Claude Code April 2026 | Breaking Down 5 Changes and New Features

Change 1/5: Desktop Redesign — From "Supplementary Screen" to "Integrated Development Environment"

New Positioning — CLI as Primary, Desktop as Integrated Environment

The core of Claude Code remains the CLI (command-line interface). The desktop app has been redefined as an integrated environment that complements and enhances the CLI experience through a graphical pane layout. The official documentation states: "The Code tab within the Claude Desktop app lets you use Claude Code through a graphical interface instead of the terminal" Source. It is not a competing product with the CLI, but a different frontend for the same Claude Code.

Workspace layout, integrated terminal, file editor, side chat, and display mode features all require Claude Desktop v1.2581.0 or later Source.

Five Major Panes Added — Integrated Terminal / Editor / Diff / Preview / Side Chat

Source: Use Claude Code Desktop — Anthropic Docs

Multi-Session Parallel Execution — Switch with Ctrl+Tab, Each Session Has Independent Context

You can open multiple sessions simultaneously in one window and switch between them with Ctrl+Tab. Each session has its own independent context and change history. Automatic Git worktree isolation is also supported, so parallel sessions do not interfere with each other's work Source.

PR Monitoring — Auto-fix / Auto-merge Automates CI Integration

When you open a pull request (PR), a CI (continuous integration) status bar appears in the session. Auto-fix (automatic correction when CI checks fail) and Auto-merge (automatic merge after all checks pass) can be enabled via toggle Source.

Change 2/5: Cloud Routines — Automation That Keeps Running Even When Your Laptop Is Closed

What Are Routines — A Research Preview Feature That Runs Autonomously on Anthropic-Managed Cloud

Routines are currently in research preview. Behavior specifications, limits, and API formats are subject to change Source.

A Routine is a configuration unit that bundles a "prompt, repository, and connectors" together and runs automatically. Because the execution environment is Anthropic-managed cloud infrastructure, it continues to run even when your laptop is closed or powered off Source. Available plans are Pro, Max, Team, and Enterprise, and Claude Code on the Web at claude.ai/code must be enabled.

Comparison with Local Scheduled Tasks — Cloud Residency Offers Dramatically Better Stability

Source: Schedule recurring tasks in Claude Code Desktop — Anthropic Docs

3 Types of Triggers — Schedule / API / GitHub Events

Multiple triggers can be combined in a single Routine Source.

Schedule Trigger

Recurring execution: hourly, daily, weekdays, weekly, etc. Custom cron expressions can be set via /schedule update in the CLI (minimum interval: 1 hour). The scheduled time is entered in local time and automatically converted on the cloud side.

API Trigger

Triggered by POSTing to an HTTP endpoint dedicated to the routine. Can be called from any external service such as alert systems, deploy pipelines, or internal tools. Runtime context such as alert message body can be passed via the text field.

curl -X POST https://api.anthropic.com/v1/claude_code/routines/trig_01.../fire \
  -H "Authorization: Bearer sk-ant-oat01-xxxxx" \
  -H "anthropic-beta: experimental-cc-routine-2026-04-01" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"text": "Sentry alert SEN-4521 fired in prod."}'

Source: Automate work with routines — Anthropic Docs

This endpoint is provided under the experimental-cc-routine-2026-04-01 beta header. The request and response formats may change during the research preview period Source.

GitHub Event Trigger

Automatically triggered by events such as opening, closing, or updating a pull request (PR), or creating or publishing a release. Filter conditions (author, title, label, base branch, draft status, etc.) can be used to narrow down the scope Source.

Note: During the research preview period, GitHub webhook events are subject to per-routine and per-account hourly limits, and events exceeding the limit will be discarded until the window resets Source.

Typical Use Cases — Nightly Batch / Monitoring / Automated PR Review

The use cases illustrated in the official documentation are as follows Source.

• Backlog management: Every night, read new issues, apply labels and assign owners, then post a summary to Slack
• Alert triage: When a monitoring tool fires an alert, use an API trigger to analyze the stack trace and automatically draft a fix PR
• Automated PR review: Triggered by pull_request.opened, apply the team's custom review checklist and add inline comments
• Deploy verification: After a CD (continuous delivery) pipeline deployment, run a smoke check via API trigger and post results to the release channel
• Documentation drift detection: Weekly, scan merged PRs and create update PRs for docs that reference changed APIs

How to Configure — Create Routines Using the routines Command from Claude Code

• Web: Click "New routine" at claude.ai/code/routines
• CLI: Run the /schedule command in conversational mode. Example: /schedule daily PR review at 9am
• Desktop: On the Schedule page, select "New task" → "New remote task"

Change 3/5: Opus 4.7 xhigh Mode — Reasoning Intensity Between high and max

Overview of xhigh — Added in Claude Code v2.1.111 (2026-04-16)

A new effort level called xhigh for Opus 4.7 was added in Claude Code v2.1.111 (2026-04-16) Source. The official changelog states: "Added xhigh effort level for Opus 4.7, sitting between high and max."

How to Use xhigh — /effort xhigh or --effort xhigh Flag

/effort xhigh

Running the /effort command without arguments launches an interactive slider, where you can select the effort level with arrow keys and confirm with Enter. You can also pass the --effort xhigh flag as a command-line argument or select it from the model picker Source.

Note: xhigh is specific to Opus 4.7. If specified for another model, it will automatically fall back to high Source.

When to Use xhigh — Recommended as a Starting Point for High-Difficulty Implementations and Agent Work

The official Opus 4.7 documentation positions it as: "For coding and agent use cases, starting with xhigh is recommended" Source. It is primarily intended for high-difficulty implementation tasks and agent work where speed is not the priority. Token consumption is expected to be higher than high, though no specific multiplier is stated in official materials [Unconfirmed]. We recommend measuring actual consumption with representative tasks before deploying to production.

Change 4/5: Auto Mode — Autonomous Execution for Long-Running Agents

Overview of Auto Mode — --enable-auto-mode Flag No Longer Required for Max + Opus 4.7

In Claude Code v2.1.111 (2026-04-16), Max subscribers no longer need the --enable-auto-mode flag when using Auto mode with Opus 4.7 Source.

Auto mode is currently in research preview Source. The official documentation states: "Claude executes all actions with background safety checks that verify alignment with your request. Reduces permission prompts while maintaining oversight."

Available Plans and Models — Max Limited to Opus 4.7, Not Available for Pro

Source: Use Claude Code Desktop — Anthropic Docs

Enable it from "Settings → Claude Code."

Auto Mode vs. Permission Bypass — Auto Mode Has Safety Checks; Bypass Is for Sandboxes Only

Auto mode reduces permission prompts with background safety checks. In contrast, Bypass permissions mode executes all operations without any prompts, and the official documentation explicitly states it should "only be used in sandboxed containers or VMs" Source. Use Auto mode in standard development environments.

When to Enable It — Long-Running Autonomous Tasks, Automatic PR Fixes, Overnight Background Execution

• Agