certificate-based Workload Identity Federation (mTLS ADC).

    +11 more show less

    • OpenTelemetry spans now include stop_reason, gen_ai.response.finish_reasons, and user_system_prompt (gated behind OTEL_LOG_USER_PROMPTS) on LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. request spans.
    • SDK: mcp_authenticate now supports redirectUri for custom scheme completion and claude.ai connectors.
    • Enables CLAUDE_CODE_FORK_SUBAGENT=1 in non-interactive sessions for SDK and claude -p.
    • --dangerously-skip-permissions no longer prompts for writes to .claude/skills/, .claude/agents/, and .claude/commands/.
    • MCP servers that hit a transient error at startup now auto-retry up to 3 times instead of staying disconnected.
    • Terminal tab session title is now generated in the configured language setting.
    • LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code. diagnostic summaries now expand on click or Ctrl+O and show an expand hint.
    • Scrollable overflow dialogs: terminal dialogs that exceed terminal height are now navigable with arrow keys, PgUp/PgDn, Home/End, and mouse wheel in both fullscreen and non-fullscreen modes.
    • Clicking any line of a long URL wrapped across rows in fullscreen mode now opens the full URL.
    • [VSCode] Voice dictation now respects the accessibility.voice.speechLanguage setting when no Claude Code language is configured.
    • [VSCode] /context now opens a native token usage dialog.

    16 details in Everything

99. v2.1.119 Apr 23, 2026 · release notes

    Claude Code v2.1.119 adds GitLab/Bitbucket PR support, persistent config, richer hooks, and OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. improvements.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1.119 https://github.com/anthropics/claude-code.git
    # already have the repo? check out this version:
    $ git checkout v2.1.119

    └──▷ TRY IT
    Load a GitLab MR or Bitbucket PR directly into a Claude Code session for review or agentic work, without converting the URL to GitHub format.
    copy

    $ claude --from-pr https://gitlab.com/myorg/myrepo/-/merge_requests/42

    Hide the working directory from the startup logo when running Claude Code in shared or recorded terminal sessions.
    copy

    $ CLAUDE_CODE_HIDE_CWD=1 claude

    • Adds prUrlTemplate setting to point the footer PR badge at a custom code-review URL instead of github.com.
    • Adds CLAUDE_CODE_HIDE_CWD environment variable to hide the working directory in the startup logo.
    • --from-pr now accepts GitLab merge-request, Bitbucket pull-request, and GitHub Enterprise PR URLs in addition to github.com.
    • /config settings (theme, editor mode, verbose, etc.) now persist to ~/.claude/settings.json and participate in project/local/policy override precedence.
    • --print mode now honors the agent's tools: and disallowedTools: frontmatter, matching interactive-mode behavior.

    +11 more show less

    • --agent <name> now honors the agent definition's permissionMode for built-in agents.
    • PowerShell tool commands can now be auto-approved in permission mode, matching Bash behavior.
    • PostToolUse and PostToolUseFailure hook inputs now include duration_ms (tool execution time, excluding permission prompts and PreToolUse hooks).
    • Subagent and SDK MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server reconfiguration now connects servers in parallel instead of serially, reducing startup latency.
    • Plugins pinned by another plugin's version constraint now auto-update to the highest satisfying git tag.
    • OpenTelemetry tool_result and tool_decision events now include tool_use_id; tool_result also includes tool_input_size_bytes.
    • Status line stdin JSON now includes effort.level and thinking.enabled fields.
    • Tool search on Vertex AI is now opt-in via ENABLE_TOOL_SEARCH environment variable.
    • Slash command suggestions now highlight the characters that matched your query.
    • Slash command picker now wraps long descriptions onto a second line instead of truncating.
    • owner/repo#N shorthand links in output now use your git remote's host instead of always pointing at github.com.

    16 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Tool search is now disabled by default on Vertex AI; opt in by setting ENABLE_TOOL_SEARCH to restore the previous behavior.
100. v2.1.118 Apr 23, 2026 · release notes

     Claude Code v2.1.118 adds vim visual mode, custom themes, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tool hooks, and a new /usage command.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.118 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.118

     └──▷ TRY IT
     Block all updates in a locked-down enterprise environment where even manual claude update must be prohibited.
     copy

     $ DISABLE_UPDATES=1 claude

     Allow a hook to call an MCP tool directly — e.g., log an event to a custom MCP server after every agent stop.
     json copy

     {
       "type": "mcp_tool",
       "server": "my-mcp-server",
       "tool": "log_event",
       "params": { "event": "agent_stop" }
     }

     Tag a plugin release with version validation before publishing.
     copy

     $ claude plugin tag v1.2.0

     • Adds vim visual mode (v) and visual-line mode (V) with selection, operators, and visual feedback in the editor.
     • Merges /cost and /stats into a unified /usage command with tabs; old names remain as shortcuts.
     • Introduces custom named themes: create/switch via /theme or hand-edit JSON in ~/.claude/themes/; plugins can ship themes via a themes/ directory.
     • Hooks can now invoke MCP tools directly using type: "mcp_tool".
     • New DISABLE_UPDATES env var completely blocks all update paths including manual claude update, stricter than DISABLE_AUTOUPDATER.

     +8 more show less

     • WSLWSLWindows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. on Windows can now inherit Windows-side managed settings via the wslInheritsWindowsSettings policy key.
     • Auto mode: include "$defaults" in autoMode.allow, autoMode.soft_deny, or autoMode.environment to extend the built-in rule list instead of replacing it.
     • Adds a Don't ask again option to the auto mode opt-in prompt.
     • New claude plugin tag subcommand creates release git tags for plugins with version validation.
     • --continue/--resume now find sessions that added the current directory via /add-dir.
     • /color now syncs the session accent color to claude.ai/code when Remote Control is connected.
     • The /model picker honors ANTHROPIC_DEFAULT_*_MODEL_NAME/_DESCRIPTION overrides when using a custom ANTHROPIC_BASE_URL gateway.
     • Plugin version-constraint skips during auto-update are now surfaced in /doctor and the /plugin Errors tab.

     13 details in Everything

101. v2.1.117 Apr 22, 2026 · release notes

     Claude Code v2.1.117 adds fork-subagent support, richer OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. attributes, and faster native search via embedded bfs/ugrep.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.117 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.117

     └──▷ TRY IT
     Surface custom/MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. slash-command names in OpenTelemetry traces for full audit visibility (they are redacted by default).
     copy

     $ OTEL_LOG_TOOL_DETAILS=1 claude

     • Enables forked subagents on external builds via CLAUDE_CODE_FORK_SUBAGENT=1.
     • Loads agent frontmatter mcpServers for main-thread agent sessions launched with --agent.
     • Persists /model selections across restarts even when a project pins a different model; startup header now indicates when the active model is from a project or managed-settings pin.
     • Adds summarization offer in /resume for stale, large sessions before re-reading them.
     • Concurrent MCP server connections now default when both local and claude.ai servers are configured, reducing startup time.

     +9 more show less

     • plugin install on an already-installed plugin now installs any missing dependencies instead of stopping.
     • Adds claude plugin marketplace add auto-resolution of missing dependencies from configured marketplaces; dependency errors now include an install hint.
     • Enforces managed-settings blockedMarketplaces and strictKnownMarketplaces on plugin install, update, refresh, and autoupdate.
     • Advisor Tool (experimental) dialog now carries an experimental label, a learn-more link, and a startup notification when enabled.
     • Extends cleanupPeriodDays retention sweep to cover ~/.claude/tasks/, ~/.claude/shell-snapshots/, and ~/.claude/backups/.
     • OpenTelemetry user_prompt events now include command_name and command_source for slash commands; cost.usage, token.usage, api_request, and api_error now include an effort attribute for models that support effort levels; custom/MCP command names are redacted unless OTEL_LOG_TOOL_DETAILS=1 is set.
     • On native macOS and Linux builds, Glob and Grep tools are replaced by embedded bfs and ugrep available through the Bash tool, enabling faster searches without a separate tool round-trip.
     • Windows builds now cache where.exe executable lookups per process for faster subprocess launches.
     • Default effort for Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 is now high (previously medium).

     14 details in Everything

102. v2.1.116 Apr 20, 2026 · release notes

     Claude Code v2.1.116 adds faster session resumption, deferred MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. startup, inline thinking progress, and smarter plugin dependency auto-install.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.116 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.116

     └──▷ TRY IT
     After adding a new plugin source, force-reload plugins so missing dependencies are pulled in automatically without manual marketplace steps.
     copy

     $ /reload-plugins

     Configure scroll sensitivity for VS Code, Cursor, or Windsurf so fullscreen sessions scroll smoothly without manual editor tweaks.
     copy

     $ /terminal-setup

     Run a main-thread agent that fires its frontmatter hooks as part of the agent lifecycle.
     copy

     $ claude --agent <path/to/agent.md>

     • Speeds up /resume on large sessions by up to 67% on sessions 40 MB and larger, with more efficient handling of dead-fork entries.
     • Defers resources/templates/list to first @-mention, reducing MCP startup time when multiple stdio servers are configured.
     • Improves fullscreen scrolling in VS Code, Cursor, and Windsurf terminals — /terminal-setup now configures scroll sensitivity for those editors.
     • Replaces the separate thinking-hint row with an inline progress indicator (still thinking, thinking more, almost done thinking) in the spinner.
     • Extends /config search to match option values (e.g. searching vim surfaces the Editor mode setting).

     +7 more show less

     • Allows /doctor to be opened while Claude is actively responding, without waiting for the current turn to finish.
     • Enables /reload-plugins and background plugin auto-update to auto-install missing plugin dependencies from already-added marketplaces.
     • Surfaces a rate-limit hint in the Bash tool when gh commands hit GitHub's API rate limit, letting agents back off instead of retrying.
     • Shows 5-hour and weekly usage immediately in the Settings Usage tab, with graceful handling when the usage endpoint is rate-limited.
     • Fires agent frontmatter hooks: when the agent is run as a main-thread agent via --agent.
     • Displays No commands match in the slash command menu when a filter has zero results instead of hiding the menu entirely.
     • Security: sandbox auto-allow no longer bypasses the dangerous-path safety check for rm/rmdir targeting /, $HOME, or other critical system directories.

     12 details in Everything

103. v2.1.113 Apr 17, 2026 · release notes

     Claude Code v2.1.113 adds domain blocklisting, exec-wrapper deny rules, and expanded Remote Control capabilities.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.113 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.113

     └──▷ USE IT
     Block a specific domain (e.g., an exfiltration endpoint) even when a wildcard allowedDomains rule would normally permit it.
     json copy

     # In your Claude Code settings (e.g., .claude/settings.json)
     {
       "sandbox": {
         "network": {
           "allowedDomains": ["*.example.com"],
           "deniedDomains": ["malicious.example.com"]
         }
       }
     }

     • New sandbox.network.deniedDomains setting blocks specific domains even when a broader allowedDomains wildcard would otherwise permit them.
     • Security: Bash deny rules now match commands wrapped in env/sudo/watch/ionice/setsid and similar exec wrappers, closing bypass paths.
     • Security: Bash(find:*) allow rules no longer auto-approve find -exec/-delete, preventing unintended file operations.
     • Security: on macOS, /private/{etc,var,tmp,home} paths are treated as dangerous removal targets under Bash(rm:*) allow rules.
     • Bash tool now shows the full command in the transcript for multi-line commands whose first line is a comment, closing a UI-spoofing vector.

     +7 more show less

     • Remote Control (mobile/web) clients can now use /extra-usage and query @-file autocomplete suggestions.
     • CLI now spawns a native Claude Code binary via a per-platform optional dependency instead of bundled JavaScript.
     • Ctrl+A and Ctrl+E move to the start/end of the current logical line in multiline input, matching readline behavior.
     • Windows: Ctrl+Backspace now deletes the previous word.
     • Long URLs in responses and bash output stay clickable when they wrap across lines in terminals with OSC 8 hyperlinks.
     • Improved /ultrareview: faster launch with parallelized checks, diffstat in the launch dialog, and animated launching state.
     • Improved /loop: pressing Esc cancels pending wakeups; wakeups display as Claude resuming /loop wakeup for clarity.

     12 details in Everything

104. v2.1.111 Apr 16, 2026 · release notes

     Claude Code v2.1.111 adds /ultrareview for cloud-based PR analysis, xhigh effort tier, PowerShell support, and a permission-prompt reducer.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.111 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.111

     └──▷ TRY IT
     Review a specific GitHub PR with multi-agent cloud analysis to get comprehensive feedback before merging.
     copy

     $ /ultrareview 4821

     Reduce permission prompts in a project by auto-generating an allowlist from your session history.
     copy

     $ /less-permission-prompts

     Emit full API request and response bodies as OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. log events to trace model calls during debugging.
     copy

     $ OTEL_LOG_RAW_API_BODIES=1 claude --output-format stream-json '<prompt>'

     • Adds xhigh effort level for Opus 4.7, sitting between high and max; selectable via /effort, --effort, or the model picker.
     • New /ultrareview command runs comprehensive code review in the cloud using parallel multi-agent analysis — no arguments reviews the current branch; /ultrareview <PR#> fetches and reviews a specific GitHub PR.
     • New /less-permission-prompts skill scans transcripts for common read-only Bash and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tool calls and proposes a prioritized allowlist for .claude/settings.json.
     • Auto mode is now available for Max subscribers using Opus 4.7, and no longer requires the --enable-auto-mode flag.
     • Adds PowerShell tool support on Windows (progressive rollout); opt in/out with CLAUDE_CODE_USE_POWERSHELL_TOOL; enable on Linux/macOS with CLAUDE_CODE_USE_POWERSHELL_TOOL=1 (requires pwsh on PATH).

     +9 more show less

     • Adds OTEL_LOG_RAW_API_BODIES environment variable to emit full API request and response bodies as OpenTelemetry log events for debugging.
     • New Auto (match terminal) theme option automatically matches the terminal's dark/light mode; selectable via /theme.
     • /effort now opens an interactive slider with arrow-key navigation and Enter to confirm when called without arguments.
     • Read-only bash commands with glob patterns (e.g. ls *.ts) and commands beginning with cd <project-dir> && no longer trigger a permission prompt.
     • /skills menu now supports sorting by estimated token count — press t to toggle.
     • Ctrl+L now forces a full screen redraw in addition to clearing the prompt input.
     • Transcript view footer now shows [ (dump to scrollback) and v (open in editor) shortcuts.
     • Headless --output-format stream-json now includes plugin_errors on the init event when plugins are demoted for unsatisfied dependencies.
     • 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 on re-run, and offer a 1M context option for supported models.

     14 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Ctrl+U behavior changed: it now clears the entire input buffer instead of deleting only to the start of the line (press Ctrl+Y to restore).
105. v2.1.110 Apr 15, 2026 · release notes

     Claude Code v2.1.110 adds flicker-free TUITUIA text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. fullscreen, mobile push notifications, and distributed trace linking for SDK sessions.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.110 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.110

     └──▷ TRY IT
     Switch to flicker-free fullscreen rendering mid-conversation without starting a new session.
     copy

     $ /tui fullscreen

     Link a headless Claude Code SDK run into your existing distributed trace for end-to-end observability.
     copy

     $ TRACEPARENT=00-4bf92f3577b34da6a3ce929d0e0e4736-00f067aa0ba902b7-01 TRACESTATE=vendor=value claude --headless --print "summarise logs"

     • Adds /tui command and tui setting — run /tui fullscreen to switch to flicker-free rendering within the same conversation.
     • Adds /focus command to toggle focus view independently from verbose transcript (previously both were on Ctrl+O).
     • Adds push notification tool so Claude can send mobile push notifications when Remote Control and Push when Claude decides config are enabled.
     • Adds autoScrollEnabled config to disable conversation auto-scroll in fullscreen mode.
     • Adds option to show Claude's last response as commented context in the Ctrl+G external editor, enabled via /config.

     +7 more show less

     • Improves /plugin Installed tab: items needing attention and favorites float to the top, disabled items collapse behind a fold, and f favorites the selected item.
     • Improves /doctor to warn when an MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server is defined in multiple config scopes with different endpoints.
     • --resume/--continue now resurrects unexpired scheduled tasks.
     • Enables /autocompact, /context, /exit, and /reload-plugins commands from Remote Control (mobile/web) clients.
     • Write tool now informs the model when you edit proposed content in the IDE diff before accepting.
     • SDK/headless sessions now read TRACEPARENT/TRACESTATE from the environment for distributed trace linking.
     • Session recap is now enabled for users with telemetry disabled (Bedrock, Vertex, Foundry, DISABLE_TELEMETRY); opt out via /config or CLAUDE_CODE_ENABLE_AWAY_SUMMARY=0.

     12 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Ctrl+O no longer toggles focus view — it now toggles between normal and verbose transcript only; focus view must be toggled with the new /focus command.
106. v2.1.108 Apr 14, 2026 · release notes

     Claude Code v2.1.108 adds 1-hour prompt cache TTLTTLTime to live, a limit on how long data remains valid or how far a packet can travel. Caches use a time limit; IP packets use a hop limit. opt-in, session recap, and slash-command discovery via Skill tool.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.108 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.108

     └──▷ TRY IT
     Opt into the 1-hour prompt cache TTL on Bedrock to cut repeated-context costs across long sessions.
     copy

     $ ENABLE_PROMPT_CACHING_1H=1 claude

     Re-orient yourself after stepping away from a long session without scrolling through the full transcript.
     copy

     $ /recap

     Resume work in the current repo — the picker now defaults to sessions from this directory; switch to all projects if needed.
     copy

     $ /resume   # Ctrl+A to expand to all projects

     • Adds ENABLE_PROMPT_CACHING_1H env var to opt into 1-hour prompt cache TTL across API key, Bedrock, Vertex, and Foundry backends; FORCE_PROMPT_CACHING_5M forces the 5-minute TTL.
     • Adds /recap command (and CLAUDE_CODE_ENABLE_AWAY_SUMMARY env var) to generate a context summary when returning to a session; also configurable via /config.
     • Enables the model to discover and invoke built-in slash commands (/init, /review, /security-review) via the Skill tool.
     • Adds /undo as an alias for /rewind.
     • Adds a warning at startup when prompt caching is disabled via DISABLE_PROMPT_CACHING* environment variables.

     +5 more show less

     • Adds a verbose indicator when viewing the detailed transcript via Ctrl+O.
     • Improves /resume picker to default to sessions from the current directory; Ctrl+A reveals all projects.
     • Improves /model to warn before switching models mid-conversation to flag the uncached full-history re-read cost.
     • Improves error messages: rate limits distinguished from plan usage limits; 5xx/529 errors link to status.claude.com; unknown slash commands suggest the closest match.
     • Reduces memory footprint for file reads, edits, and syntax highlighting by loading language grammars on demand.

     10 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !ENABLE_PROMPT_CACHING_1H_BEDROCK is deprecated (still honored) in favor of the unified ENABLE_PROMPT_CACHING_1H env var.
107. v2.1.105 Apr 13, 2026 · release notes

     Claude Code v2.1.105 adds PreCompact hooks, plugin background monitors, worktree path switching, and a smarter /doctor with auto-fix.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.105 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.105

     └──▷ TRY IT
     Block compaction at a critical point in a long session by returning a structured decision from a PreCompact hook.
     copy

     $ #!/usr/bin/env sh
     # .claude/hooks/pre-compact.sh
     # Exit code 2 tells Claude Code to block compaction
     exit 2

     • Adds path parameter to the EnterWorktree tool to switch into an existing worktree of the current repository.
     • Adds PreCompact hook support: hooks can block compaction by exiting with code 2 or returning {"decision":"block"}.
     • Adds background monitor support for plugins via a top-level monitors manifest key that auto-arms at session start or on skill invoke.
     • Adds /proactive as an alias for /loop.
     • Improves /doctor layout with status icons and adds f keypress to have Claude automatically fix reported issues.

     +3 more show less

     • Raises plugin skill description listing cap from 250 to 1,536 characters, with a startup warning when descriptions are truncated.
     • Improves WebFetch to strip <style> and <script> contents from fetched pages, preventing CSS-heavy pages from exhausting the content budget.
     • Improves MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. large-output truncation prompt with format-specific recipes (e.g. jq for JSON, computed Read chunk sizes for text).

     8 details in Everything

108. v2.1.101 Apr 10, 2026 · release notes

     Claude Code v2.1.101 adds /team-onboarding, OS CA store trust by default, and auto-create cloud environments for /ultraplan.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.101 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.101

     └──▷ TRY IT
     Generate a ramp-up guide for a new teammate joining your project, based on your existing Claude Code usage patterns.
     copy

     $ /team-onboarding

     Resume a named session by its human-readable title (set via /rename) rather than an opaque session ID.
     copy

     $ claude -p --resume my-project-audit

     Restrict TLS trust to bundled CAs only, opting out of the new OS CA store default (e.g., in locked-down environments where OS trust anchors should not be used).
     copy

     $ CLAUDE_CODE_CERT_STORE=bundled claude

     • Adds /team-onboarding command to generate a teammate ramp-up guide from your local Claude Code usage.
     • Trusts OS CA certificate store by default, enabling enterprise TLS proxies without extra configuration (set CLAUDE_CODE_CERT_STORE=bundled to restrict to bundled CAs only).
     • Enables /ultraplan and other remote-session features to auto-create a default cloud environment, removing the requirement to complete web setup first.
     • Beta tracing now honors OTEL_LOG_USER_PROMPTS, OTEL_LOG_TOOL_DETAILS, and OTEL_LOG_TOOL_CONTENT; sensitive span attributes are opt-in only.
     • claude -p --resume <name> now accepts session titles set via /rename or --name.

     +5 more show less

     • Rate-limit retry messages now show which limit was hit and when it resets.
     • Refusal error messages now include the API-provided explanation when available.
     • Tool-not-available errors now explain why the tool is unavailable and how to proceed.
     • /plugin and claude plugin update now warn when the marketplace could not be refreshed instead of silently showing stale versions.
     • SDK query() now cleans up subprocesses and temp files when consumers break from for await or use await using.

     10 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !OS CA certificate store is now trusted by default; set CLAUDE_CODE_CERT_STORE=bundled to revert to bundled CAs only.
109. v2.1.98 Apr 9, 2026 · release notes

     Claude Code v2.1.98 adds Vertex AI setup wizard, Perforce mode, subprocess sandboxingsandboxingRunning code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls., and a new Monitor tool for background scripts.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.98 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.98

     └──▷ TRY IT
     Protect read-only Perforce-managed files so writes fail loudly with a p4 edit hint instead of silently overwriting them.
     copy

     $ CLAUDE_CODE_PERFORCE_MODE=1 claude

     Sandbox background script subprocesses with PID namespace isolation and cap the number of script invocations per session.
     copy

     $ CLAUDE_CODE_SUBPROCESS_ENV_SCRUB=1 CLAUDE_CODE_SCRIPT_CAPS=10 claude

     Print the current prompt in a cache-friendly form that strips dynamic sections, useful for sharing stable prompt hashes across users in CI.
     copy

     $ claude --print --exclude-dynamic-system-prompt-sections

     • Adds interactive Google Vertex AI setup wizard from the login screen under 3rd-party platform, guiding through GCP auth, project/region config, credential verification, and model pinning.
     • New CLAUDE_CODE_PERFORCE_MODE env var: Edit/Write/NotebookEdit fail on read-only files with a p4 edit hint instead of silently overwriting them.
     • New Monitor tool for streaming events from background scripts.
     • Adds subprocess sandboxing with PID namespace isolation on Linux when CLAUDE_CODE_SUBPROCESS_ENV_SCRUB is set; new CLAUDE_CODE_SCRIPT_CAPS env var limits per-session script invocations.
     • New --exclude-dynamic-system-prompt-sections flag for print mode to improve cross-user prompt caching.

     +8 more show less

     • Adds workspace.git_worktree to status line JSON input, set when the current directory is inside a linked git worktree.
     • Propagates W3C TRACEPARENT env var to Bash tool subprocesses when OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. tracing is enabled, correctly parenting child-process spans in the trace tree.
     • LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code.: Claude Code now identifies itself to language servers via clientInfo in the initialize request.
     • Improved /agents with a tabbed layout: Running tab shows live subagents; Library tab adds Run agent and View running instance actions.
     • Improved /reload-plugins to pick up plugin-provided skills without requiring a restart.
     • Improved Vim mode: j/k in NORMAL mode now navigate history and select the footer pill at the input boundary.
     • Improved OTEL tracing: interaction spans now correctly wrap full turns under concurrent SDK calls, and headless turns end spans per-turn.
     • Updated /claude-api skill to cover Managed Agents alongside Claude API.

     13 details in Everything

110. v2.1.97 Apr 8, 2026 · release notes

     Claude Code v2.1.97 adds focus view toggle, refreshable status lines, git worktree context, live subagent indicators, and Cedar policy syntax highlighting.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.97 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.97

     └──▷ USE IT
     Keep a live clock or custom metric in the status line without restarting — useful for monitoring session-level stats in long-running agentic runs.
     json copy

     # In settings.json
     {
       "statusLine": {
         "command": "date '+%H:%M:%S'",
         "refreshInterval": 5
       }
     }

     Distinguish which git worktree a subagent is operating in from a shared status-line script — lets you label the session card with the worktree path.
     python copy

     # Example status-line script that reads workspace.git_worktree from stdin
     import sys, json
     data = json.load(sys.stdin)
     worktree = data.get("workspace", {}).get("git_worktree", "")
     print(f"worktree: {worktree}" if worktree else "main worktree")

     Toggle focus view during a long tool-heavy session to see just the prompt, a compact tool summary, and the final response without visual noise.
     copy

     $ # While Claude Code is running in NO_FLICKER mode, press:
     Ctrl+O

     • Adds focus view toggle (Ctrl+O) in NO_FLICKER mode showing prompt, one-line tool summary with edit diffstats, and final response
     • Adds refreshInterval status line setting to re-run the status line command every N seconds
     • Adds workspace.git_worktree field to status line JSON input, populated when inside a linked git worktree
     • Adds ● N running indicator in /agents next to agent types with live subagent instances
     • Adds syntax highlighting for Cedar policy files (.cedar, .cedarpolicy)

     +8 more show less

     • Accept Edits mode now auto-approves filesystem commands prefixed with safe env vars or process wrappers (e.g. LANG=C rm foo, timeout 5 mkdir out)
     • Auto mode and bypass-permissions mode now auto-approve sandbox network access prompts
     • sandbox.network.allowMachLookup now takes effect on macOS
     • Pasted and attached images are now compressed to the same token budget as images read via the Read tool
     • Slash command and @-mention completion now triggers after CJK sentence punctuation, enabling Japanese/Chinese input without a leading space
     • Bridge sessions now display the local git repo, branch, and working directory on the claude.ai session card
     • Bash tool OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. tracing now injects a W3C TRACEPARENT env var into subprocesses when tracing is enabled
     • Updated /claude-api skill to cover Managed Agents alongside the Claude API

     13 details in Everything

111. v2.1.94 Apr 7, 2026 · release notes

     Claude Code v2.1.94 adds Mantle-powered Bedrock support, raises default effort to high, and expands hook/plugin capabilities.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.94 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.94

     └──▷ TRY IT
     Route Claude Code through Amazon Bedrock powered by Mantle instead of the default endpoint.
     copy

     $ CLAUDE_CODE_USE_MANTLE=1 claude

     Set a custom session title from a UserPromptSubmit hook by returning hookSpecificOutput.sessionTitle.
     json copy

     {
       "hookSpecificOutput": {
         "sessionTitle": "Incident triage — 2025-07-01"
       }
     }

     • Adds Amazon Bedrock powered by Mantle support via CLAUDE_CODE_USE_MANTLE=1.
     • Raises default effort level from medium to high for API-key, Bedrock/Vertex/Foundry, Team, and Enterprise users (adjustable with /effort).
     • Adds compact Slacked #channel header with a clickable channel link for Slack MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. send-message tool calls.
     • Adds keep-coding-instructions frontmatter field support for plugin output styles.
     • Adds hookSpecificOutput.sessionTitle to UserPromptSubmit hooks for setting the session title programmatically.

     +3 more show less

     • Plugin skills declared via "skills": ["./"] now use the skill frontmatter name as the invocation name instead of the directory basename, providing a stable name across install methods.
     • Improves --resume to resume sessions from other worktrees of the same repo directly instead of printing a cd command.
     • [VSCode] Adds a warning banner when settings.json files fail to parse so users know their permission rules are not being applied.

     8 details in Everything

112. v2.1.92 Apr 4, 2026 · release notes

     Claude Code v2.1.92 adds fail-closed remote policy enforcement, an interactive Bedrock setup wizard, and richer cost breakdowns.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.92 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.92

     └──▷ TRY IT
     Name a Remote Control session with your own prefix so it's identifiable across machines in a shared team environment.
     copy

     $ claude --remote-control-session-name-prefix my-team

     • Adds forceRemoteSettingsRefresh policy setting that blocks CLI startup until remote managed settings are freshly fetched and exits on fetch failure (fail-closed).
     • Adds an interactive Bedrock setup wizard from the login screen (3rd-party platform) guiding AWS auth, region config, credential verification, and model pinning.
     • Adds per-model and cache-hit cost breakdown to the /cost command for subscription users.
     • Upgrades /release-notes to an interactive version picker.
     • Remote Control session names now default to your hostname as prefix (e.g. myhost-graceful-unicorn), overridable with --remote-control-session-name-prefix.

     +2 more show less

     • Pro users see a footer hint when returning to a session after prompt cache expiry, showing estimated uncached tokens for the next turn.
     • Write tool diff computation is 60% faster on large files containing tabs, &, or $.

     7 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The /tag command has been removed.
     • !The /vim command has been removed; vim mode must now be toggled via /config → Editor mode.
113. v2.1.91 Apr 2, 2026 · release notes

     Claude Code v2.1.91 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. large-result passthrough, plugin executables, and multi-line deep link support.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.91 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.91

     └──▷ USE IT
     Pass a large DB schema or other oversized MCP tool result through without truncation by annotating the result metadata.
     json copy

     { "_meta": { "anthropic/maxResultSizeChars": 500000 }, "content": "<full schema here>" }

     Open Claude Code from a browser or launcher with a multi-line prompt pre-filled — useful for scripted handoffs or automation.
     copy

     $ open 'claude-cli://open?q=Summarize%20this%20file%0AHere%20is%20the%20context'

     Lock down a shared or CI environment so plugin and slash commands cannot execute inline shell code.
     json copy

     # In your Claude Code settings.json
     { "disableSkillShellExecution": true }

     • Adds MCP tool result persistence override via _meta["anthropic/maxResultSizeChars"] annotation (up to 500K), allowing large results like DB schemas to pass through without truncation.
     • New disableSkillShellExecution setting to disable inline shell execution in skills, custom slash commands, and plugin commands.
     • Supports multi-line prompts in claude-cli://open?q= deep links — encoded newlines (%0A) are no longer rejected.
     • Plugins can now ship executables under bin/ and invoke them as bare commands from the Bash tool.

     4 details in Everything

114. v2.1.90 Apr 1, 2026 · release notes

     Claude Code v2.1.90 adds /powerup interactive lessons, offline plugin cache control, and faster /resume loading.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.90 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.90

     └──▷ TRY IT
     Launch interactive feature lessons to onboard yourself or a teammate to Claude Code capabilities.
     copy

     $ /powerup

     Keep plugin marketplace working in an offline or network-restricted environment where git pull fails.
     copy

     $ CLAUDE_CODE_PLUGIN_KEEP_MARKETPLACE_ON_FAILURE=1 claude

     • Adds /powerup command: interactive lessons teaching Claude Code features with animated demos.
     • Adds CLAUDE_CODE_PLUGIN_KEEP_MARKETPLACE_ON_FAILURE env var to preserve the existing marketplace cache when git pull fails, enabling plugin use in offline or air-gapped environments.
     • Adds .husky to the list of protected directories in acceptEdits mode.
     • Improves /resume all-projects view to load project sessions in parallel, reducing wait times for users with many projects.
     • Changes --resume picker to hide sessions created by claude -p or SDK invocations, keeping the list focused on interactive sessions.

     +1 more show less

     • Removes Get-DnsClientCache and ipconfig /displaydns from the auto-allow list to protect DNS cache privacy.

     6 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Get-DnsClientCache and ipconfig /displaydns are no longer auto-allowed; setups or scripts that relied on these commands running without a permission prompt will now require explicit approval.
115. v2.1.89 Apr 1, 2026 · release notes

     Claude Code v2.1.89 adds deferred-permission hooks, flicker-free rendering, PermissionDenied hook, and faster MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. connections in headless mode.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.89 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.89

     └──▷ TRY IT
     Pause a headless session at a sensitive tool call and let a hook re-evaluate before proceeding — useful for human-in-the-loophuman-in-the-loopHuman in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. approval pipelines.
     copy

     $ claude -p --resume <session-id>

     Speed up CI/automation runs by skipping MCP server wait entirely and capping individual server connect time at 5 s.
     copy

     $ MCP_CONNECTION_NONBLOCKING=true claude -p --mcp-config mcp.json '<prompt>'

     Eliminate terminal flicker in long streaming sessions or when running inside a multiplexer like tmux.
     copy

     $ CLAUDE_CODE_NO_FLICKER=1 claude

     • Adds "defer" permission decision to PreToolUse hooks, letting headless sessions pause at a tool call and resume via -p --resume for hook re-evaluation.
     • Adds CLAUDE_CODE_NO_FLICKER=1 environment variable to enable flicker-free alt-screen rendering with virtualized scrollback.
     • Adds PermissionDenied hook that fires after auto-mode classifier denials — return {retry: true} to signal the model it can retry the tool call.
     • Adds named subagents to @ mention typeahead suggestions.
     • Adds MCP_CONNECTION_NONBLOCKING=true for -p mode to skip the MCP connection wait entirely; bounds --mcp-config server connections at 5 s instead of blocking on the slowest server.

     +5 more show less

     • Auto mode: denied commands now surface a notification and appear in /permissions → Recent tab where you can retry with r.
     • Enables Edit to work on files viewed via Bash with sed -n or cat, without requiring a separate Read call first.
     • Hook output exceeding 50 K characters is now saved to disk with a file path and preview instead of being injected directly into context.
     • /env now applies to PowerShell tool commands (previously only affected Bash).
     • Pasting !command into an empty prompt now enters bash mode, matching typed ! behavior.

     10 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !cleanupPeriodDays: 0 in settings.json is now rejected with a validation error (it previously silently disabled transcript persistence).
     • !Thinking summaries are no longer generated by default in interactive sessions — existing workflows that relied on them must set showThinkingSummaries: true in settings.json.
116. v2.1.86 Mar 27, 2026 · release notes

     Claude Code v2.1.86 adds session-ID headers for proxy aggregation and expands VCSVCSVersion control system: software that tracks changes to files over time, letting teams collaborate and roll back mistakes. Cyber tools use it to manage source code, configs, and signatures. exclusions to Jujutsu and Sapling.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.86 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.86

     └──▷ TRY IT
     Filter and group Claude Code traffic in a reverse proxy or SIEMSIEMSecurity information and event management, collecting logs from many systems and correlating events to find suspicious activity. Analysts use it to investigate alerts and incidents. by session without decoding request bodies.
     copy

     $ # In your proxy or log pipeline, match on the new header:
     X-Claude-Code-Session-Id: <session-id>

     • Adds X-Claude-Code-Session-Id header to all API requests so proxies can correlate and aggregate a session's requests without body inspection.
     • Adds .jj and .sl to VCS directory exclusion lists, preventing Grep and file autocomplete from descending into Jujutsu or Sapling metadata directories.
     • Read tool now uses a compact line-number format and deduplicates unchanged re-reads, reducing token usage in file-heavy workflows.
     • Memory filenames in the Saved N memories notice now highlight on hover and are clickable to open the file directly.
     • Skill descriptions in the /skills listing are capped at 250 characters to reduce context consumption.

     +5 more show less

     • /skills menu now sorts alphabetically for easier scanning.
     • Auto mode now displays 'unavailable for your plan' when restricted by plan limits instead of the misleading temporarily unavailable.
     • Improves prompt cache hit rate for Bedrock, Vertex, and Foundry users by removing dynamic content from tool descriptions.
     • Reduces startup event-loop stalls when many claude.ai MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. connectors are configured by extending the macOS keychain cache from 5 s to 30 s.
     • Reduces token overhead when mentioning files with @ by no longer JSON-escaping raw string content.

     10 details in Everything

117. v2.1.85 Mar 26, 2026 · release notes

     Claude Code v2.1.85 adds conditional hook filtering, richer PreToolUse headless flows, and expanded deep-link query support.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.85 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.85

     └──▷ USE IT
     Run a git-related hook only when Bash commands match git *, avoiding overhead for every other tool call.
     json copy

     # In your hooks config (e.g., .claude/settings.json)
     {
       "hooks": {
         "PreToolUse": [
           {
             "if": "Bash(git *)",
             "command": "./scripts/audit-git-command.sh"
           }
         ]
       }
     }

     Allow a headless CI integration to answer tool-permission questions programmatically without interactive prompts.
     json copy

     # PreToolUse hook returns both a permission decision and updated input
     # so AskUserQuestion is satisfied without user interaction
     {
       "permissionDecision": "allow",
       "updatedInput": { "answer": "yes" }
     }

     Enable full tool-parameter logging in OTELOTELAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. tool_result events for detailed pipeline observability.
     copy

     $ OTEL_LOG_TOOL_DETAILS=1 claude --mcp-config mcp.json 'audit my repo'

     • Adds CLAUDE_CODE_MCP_SERVER_NAME and CLAUDE_CODE_MCP_SERVER_URL environment variables to MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. headersHelper scripts, enabling a single helper script to serve multiple MCP servers.
     • Adds conditional if field to hooks using permission rule syntax (e.g., Bash(git *)) so hooks only fire for matching tool calls, reducing unnecessary process spawning.
     • Adds timestamp markers in transcripts when scheduled tasks (/loop, CronCreate) fire, making it easier to audit timing of automated runs.
     • Expands deep link query support (claude-cli://open?q=…) to 5,000 characters, with a scroll to review warning for long pre-filled prompts.
     • MCP OAuthMCP OAuthAn authentication layer for the Model Context Protocol that lets MCP clients obtain delegated access tokens so cyber tools can call MCP servers on a user's behalf without handling raw credentials. now follows RFC 9728 Protected Resource Metadata discovery to locate the authorization server automatically.

     +4 more show less

     • Organization-policy-blocked plugins (managed-settings.json) are now hidden from marketplace views and cannot be installed or enabled.
     • PreToolUse hooks can now satisfy AskUserQuestion by returning updatedInput alongside permissionDecision: "allow", enabling headless integrations to supply answers programmatically.
     • Gates tool_parameters in OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. tool_result events behind the OTEL_LOG_TOOL_DETAILS=1 environment variable to reduce log verbosity by default.
     • Improves PowerShell dangerous command detection coverage.

     9 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !tool_parameters in OpenTelemetry tool_result events are no longer emitted by default; set OTEL_LOG_TOOL_DETAILS=1 to restore them.
118. v2.1.84 Mar 26, 2026 · release notes

     Claude Code v2.1.84 adds PowerShell tool preview, new hooks, env-var overrides for 3p model capability detection, and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. deduplication.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.84 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.84

     └──▷ TRY IT
     Override capability detection when using a pinned Bedrock/Vertex/Foundry model that Claude Code doesn't recognize as supporting extended thinkingextended thinkingIntermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens..
     copy

     $ ANTHROPIC_DEFAULT_OPUS_MODEL_SUPPORTS=thinking claude

     Raise the streaming idle watchdog timeout to 3 minutes for slow-responding enterprise API gateways.
     copy

     $ CLAUDE_STREAM_IDLE_TIMEOUT_MS=180000 claude

     • Adds PowerShell tool for Windows as an opt-in preview, enabling native PowerShell execution alongside the existing bash tool.
     • Adds ANTHROPIC_DEFAULT_{OPUS,SONNET,HAIKU}_MODEL_SUPPORTS env vars to override effort/thinking capability detection for pinned default models on Bedrock, Vertex, and Foundry; companion _MODEL_NAME/_DESCRIPTION vars customize the /model picker label.
     • Adds CLAUDE_STREAM_IDLE_TIMEOUT_MS env var to configure the streaming idle watchdog threshold (default 90 s).
     • Adds TaskCreated hook that fires when a task is created via TaskCreate.
     • Adds type: "http" support for the WorktreeCreate hook — return the created worktree path via hookSpecificOutput.worktreePath in the response JSON.

     +10 more show less

     • Adds allowedChannelPlugins managed setting for team/enterprise admins to define a channel plugin allowlist.
     • Adds x-client-request-id header to all API requests to aid timeout debugging.
     • Adds idle-return prompt that nudges users back after 75+ minutes of inactivity to /clear, reducing unnecessary token re-caching on stale sessions.
     • Deep links (claude-cli://) now open in the user's preferred terminal instead of the first terminal found in the detection list.
     • Rules and skills paths: frontmatter now accepts a YAML list of globs, not just a single pattern.
     • MCP servers configured both locally and via claude.ai connectors are now deduplicated — local config wins.
     • Background bash tasks stuck on an interactive prompt now surface a notification after ~45 seconds.
     • Global system-prompt caching now works when ToolSearch is enabled, including for users with MCP tools configured.
     • [VSCode] Adds a rate-limit warning banner showing usage percentage and reset time.
     • Stats screenshot (Ctrl+S in /stats) now works in all builds and renders 16× faster.

     15 details in Everything

119. v2.1.83 Mar 25, 2026 · release notes

     Claude Code v2.1.83 adds drop-in policy fragments, new hook events, credential scrubbing, transcript search, and agent initialPrompt support.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.83 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.83

     └──▷ TRY IT
     Strip cloud credentials from all subprocesses to prevent accidental credential leakage to spawned tools and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers.
     copy

     $ CLAUDE_CODE_SUBPROCESS_ENV_SCRUB=1 claude

     Prevent Claude Code from registering the claude-cli:// URL handler on a locked-down or shared workstation.
     json copy

     # In managed-settings.json or a managed-settings.d/ fragment:
     { "disableDeepLinkRegistration": true }

     • Adds managed-settings.d/ drop-in directory so separate teams can deploy independent policy fragments that merge alphabetically alongside managed-settings.json.
     • Adds CwdChanged and FileChanged hook events for reactive environment management (e.g., direnv integration).
     • Adds sandbox.failIfUnavailable setting to exit with an error when sandbox is enabled but cannot start, instead of silently running unsandboxed.
     • Adds disableDeepLinkRegistration setting to prevent claude-cli:// protocol handler registration.
     • Adds CLAUDE_CODE_SUBPROCESS_ENV_SCRUB=1 environment variable to strip Anthropic and cloud provider credentials from subprocess environments (Bash tool, hooks, MCP stdio servers).

     +10 more show less

     • Adds transcript search — press / in transcript mode (Ctrl+O) to search, then n/N to step through matches.
     • Adds Ctrl+X Ctrl+E as a readline-native alias for opening the external editor (Ctrl+G still works).
     • Pasted images now insert an [Image #N] chip at the cursor for positional referencing in prompts.
     • Agents can now declare initialPrompt in frontmatter to auto-submit a first turn.
     • chat:killAgents and chat:fastMode keybindings are now rebindable via ~/.claude/keybindings.json.
     • /status now works while Claude is responding, instead of being queued until the turn finishes.
     • Linux now respects XDG_DATA_HOME when registering the claude-cli:// protocol handler.
     • WebFetch now identifies as Claude-User so site operators can recognize and allowlist Claude Code traffic via robots.txt.
     • Increased non-streaming fallback token cap from 21k to 64k and timeout from 120s to 300s so fallback requests are less likely to be truncated.
     • [VSCode] Spinner now turns red with Not responding when the backend hasn't responded for 60 seconds.

     15 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The 'stop all background agents' keybinding is changed from Ctrl+F to Ctrl+X Ctrl+K.
     • !TaskOutput tool is deprecated in favor of using Read on the background task's output file path.
120. v2.1.81 Mar 20, 2026 · release notes

     Claude Code v2.1.81 adds --bare flag for headless scripting and --channels permission relay to mobile devices.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.81 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.81

     └──▷ TRY IT
     Run a fully headless, hook-free scripted prompt in CI where only an API key is available — no OAuth, no plugin overhead.
     copy

     $ claude --bare -p "Summarize the security findings in findings.json" --settings ci-settings.json

     • Adds --bare flag for scripted -p calls — skips hooks, LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code., plugin sync, and skill directory walks; requires ANTHROPIC_API_KEY or apiKeyHelper via --settings (OAuth/keychain and auto-memory disabled).
     • Adds --channels permission relay so channel servers that declare the permission capability can forward tool-approval prompts to your phone.
     • Updates MCP OAuthMCP OAuthAn authentication layer for the Model Context Protocol that lets MCP clients obtain delegated access tokens so cyber tools can call MCP servers on a user's behalf without handling raw credentials. to support Client ID Metadata Document (CIMDCIMDA text-based Short Message Service protocol from Nokia for exchanging SMS messages between applications and SMS centers, used to build tools that send, receive, or test carrier messaging flows. / SEP-991) for servers without Dynamic Client Registration.
     • Improves MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. read/search tool calls to collapse into a single Queried {server} line, expandable with Ctrl+O.
     • Improves ! bash mode discoverability — Claude now suggests it when you need to run an interactive command.

     +2 more show less

     • Improves ref-tracked plugin freshness — plugins now re-clone on every load to pick up upstream changes.
     • Resuming a session that was in a worktree now automatically switches back to that worktree.

     7 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Plan mode now hides the clear context option by default; restore it by setting "showClearContextOnPlanAccept": true in settings.
     • !Line-by-line response streaming is disabled on Windows (including WSLWSLWindows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. in Windows Terminal) due to rendering issues.
121. v2.1.80 Mar 19, 2026 · release notes

     Claude Code v2.1.80 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. push channels, inline plugin settings, effort frontmatter, and rate-limit statusline fields.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.80 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.80

     └──▷ TRY IT
     Push messages into your session from an MCP server during a long-running workflow — currently in research preview.
     copy

     $ claude --channels

     Set a specific effort level for a skill or slash command so it always runs at the right cost/quality trade-off when invoked.
     yaml copy

     # In your skill or slash-command frontmatter:
     ---
     effort: low
     ---

     • Adds rate_limits field to statusline scripts exposing Claude.ai rate-limit usage across 5-hour and 7-day windows, including used_percentage and resets_at.
     • Adds source: 'settings' plugin marketplace source so plugin entries can be declared inline in settings.json without a separate marketplace lookup.
     • Adds CLI tool usage detection to plugin tips, supplementing existing file-pattern matching for smarter plugin suggestions.
     • Adds effort frontmatter key for skills and slash commands to override the model effort level on a per-command basis.
     • Adds --channels (research preview) enabling MCP servers to push messages directly into a running session.

     5 details in Everything

122. v2.1.79 Mar 18, 2026 · release notes

     Claude Code v2.1.79 adds Console auth, turn-duration display, remote-control bridge, and multi-directory plugin seeds.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.79 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.79

     └──▷ TRY IT
     Authenticate with the Anthropic Console (API billing) instead of the default Claude.ai login flow.
     copy

     $ claude auth login --console

     Load plugins from multiple seed directories at once — useful when team-wide and project-local plugins live in separate paths.
     copy

     $ CLAUDE_CODE_PLUGIN_SEED_DIR=/shared/plugins:/project/.claude/plugins claude

     Bridge a running VSCode session to claude.ai/code so you can continue the conversation from a browser or phone.
     copy

     $ /remote-control

     • Adds --console flag to claude auth login for Anthropic Console (API billing) authentication.
     • Adds Show turn duration toggle to the /config menu to display how long each turn takes.
     • Adds CLAUDE_CODE_PLUGIN_SEED_DIR support for multiple seed directories separated by the platform path delimiter (: on Unix, ; on Windows).
     • [VSCode] Adds /remote-control command to bridge your session to claude.ai/code for continuation from a browser or phone.
     • [VSCode] Session tabs now receive AI-generated titles based on your first message.

     5 details in Everything

123. v2.1.78 Mar 17, 2026 · release notes

     Claude Code v2.1.78 adds a StopFailure hook, plugin persistent state, and frontmatter-driven agent configuration.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.78 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.78

     └──▷ USE IT
     Trigger a Slack alert or pager when a session dies mid-run due to a rate limit or auth failure.
     json copy

     # In your Claude Code hooks config, add a StopFailure handler:
     {
       "hooks": {
         "StopFailure": "notify-on-error.sh"
       }
     }

     Add a custom internal model (e.g. a fine-tune) to the /model picker so teammates can select it by name.
     copy

     $ export ANTHROPIC_CUSTOM_MODEL_OPTION=my-org/claude-fine-tune-v1
     export ANTHROPIC_CUSTOM_MODEL_OPTION_NAME="FineTune v1"
     export ANTHROPIC_CUSTOM_MODEL_OPTION_DESCRIPTION="Internal security-tuned variant"
     claude

     Store plugin state across updates so that user preferences or cached data persist after a /plugin upgrade.
     copy

     $ # Inside your plugin script, write persistent data to the provided path:
     echo '{"last_scan": "2025-01-01"}' > "${CLAUDE_PLUGIN_DATA}/state.json"

     • New StopFailure hook event fires when a turn ends due to an API error (rate limit, auth failure, etc.), enabling automated error-handling workflows.
     • New ${CLAUDE_PLUGIN_DATA} variable provides plugins a persistent state directory that survives plugin updates; /plugin uninstall now prompts before deleting it.
     • Plugin-shipped agents can now declare effort, maxTurns, and disallowedTools in frontmatter for per-agent configuration.
     • New ANTHROPIC_CUSTOM_MODEL_OPTION env var adds a custom entry to the /model picker, with optional _NAME and _DESCRIPTION suffix vars for display.
     • Terminal notifications (iTerm2/Kitty/Ghostty popups, progress bar) now propagate to the outer terminal when Claude Code runs inside tmux with set -g allow-passthrough on.

     +1 more show less

     • Response text now streams line-by-line as it is generated.

     6 details in Everything

124. v2.1.77 Mar 17, 2026 · release notes

     Claude Code v2.1.77 raises output token limits to 128k, adds allowRead sandbox setting, and improves /copy, plugins, and agent resumption.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.77 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.77

     └──▷ USE IT
     Re-allow read access to a specific path inside a broader denyRead region for fine-grained sandbox control.
     yaml copy

     # In your Claude Code sandbox config:
     denyRead: ["/home/user"]
     allowRead: ["/home/user/project"]

     Retrieve a specific earlier assistant response — useful when you want to copy a response from several turns ago without scrolling.
     copy

     $ /copy 3

     Validate a plugin's frontmatter and hooks config before publishing to catch schema and YAML errors early.
     copy

     $ claude plugin validate ./my-plugin

     • Raises default max output tokens for Claude Opus 4.6 to 64k, and the upper bound for Opus 4.6 and Sonnet 4.6 models to 128k tokens.
     • New allowRead sandbox filesystem setting lets you re-allow read access within denyRead regions for fine-grained filesystem control.
     • /copy now accepts an optional index — /copy N copies the Nth-latest assistant response.
     • Renames /fork to /branch (/fork remains as an alias).
     • Background bash tasks are now killed if output exceeds 5 GB, preventing runaway processes from filling disk.

     +8 more show less

     • Sessions are now auto-named from plan content when you accept a plan.
     • Improved claude plugin validate to check skill, agent, and command frontmatter plus hooks/hooks.json, catching YAML parse errors and schema violations.
     • Headless mode plugin installation now composes correctly with CLAUDE_CODE_PLUGIN_SEED_DIR.
     • Shows a notice when apiKeyHelper takes longer than 10s, preventing it from blocking the main loop.
     • Faster startup on macOS (~60 ms) by reading keychain credentials in parallel with module loading.
     • Faster --resume on fork-heavy and very large sessions — up to 45% faster loading and ~100–150 MB less peak memory.
     • [VSCode] Plan preview tab titles now use the plan's heading instead of Claude's Plan.
     • [VSCode] Footer now points to the macOptionClickForcesSelection setting when option+click doesn't trigger native selection on macOS.

     13 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The Agent tool no longer accepts a resume parameter — use SendMessage({to: agentId}) to continue a previously spawned agent.
125. v2.1.76 Mar 14, 2026 · release notes

     Claude Code v2.1.76 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. elicitation, new hooks, sparse-checkout worktrees, /effort command, and session naming.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.76 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.76

     └──▷ TRY IT
     Name a session at startup so it's identifiable in Remote Control session lists and logs.
     copy

     $ claude -n "incident-2025-investigation" --worktree

     Load plugins from two separate directories (e.g. company-wide and personal) using the updated --plugin-dir flag.
     copy

     $ claude --plugin-dir /opt/company-plugins --plugin-dir ~/.local/claude-plugins

     • Adds MCP elicitation support — MCP servers can now request structured input mid-task via an interactive dialog (form fields or browser URL).
     • Adds Elicitation and ElicitationResult hooks to intercept and override MCP elicitation responses before they're sent back.
     • Adds -n / --name <name> CLI flag to set a display name for a session at startup.
     • Adds worktree.sparsePaths setting for claude --worktree to check out only specified directories via git sparse-checkout in large monorepos.
     • Adds PostCompact hook that fires after conversation compaction completes.

     +4 more show less

     • Adds /effort slash command to set the model effort level interactively.
     • Adds session quality survey with configurable sample rate via the feedbackSurveyRate setting for enterprise admins.
     • Improves background agent behavior — killing a background agent now preserves its partial results in the conversation context.
     • Updates --plugin-dir to support repeated use for multiple directories (one path per flag).

     9 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !--plugin-dir now only accepts one path per flag; use repeated --plugin-dir flags to specify multiple plugin directories.
126. v2.1.74 Mar 12, 2026 · release notes

     Claude Code v2.1.74 adds actionable /context optimization tips and a configurable auto-memory directory.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.74 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.74

     └──▷ USE IT
     Redirect auto-memory storage to a shared or project-specific path rather than the default location.
     json copy

     # In your Claude Code settings (e.g., settings.json)
     {
       "autoMemoryDirectory": "/home/user/projects/myproject/.claude-memory"
     }

     Spot which tools or memories are eating your context budget and get specific tips to reclaim space before hitting limits.
     copy

     $ /context

     Sideload a local plugin build that overrides a same-named marketplace plugin during development.
     copy

     $ claude --plugin-dir ./my-plugin-dev-copy

     • Adds actionable suggestions to the /context command — identifies context-heavy tools, memory bloat, and capacity warnings with specific optimization tips.
     • Adds autoMemoryDirectory setting to configure a custom directory for auto-memory storage.
     • Changes --plugin-dir so local dev copies now override installed marketplace plugins with the same name (unless force-enabled by managed settings).

     3 details in Everything

127. v2.1.73 Mar 11, 2026 · release notes

     Claude Code v2.1.73 adds modelOverrides for custom provider model IDs and actionable SSLSSLA cryptographic protocol for encrypting network connections, now deprecated in favor of TLS but whose name persists colloquially. Cyber tools reference it when handling HTTPS or other encrypted-channel inspection and termination. error guidance for corporate proxy users.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.73 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.73

     └──▷ USE IT
     Route model picker selections to a specific Bedrock inference profile ARN instead of the default model ID.
     json copy

     # In your Claude Code settings (e.g. ~/.claude/settings.json)
     {
       "modelOverrides": {
         "claude-opus-4-5": "arn:aws:bedrock:us-east-1::foundation-model/your-inference-profile-arn"
       }
     }

     • Adds modelOverrides setting to map model picker entries to custom provider model IDs (e.g., Bedrock inference profile ARNs).
     • Adds actionable guidance when OAuth login or connectivity checks fail due to SSL certificate errors (corporate proxies, NODE_EXTRA_CA_CERTS).
     • Updates default Opus model on Bedrock, Vertex, and Microsoft Foundry to Opus 4.6.
     • Enables /effort command to work while Claude is responding, matching /model behavior.

     4 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The /output-style command is deprecated; output style is now fixed at session start for better prompt caching — use /config instead.
128. v2.1.72 Mar 10, 2026 · release notes

     Claude Code v2.1.72 adds file-write in /copy, /plan descriptions, ExitWorktree tool, cron control, and a VS Code URI handler.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.72 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.72

     └──▷ TRY IT
     Write a focused selection to a file over SSH without relying on clipboard access.
     copy

     $ # In an active Claude Code session, open the copy picker with /copy, highlight the desired block, then press `w` to write it directly to a file.

     Jump straight into plan mode with context so Claude starts immediately rather than waiting for a follow-up prompt.
     copy

     $ claude> /plan fix the auth bug

     Stop all scheduled cron jobs mid-session without restarting Claude Code.
     copy

     $ CLAUDE_CODE_DISABLE_CRON=1 claude

     Open a new Claude Code tab in VS Code programmatically with a pre-filled prompt from an external script or tool.
     copy

     $ open 'vscode://anthropic.claude-code/open?prompt=explain+this+file&session=abc123'

     • Adds w key in /copy to write the focused selection directly to a file, bypassing the clipboard — useful over SSH.
     • Adds optional description argument to /plan (e.g., /plan fix the auth bug) to enter plan mode and immediately start working.
     • Adds ExitWorktree tool to cleanly leave an EnterWorktree session.
     • Adds CLAUDE_CODE_DISABLE_CRON environment variable to immediately stop scheduled cron jobs mid-session.
     • Adds lsof, pgrep, tput, ss, fd, and fdfind to the bash auto-approval allowlist, reducing permission prompts for common read-only operations.

     +9 more show less

     • Adds support for marketplace git URLs without a .git suffix (Azure DevOps, AWS CodeCommit).
     • Adds claude plugins as an alias for claude plugin.
     • Simplifies effort levels to low/medium/high (removed max) with new symbols (○ ◐ ●); adds /effort auto to reset to default.
     • Improves /config keyboard UX — Escape cancels changes, Enter saves and closes, Space toggles settings.
     • Improves up-arrow history to show the current session's messages first when running multiple concurrent sessions.
     • Improves voice input transcription accuracy for repo names and common dev terms (regex, OAuth, JSON).
     • Hides CLAUDE.md HTML comments (<!-- ... -->) from Claude when auto-injected; comments remain visible when read with the Read tool.
     • VSCode: Adds vscode://anthropic.claude-code/open URI handler to open a new Claude Code tab programmatically, with optional prompt and session query parameters.
     • VSCode: Adds effort level indicator on the input border.

     14 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The CLAUDE_CODE_PROXY_SUPPORTS_TOOL_REFERENCE environment variable has been removed; the tool search proxy bypass is now controlled by a different environment variable.
     • !Effort level max has been removed; valid levels are now low, medium, and high only.
129. v2.1.71 Mar 7, 2026 · release notes

     Claude Code v2.1.71 adds /loop scheduling, rebindable push-to-talk voice, and expands bash auto-approval allowlist

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.71 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.71

     └──▷ TRY IT
     Continuously poll a deploy status every 5 minutes without manually re-running the prompt — useful for watching long-running CI/CD pipelines.
     copy

     $ /loop 5m check the deploy

     Rebind push-to-talk to a modifier+letter combo that won't interfere with typing, then activate voice hands-free.
     json copy

     // In keybindings.json
     { "voice:pushToTalk": "meta+k" }

     • New /loop command runs a prompt or slash command on a recurring interval (e.g. /loop 5m check the deploy)
     • Adds cron scheduling tools for recurring prompts within a session
     • New voice:pushToTalk keybinding makes the voice activation key rebindable in keybindings.json (default: space); supports modifier+letter combos like meta+k
     • Expands bash auto-approval allowlist with fmt, comm, cmp, numfmt, expr, test, printf, getconf, seq, tsort, and pr
     • Updates /debug to toggle debug logging on mid-session without requiring a restart

     +2 more show less

     • Improves /plugin uninstall to disable project-scoped plugins in .claude/settings.local.json instead of .claude/settings.json, keeping changes out of shared teammate config
     • Plugin-provided MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers that duplicate a manually-configured server (same command/URL) are now automatically skipped, with suppressions shown in the /plugin menu

     7 details in Everything

130. v2.1.70 Mar 6, 2026 · release notes

     Claude Code v2.1.70 adds VS Code session browser, plan markdown view, and native MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server management dialog.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.70 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.70

     └──▷ TRY IT
     Manage MCP servers (enable/disable, reconnect, OAuth) directly from the VS Code chat panel without switching to a terminal.
     copy

     $ /mcp

     • Adds a spark icon in the VS Code activity bar listing all Claude Code sessions, with sessions opening as full editors.
     • Adds a full markdown document view for plans in VS Code, including support for inline comments to provide feedback.
     • Adds a native MCP server management dialog via /mcp in the VS Code chat panel — enable/disable servers, reconnect, and manage OAuth authentication without leaving the editor.

     3 details in Everything

131. v2.1.69 Mar 5, 2026 · release notes

     Claude Code v2.1.69 adds Voice STT in 20 languages, new hook events, plugin hot-reload, and a suite of new settings and CLI options.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.69 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.69

     └──▷ TRY IT
     Strip Claude's default git commit/PR instructions from the system prompt when your team enforces its own commit conventions via a CLAUDE.md rule.
     copy

     $ CLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS=1 claude

     Hot-reload plugins mid-session after editing a SKILL.md or plugin manifest, without restarting Claude Code.
     copy

     $ /reload-plugins

     • Adds /claude-api skill for building applications with the Claude API and Anthropic SDK
     • Adds Ctrl+U on an empty bash prompt (!) to exit bash mode, matching Escape and Backspace
     • Adds numeric keypad support for selecting options in Claude's interview questions
     • Adds optional --name argument to /remote-control and claude remote-control to set a custom session title visible in claude.ai/code
     • Expands Voice STT support to 20 total languages, adding Russian, Polish, Turkish, Dutch, Ukrainian, Greek, Czech, Danish, Swedish, and Norwegian

     +16 more show less

     • Adds effort level display (e.g., with low effort) to the logo and spinner so the active effort setting is always visible
     • Adds agent name display in the terminal title when using claude --agent
     • Adds sandbox.enableWeakerNetworkIsolation setting (macOS only) to allow Go programs like gh, gcloud, and terraform to verify TLS certificates when using a custom MITM proxy with httpProxyPort
     • Adds includeGitInstructions setting and CLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS env var to remove built-in commit and PR workflow instructions from Claude's system prompt
     • Adds /reload-plugins command to activate pending plugin changes without restarting
     • Adds ${CLAUDE_SKILL_DIR} variable so skills can reference their own directory in SKILL.md content
     • Adds InstructionsLoaded hook event that fires when CLAUDE.md or .claude/rules/*.md files are loaded into context
     • Adds agent_id and agent_type fields to hook events for subagent and --agent sessions
     • Adds worktree field to status line hook commands with name, path, branch, and original repo directory in --worktree sessions
     • Adds pluginTrustMessage in managed settings to append organization-specific context to the plugin trust warning
     • Adds pathPattern to strictKnownMarketplaces for regex-matching file/directory marketplace sources alongside hostPattern restrictions
     • Adds plugin source type git-subdir to point to a subdirectory within a git repo
     • Adds oauth.authServerMetadataUrl config option for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers to specify a custom OAuth metadata discovery URL when standard discovery fails
     • Extends policy limit fetching (e.g., remote control restrictions) to Team plan OAuth users, not just Enterprise
     • [VSCode] Adds compaction display as a collapsible Compacted chat card with the summary inside
     • [VSCode] Permission mode picker now respects permissions.disableBypassPermissionsMode from effective Claude Code settings

     21 details in Everything

132. v2.1.68 Mar 4, 2026 · release notes

     Claude Code v2.1.68 adds ultrathink keyword for on-demand high-effort reasoning and defaults Opus 4.6 to medium effort.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.68 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.68

     └──▷ TRY IT
     Force maximum reasoning depth on a single turn when tackling a complex architectural or security analysis problem.
     copy

     $ ultrathink — type this keyword in your message to Claude Code to elevate the next turn to high effort

     • Re-introduces the ultrathink keyword to trigger high-effort reasoning mode for the next turn.
     • Opus 4.6 now defaults to medium effort for Max and Team subscribers, balancing speed and thoroughness.

     2 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Opus 4 and Opus 4.1 are removed from Claude Code on the first-party API; users with either model pinned are automatically migrated to Opus 4.6.
133. v2.1.63 Feb 28, 2026 · release notes

     Claude Code v2.1.63 adds HTTP hooks, /simplify & /batch slash commands, and cross-worktree config sharing.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.63 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.63

     └──▷ TRY IT
     Opt out of claude.ai MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers entirely for a session — useful in air-gapped or compliance-restricted environments.
     copy

     $ ENABLE_CLAUDEAI_MCP_SERVERS=false claude

     • Adds /simplify and /batch bundled slash commands.
     • Adds HTTP hooks: POST JSON to a URL and receive JSON back, as an alternative to shell command hooks.
     • Adds ENABLE_CLAUDEAI_MCP_SERVERS=false environment variable to opt out of claude.ai MCP servers.
     • Improves /model command to display the currently active model in the slash command menu.
     • Project configs and auto memory are now shared across git worktrees of the same repository.

     +3 more show less

     • Adds manual URL paste fallback during MCP OAuthMCP OAuthAn authentication layer for the Model Context Protocol that lets MCP clients obtain delegated access tokens so cyber tools can call MCP servers on a user's behalf without handling raw credentials. authentication when the automatic localhost redirect fails.
     • Adds Always copy full response option to the /copy picker, skipping the code block picker for future /copy calls.
     • VSCode: Adds session rename and remove actions to the sessions list.

     8 details in Everything

134. v2.1.59 Feb 26, 2026 · release notes

     Claude Code v2.1.59 adds auto-memory saving, a /copy picker for code blocks, and smarter always-allow bash prefixes.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.59 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.59

     └──▷ TRY IT
     After a session, review and edit what Claude auto-saved to persistent memory so future sessions inherit the right context.
     copy

     $ /memory

     When a response contains multiple code blocks, pick exactly the one snippet you want on your clipboard instead of manually selecting text.
     copy

     $ /copy

     • Adds auto-memory: Claude automatically saves useful context to persistent memory, manageable via the /memory command.
     • New /copy command opens an interactive picker to select and copy individual code blocks or the full response.
     • Smarter always allow prefix suggestions for compound bash commands — computes per-subcommand prefixes (e.g. each part of cd /tmp && git fetch && git push) instead of treating the whole command as one unit.
     • Reduces memory usage in multi-agent sessions by releasing completed subagent task state.

     4 details in Everything

135. v2.1.51 Feb 24, 2026 · release notes

     Claude Code v2.1.51 adds remote-control subcommand, custom npm registry support for plugins, and new SDK account environment variables.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.51 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.51

     └──▷ TRY IT
     Serve your local Claude Code environment to an external build system using the new remote-control subcommand.
     copy

     $ claude remote-control

     Extend the plugin git clone timeout when working behind a slow corporate proxy to avoid installation failures.
     copy

     $ CLAUDE_CODE_PLUGIN_GIT_TIMEOUT_MS=300000 claude

     Provide account metadata to Claude Code synchronously in SDK/CI contexts so telemetry events are correctly attributed from the very first event.
     copy

     $ CLAUDE_CODE_ACCOUNT_UUID=<uuid> CLAUDE_CODE_USER_EMAIL=user@example.com CLAUDE_CODE_ORGANIZATION_UUID=<org-uuid> claude --print "<prompt>"

     • Adds claude remote-control subcommand for external builds, enabling local environment serving for all users.
     • Adds CLAUDE_CODE_PLUGIN_GIT_TIMEOUT_MS environment variable to configure plugin marketplace git timeout (default raised from 30s to 120s).
     • Supports custom npm registries and specific version pinning when installing plugins from npm sources.
     • Reduces the tool-result disk-persistence threshold from 100K to 50K characters, lowering context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. consumption and extending conversation longevity.
     • Adds CLAUDE_CODE_ACCOUNT_UUID, CLAUDE_CODE_USER_EMAIL, and CLAUDE_CODE_ORGANIZATION_UUID environment variables so SDK callers can supply account metadata synchronously, eliminating a race condition in early telemetry.

     +2 more show less

     • Routes HTTP hooks through the sandbox network proxy when sandboxingsandboxingRunning code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. is enabled, enforcing the domain allowlist.
     • The /model picker now shows human-readable labels (e.g., Sonnet 4.5) instead of raw model IDs for pinned versions, with an upgrade hint when a newer version is available.

     7 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !BashTool now skips the login shell (-l flag) by default when a shell snapshot is available (previously required setting CLAUDE_BASH_NO_LOGIN=true to achieve this); environments that relied on login-shell initialization running unconditionally may behave differently.
     • !HTTP hook header values no longer interpolate arbitrary environment variables — env var interpolation now requires an explicit allowedEnvVars list in the hook configuration, so existing hooks using env var interpolation in headers will stop expanding those variables.
136. v2.1.50 Feb 20, 2026 · release notes

     Claude Code v2.1.50 adds worktree isolation for agents, new hook events, claude agents CLI command, and 1M context controls.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.50 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.50

     └──▷ TRY IT
     List all configured agents to audit what's available in your Claude Code setup.
     copy

     $ claude agents

     Declare an agent that runs in its own isolated git worktree, keeping changes sandboxed from your main branch.
     yaml copy

     isolation: worktree

     Disable 1M context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. support to control token usage or costs in environments where smaller context is preferred.
     copy

     $ CLAUDE_CODE_DISABLE_1M_CONTEXT=1 claude -p "summarize this repo"

     • Adds WorktreeCreate and WorktreeRemove hook events for custom VCSVCSVersion control system: software that tracks changes to files over time, letting teams collaborate and roll back mistakes. Cyber tools use it to manage source code, configs, and signatures. setup and teardown during agent worktree isolation.
     • Supports isolation: worktree in agent definitions, letting agents declaratively run in isolated git worktrees.
     • New claude agents CLI command lists all configured agents.
     • Adds startupTimeout configuration option for LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code. servers.
     • Adds CLAUDE_CODE_DISABLE_1M_CONTEXT environment variable to disable 1M context window support.

     +5 more show less

     • Opus 4.6 (fast mode) now includes the full 1M context window.
     • CLAUDE_CODE_SIMPLE mode now also disables MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools, attachments, hooks, and CLAUDE.md file loading for a fully minimal experience.
     • VSCode: Adds /extra-usage command support in VS Code sessions.
     • Improves startup performance for headless mode (-p flag) by deferring Yoga WASMWASMA binary instruction format for a stack-based virtual machine, maintained by the W3C, that lets code compiled from any language run at near-native speed inside a sandboxed environment across browsers and servers. and UI component imports.
     • Improves memory efficiency during long sessions by clearing internal caches after compaction and clearing large tool results after processing.

     10 details in Everything

137. v2.1.49 Feb 19, 2026 · release notes

     Claude Code v2.1.49 adds file editing in Simple mode, ConfigChange hooks for auditing, and richer SDK model capability discovery.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.49 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.49

     └──▷ TRY IT
     Use Simple mode for lightweight file-editing tasks without needing a full agent setup.
     copy

     $ CLAUDE_CODE_SIMPLE=1 claude

     • Simple mode (CLAUDE_CODE_SIMPLE) now includes the file edit tool alongside the Bash tool, enabling direct file editing without leaving simple mode.
     • New ConfigChange hook event fires when configuration files change mid-session, enabling enterprise security auditing and optional blocking of settings changes.
     • SDK model info now exposes supportsEffort, supportedEffortLevels, and supportsAdaptiveThinking fields so consumers can programmatically discover model capabilities.
     • Permission suggestions are now populated when safety checks trigger an ask response, enabling SDK consumers to surface permission options to users.
     • Pressing Ctrl+C or ESC twice within 3 seconds now kills all background agents when the main thread is idle.

     +1 more show less

     • Improved permission prompts for path safety and working directory blocks now display the reason for the restriction.

     6 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Sonnet 4.5 with 1M context is being removed from the Max plan in favor of Sonnet 4.6 (which now has 1M context); users must switch via /model.
138. v2.1.47 Feb 18, 2026 · release notes

     Claude Code v2.1.47 adds new hook fields, keybindings, statusline data, and smarter background-agent lifecycle controls.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.47 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.47

     └──▷ TRY IT
     Kill all running background agents at once without interrupting the main thread.
     copy

     $ # While Claude Code is running, press ctrl+f to terminate all background agents instantly.

     • Adds last_assistant_message field to Stop and SubagentStop hook inputs, letting hooks read the final assistant response without parsing transcript files.
     • Adds chat:newline keybinding action for configurable multi-line input.
     • Adds added_dirs to the statusline JSON workspace section, exposing directories added via /add-dir to external scripts.
     • New ctrl+f shortcut kills all background agents; pressing ESC now cancels only the main thread while background agents keep running, giving finer control over agent lifecycle.
     • The /rename command now also updates the terminal tab title.

     +3 more show less

     • Increases initial session count in the /resume picker from 10 to 50 for faster session discovery.
     • Config backup files are now stored in ~/.claude/backups/ instead of the home directory root.
     • VS Code plan preview auto-updates as Claude iterates and keeps the preview open on rejection so Claude can revise.

     8 details in Everything

139. v2.1.45 Feb 17, 2026 · release notes

     Claude Code v2.1.45 adds Claude Sonnet 4.6 support, SDK rate-limit types, and richer plugin/spinner customization.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.45 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.45

     └──▷ USE IT
     Suppress all default spinner tips and show only your team's custom messages during long operations.
     json copy

     In your Claude Code settings (e.g. .claude/settings.json), add:
     {
       "spinnerTipsOverride": {
         "tips": ["Hang tight — running your scan…", "Almost there…"],
         "excludeDefault": true
       }
     }

     Load a shared plugin and marketplace config from a team-wide directory so all members get the same plugins without touching per-repo config.
     copy

     $ claude --add-dir /shared/team-claude-config

     • Supports Claude Sonnet 4.6 as a selectable model.
     • Reads enabledPlugins and extraKnownMarketplaces from --add-dir directories, enabling plugin and marketplace configuration via added directories.
     • New spinnerTipsOverride setting lets you supply custom spinner tip strings and optionally suppress built-in tips with excludeDefault: true.
     • New SDKRateLimitInfo and SDKRateLimitEvent types expose rate-limit utilization, reset times, and overage data to SDK consumers.
     • [VSCode] Permission destination choice (project/user/session) now persists across sessions.

     5 details in Everything

140. v2.1.41 Feb 13, 2026 · release notes

     Claude Code v2.1.41 adds auth subcommands, Windows ARM64ARM64The 64-bit instruction set architecture from ARM Holdings, used in modern mobile chips, Apple Silicon, and cloud server CPUs; tools target it to run natively on those platforms without emulation overhead. support, and smarter /rename.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.41 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.41

     └──▷ TRY IT
     Check your current authentication state without opening a browser — useful in CI or remote SSH sessions.
     copy

     $ claude auth status

     Log in to Claude Code from the CLI, e.g. when setting up a new machine or switching accounts.
     copy

     $ claude auth login

     Let Claude name the current session automatically based on what you've discussed, instead of typing a name manually.
     copy

     $ /rename

     • Adds claude auth login, claude auth status, and claude auth logout CLI subcommands for managing authentication from the terminal.
     • Adds native Windows ARM64 (win32-arm64) binary support.
     • Improves /rename to auto-generate a session name from conversation context when called without arguments.

     3 details in Everything

141. v2.1.33 Feb 6, 2026 · release notes

     Claude Code v2.1.33 adds multi-agent hook events, sub-agent spawn restrictions, persistent agent memory, and VSCode remote sessions.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.33 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.33

     └──▷ USE IT
     Restrict an orchestrator agent so it can only spawn specific sub-agent types, reducing blast radius in automated pipelines.
     yaml copy

     # In agent frontmatter (tools section):
     ---
     tools:
       - Task(researcher)
       - Task(summarizer)
     ---

     Give an agent persistent memory scoped to the current project so context survives across sessions.
     yaml copy

     # In agent frontmatter:
     ---
     memory: project
     ---

     • Adds TeammateIdle and TaskCompleted hook events for multi-agent workflows, enabling automation when agents become idle or complete tasks.
     • Supports restricting which sub-agents can be spawned via Task(agent_type) syntax in agent tools frontmatter.
     • Adds memory frontmatter field for agents with user, project, or local scope for persistent memory across sessions.
     • Adds plugin name to skill descriptions and the /skills menu for improved discoverability.
     • Improves API connection failure error messages to show specific causes (e.g., ECONNREFUSED, SSLSSLA cryptographic protocol for encrypting network connections, now deprecated in favor of TLS but whose name persists colloquially. Cyber tools reference it when handling HTTPS or other encrypted-channel inspection and termination. errors) instead of generic messages.

     +3 more show less

     • Surfaces errors from invalid managed settings.
     • VSCode: Adds support for remote sessions, allowing OAuth users to browse and resume sessions from claude.ai.
     • VSCode: Adds git branch and message count to the session picker with branch-name search support.

     8 details in Everything

142. v2.1.32 Feb 5, 2026 · release notes

     Claude Code v2.1.32 adds Opus 4.6, experimental multi-agent teams, automatic memory, and partial conversation summarization.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.32 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.32

     └──▷ TRY IT
     Spin up an experimental multi-agent team session to tackle complex, parallelizable tasks.
     copy

     $ CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 claude

     • Adds Claude Opus 4.6 model support.
     • Introduces experimental multi-agent collaboration via agent teams (enable with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1).
     • Claude now automatically records and recalls memories as it works across sessions.
     • Adds Summarize from here option in the message selector for partial conversation summarization.
     • Skills defined in .claude/skills/ inside --add-dir directories are now loaded automatically.

     +2 more show less

     • Skill character budget now scales with context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. size (2% of context), allowing more skill descriptions to appear untruncated for users with larger context windowscontext windowsThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it..
     • VSCode: Adds a spinner when loading the past conversations list.

     7 details in Everything

143. v2.1.31 Feb 4, 2026 · release notes

     Claude Code v2.1.31 adds session resume hints and Japanese IME support in checkbox selection.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.31 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.31

     • Adds a session resume hint on exit, showing the command to continue a conversation later.
     • Supports full-width (zenkaku) space input from Japanese IME in checkbox selection.

     2 details in Everything

144. v2.1.30 Feb 3, 2026 · release notes

     Claude Code v2.1.30 adds PDF page-range reading, OAuth credentials for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, and a new /debug command

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.30 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.30

     └──▷ TRY IT
     Register a Slack MCP server that requires pre-configured OAuth credentials instead of Dynamic Client Registration.
     copy

     $ claude mcp add slack-mcp --client-id <your-client-id> --client-secret <your-client-secret>

     • Adds pages parameter to the Read tool for PDFs, enabling specific page ranges (e.g., pages: "1-5"); large PDFs (>10 pages) now return a lightweight reference when @-mentioned instead of being inlined into context
     • Adds pre-configured OAuth client credentials for MCP servers that lack Dynamic Client Registration support (e.g., Slack) via new --client-id and --client-secret flags on claude mcp add
     • Adds /debug command for Claude to help troubleshoot the current session
     • Supports additional git log and git show flags in read-only mode, including --topo-order, --cherry-pick, --format, and --raw
     • Adds token count, tool uses, and duration metrics to Task tool results

     +4 more show less

     • Adds reduced motion mode to the config
     • Improves --resume memory usage by 68% for users with many sessions via lightweight stat-based loading and progressive enrichment
     • Changes /model to execute immediately instead of being queued
     • [VSCode] Adds multiline input support to the Other text input in question dialogs via Shift+Enter

     9 details in Everything

145. v2.1.27 Jan 30, 2026 · release notes

     Claude Code v2.1.27 adds --from-pr flag for PR-linked session resumption and automatic PR linking via gh pr create.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.27 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.27

     └──▷ TRY IT
     Pick up a Claude Code session exactly where you left off for a specific pull request, by PR number or full URL.
     copy

     $ claude --from-pr 42

     Resume a session using a full PR URL instead of a number — useful when working across multiple repos.
     copy

     $ claude --from-pr https://github.com/myorg/myrepo/pull/42

     • New --from-pr flag lets you resume a session linked to a specific GitHub PR number or URL.
     • Sessions created via gh pr create are now automatically linked to the corresponding PR.
     • Tool call failures and denials are now recorded in debug logs for easier troubleshooting workflows.
     • VSCode: Enables the Claude in Chrome integration.
     • Permissions now respect content-level ask over tool-level allow, so granular deny patterns like ask: ["Bash(rm *)"] correctly override a broad allow: ["Bash"].

     5 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Permission evaluation order has changed: a content-level ask rule now takes precedence over a tool-level allow. Configurations using allow: ["Bash"], ask: ["Bash(rm *)"] that previously permitted all Bash commands will now prompt for permission on rm commands.
146. v2.1.23 Jan 29, 2026 · release notes

     Claude Code v2.1.23 adds customizable spinner verbs and shows PR merge status and timeout durations in the UI.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.23 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.23

     • Adds spinnerVerbs setting to customize the verbs displayed in the activity spinner.
     • Shows timeout duration alongside elapsed time for Bash commands.
     • Displays a purple status indicator in the prompt footer for merged pull requests.

     3 details in Everything

147. v2.1.21 Jan 28, 2026 · release notes

     Claude Code v2.1.21 adds Japanese IME input support and automatic Python venv activation in VS Code.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.21 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.21

     └──▷ USE IT
     Opt out of automatic Python venv activation in VS Code if you manage your interpreter another way.
     json copy

     // In VS Code settings.json
     { "claudeCode.usePythonEnvironment": false }

     • Supports full-width (zenkaku) number input from Japanese IME in option selection prompts.
     • Adds automatic Python virtual environment activation in the VS Code extension, ensuring python and pip use the correct interpreter (configurable via claudeCode.usePythonEnvironment).
     • Improves read/search progress indicators to show Reading… while in progress and Read when complete.
     • Improves Claude's preference for file operation tools (Read, Edit, Write) over bash equivalents (cat, sed, awk).

     4 details in Everything

148. v2.1.20 Jan 27, 2026 · release notes

     Claude Code v2.1.20 adds PR review status in the footer, vim history nav, TaskUpdate deletions, and per-directory CLAUDE.md loading.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.20 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.20

     └──▷ TRY IT
     Load project-specific CLAUDE.md instructions from a shared config directory alongside the default working directory.
     copy

     $ CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1 claude --add-dir /path/to/shared-config

     • Adds arrow key history navigation in vim normal mode when the cursor cannot move further.
     • Adds external editor shortcut (Ctrl+G) to the help menu for easier discovery.
     • Adds PR review status indicator to the prompt footer showing the current branch's PR state (approved, changes requested, pending, or draft) as a colored dot with a clickable link.
     • Supports loading CLAUDE.md files from additional directories specified via --add-dir (requires CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1).
     • Adds ability to delete tasks via the TaskUpdate tool.

     +7 more show less

     • Improves /sandbox command UI to show dependency status with installation instructions when dependencies are missing.
     • Enables the /commit-push-pr skill to automatically post PR URLs to configured Slack channels via MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools.
     • Makes the /copy command available to all users.
     • Changes background agents to prompt for tool permissions before launching.
     • Accepts wildcard permission rules like Bash(*) and treats them as equivalent to Bash.
     • Renders teammate messages with rich Markdown formatting (bold, code blocks, lists, etc.) instead of plain text.
     • Changes ToolSearch results to appear as a brief notification instead of inline in the conversation.

     12 details in Everything

149. v2.1.19 Jan 23, 2026 · release notes

     Claude Code v2.1.19 adds a Tasks system, argument shorthands for custom commands, and session forking/rewind for all VS Code users.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.19 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.19

     └──▷ TRY IT
     Use positional argument shorthands in a custom command to pass the first two arguments directly, e.g. a file path and a label.
     copy

     $ # In a custom command definition:
     # Analyze $0 and tag the result as $1
     analyze_file $0 --tag $1

     • Adds $0, $1, etc. shorthand syntax for accessing individual arguments in custom commands.
     • Changes indexed argument syntax from $ARGUMENTS.0 to $ARGUMENTS[0] (bracket notation).
     • Skills without additional permissions or hooks are now allowed without requiring approval.
     • [SDK] Adds replay of queued_command attachment messages as SDKUserMessageReplay events when replayUserMessages is enabled.
     • [VS Code] Session forking and rewind functionality is now enabled for all users.

     5 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !Indexed argument syntax for custom commands changed from $ARGUMENTS.0 to $ARGUMENTS[0]; existing custom commands using the old dot notation will need to be updated.
150. v2.1.16 Jan 22, 2026 · release notes

     Claude Code v2.1.16 adds task management with dependency tracking, native VSCode plugin management, and remote session browsing for OAuth users.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.16 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.16

     • New task management system with dependency tracking between tasks.
     • [VSCode] Native plugin management support added to the IDE extension.
     • [VSCode] OAuth users can now browse and resume remote Claude sessions from the Sessions dialog.

     3 details in Everything

151. v2.1.14 Jan 20, 2026 · release notes

     Claude Code v2.1.14 adds history-based Tab autocomplete in bash mode, plugin search, SHA pinning, and a VSCode /usage command.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.14 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.14

     └──▷ TRY IT
     Quickly re-run a recent shell command in bash mode without retyping it — type the start of the command after ! and press Tab to pull from history.
     copy

     $ ! git log --on<Tab>   # completes to a previously run git command from your bash history

     • Adds history-based Tab autocomplete in bash mode (!) — type a partial command and press Tab to complete from your shell history.
     • Adds search/filter to the installed plugins list — type to filter by plugin name or description.
     • Supports pinning plugins to specific git commit SHAs, enabling marketplace entries to install exact versions.
     • Improves backspace to delete pasted text as a single token rather than one character at a time.
     • [VSCode] Adds /usage command to display current plan usage.

     5 details in Everything

152. v2.1.9 Jan 16, 2026 · release notes

     Claude Code v2.1.9 adds auto:N MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. threshold syntax, plansDirectory setting, external editor in prompts, and richer hook/skill APIs.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.9 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.9

     └──▷ USE IT
     Store plan files in a project-local directory so plans are version-controlled alongside your code.
     json copy

     # In your Claude Code settings (settings.json)
     { "plansDirectory": ".claude/plans" }

     Inject the current session ID into a skill command so downstream scripts can correlate activity back to the Claude session.
     json copy

     # In a skill definition
     { "command": "my-audit-tool --session ${CLAUDE_SESSION_ID}" }

     • Adds auto:N syntax to set the MCP tool search auto-enable threshold as a context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. percentage (0–100).
     • Adds plansDirectory setting to customize where plan files are stored.
     • Adds external editor support (Ctrl+G) inside AskUserQuestion Other input fields.
     • Adds session URL attribution to commits and PRs created from web sessions.
     • Enables PreToolUse hooks to return additionalContext directly to the model.

     +1 more show less

     • Adds ${CLAUDE_SESSION_ID} string substitution so skills can access the current session ID.

     6 details in Everything

153. v2.1.7 Jan 14, 2026 · release notes

     Claude Code v2.1.7 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tool search auto mode, permission feedback, and task notification previews.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.7 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.7

     └──▷ TRY IT
     Opt out of MCP tool search auto mode if you need all tool descriptions loaded upfront rather than discovered on demand.
     copy

     $ claude config set disallowedTools '["MCPSearch"]'

     • Adds showTurnDuration setting to hide turn-duration messages like Cooked for 1m 6s.
     • Enables feedback submission when accepting permission prompts.
     • Adds inline display of the agent's final response in task notifications, so results are visible without opening the full transcript.
     • Enables MCP tool search auto mode by default: when MCP tool descriptions exceed 10% of the context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it., they are deferred and discovered on-demand via the MCPSearch tool, reducing upfront context consumption for users with many MCP tools.
     • Updates OAuth and API Console URLs from console.anthropic.com to platform.claude.com.

     5 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !MCP tool search auto mode is now on by default; users who relied on all MCP tool descriptions being loaded upfront must add MCPSearch to disallowedTools in their settings to restore the previous behaviour.
154. v2.1.6 Jan 13, 2026 · release notes

     Claude Code v2.1.6 adds config search, date-ranged stats, nested skills discovery, and richer context windowcontext windowThe maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. metrics.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.6 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.6

     └──▷ TRY IT
     Review spending or usage trends over the last week vs. month without leaving the terminal.
     copy

     $ /stats   # then press 'r' to cycle through Last 7 days / Last 30 days / All time

     Display remaining context window capacity in the status line by referencing the new percentage fields.
     copy

     $ # In your status line config, reference the new fields:
     # context_window.used_percentage
     # context_window.remaining_percentage

     • Adds search/filter functionality inside the /config command for quickly locating settings.
     • Adds an Updates section to /doctor showing the auto-update channel and available npm versions (stable/latest).
     • Adds date range filtering to /stats — press r to cycle between Last 7 days, Last 30 days, and All time.
     • Adds automatic discovery of skills from nested .claude/skills directories when working with files in subdirectories.
     • Adds context_window.used_percentage and context_window.remaining_percentage fields to the status line for easier context window display.

     +5 more show less

     • Improves the external CLAUDE.md imports approval dialog to show which files are being imported and from where.
     • Improves /tasks dialog to go directly to task details when only one background task is running.
     • Improves @ autocomplete with icons for different suggestion types and single-line formatting.
     • Changes terminal title to Claude Code on startup for better window identification.
     • Caps task notification display at 3 lines with an overflow summary when multiple background tasks complete simultaneously.

     10 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The ability to @-mention MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers to enable/disable them has been removed — use /mcp enable <name> instead.
155. v2.1.4 Jan 11, 2026 · release notes

     Claude Code v2.1.4 adds env var to fully disable background tasks and the Ctrl+B shortcut.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.4 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.4

     └──▷ TRY IT
     Prevent Claude Code from ever backgrounding tasks — useful in CI pipelines or restricted environments where background processes are undesirable.
     copy

     $ CLAUDE_CODE_DISABLE_BACKGROUND_TASKS=1 claude

156. v2.1.3 Jan 9, 2026 · release notes

     Claude Code v2.1.3 adds release channel switching, unreachable-rule warnings, and a 10-minute hook timeout.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.3 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.3

     └──▷ TRY IT
     Switch between stable and latest release channels without leaving the CLI.
     copy

     $ /config

     Audit your permission rules for unreachable entries and get fix guidance before they silently do nothing.
     copy

     $ /doctor

     • Adds release channel toggle (stable or latest) to /config, letting you switch update tracks in-product.
     • New detection and warnings for unreachable permission rules surfaced in /doctor and after saving rules, with source attribution and actionable fix guidance.
     • Extends tool hook execution timeout from 60 seconds to 10 minutes, enabling longer-running hook scripts.
     • [VSCode] Adds clickable destination selector for permission requests, letting you choose where settings are saved: this project, all projects, shared with team, or session only.
     • Merges slash commands and skills into a unified concept, simplifying the mental model.

     5 details in Everything

157. v2.1.2 Jan 9, 2026 · release notes

     Claude Code v2.1.2 adds OSC 8 hyperlinks, winget support, plan-mode shortcut, and full large-output persistence to disk.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.1.2 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.1.2

     └──▷ TRY IT
     Force plugin autoupdates in CI or locked-down environments where the main auto-updater is disabled.
     copy

     $ FORCE_AUTOUPDATE_PLUGINS=1 claude

     • Adds clickable OSC 8 hyperlinks for file paths in tool output in supported terminals (e.g., iTerm).
     • Adds Windows Package Manager (winget) installation detection with automatic update instructions.
     • Adds Shift+Tab keyboard shortcut in plan mode to instantly select the auto-accept edits option.
     • Adds FORCE_AUTOUPDATE_PLUGINS environment variable to enable plugin autoupdate even when the main auto-updater is disabled.
     • Adds agent_type field to SessionStart hook input when --agent is specified.

     +3 more show less

     • Adds source path metadata to images dragged onto the terminal so Claude knows where images originated.
     • Large bash command outputs and tool outputs are now saved to disk instead of truncated, giving Claude access to full content via file references.
     • Unifies plugins and MCPs in the /plugins installed tab with scope-based grouping.

     8 details in Everything

     └──▷ BREAKING ON UPGRADE

     • !The Windows managed settings path C:\ProgramData\ClaudeCode\managed-settings.json is deprecated; administrators must migrate to C:\Program Files\ClaudeCode\managed-settings.json.
     • !The minimum zod peer dependency is now ^4.0.0 (SDK).
158. v2.0.74 Dec 19, 2025 · release notes

     Claude Code v2.0.74 adds LSPLSPLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code. code-intelligence tools, expands terminal setup support, and improves theme controls.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.0.74 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.0.74

     └──▷ TRY IT
     Configure optimal key bindings and shell integration for Kitty, Alacritty, Zed, or Warp in one step.
     copy

     $ /terminal-setup

     Quickly toggle syntax highlighting off for a distraction-free diff review while browsing themes.
     copy

     $ /theme  # then press ctrl+t to toggle syntax highlighting on/off

     • Adds LSP (Language Server ProtocolLanguage Server ProtocolLanguage Server Protocol, the standard way editors get code intelligence such as completions, go-to-definition and errors from a separate per-language server. Coding agents use it to understand code.) tool enabling code-intelligence features: go-to-definition, find references, and hover documentation.
     • Adds /terminal-setup support for Kitty, Alacritty, Zed, and Warp terminals.
     • Adds ctrl+t shortcut in /theme to toggle syntax highlighting on/off.
     • Adds syntax highlighting info display inside the theme picker.
     • Improves /context command visualization with skills and agents grouped by source, slash commands listed, and token counts sorted.

     5 details in Everything

159. v2.0.73 Dec 19, 2025 · release notes

     Claude Code v2.0.73 adds clickable image links, kill-ring cycling, plugin search, custom fork session IDs, and VS Code tab badges.

     └──▷ GET THIS VERSION

     copy

     $ git clone --branch v2.0.73 https://github.com/anthropics/claude-code.git
     # already have the repo? check out this version:
     $ git checkout v2.0.73

     └──▷ TRY IT
     Fork an existing session and assign a meaningful custom ID so you can reference it later by name rather than an auto-generated hash.
     copy

     $ claude --resume <existing-session-id> --fork-session --session-id my-audit-branch

     • Adds clickable [Image #N] links in the terminal that open attached images in the default viewer.
     • Adds alt-y yank-pop to cycle through kill-ring history after a ctrl-y yank.
     • Adds search filtering to the plugin discover screen — filter by name, description, or marketplace in real time.
     • Supports custom session IDs when forking sessions via --session-id combined with --resume or --continue and --fork-session.
     • Improves /theme command to open the theme picker directly.

     +1 more show less

     • [VS Code] Adds tab icon badges showing pending permissions (blue) and unread completions (orange).

     6 details in Everything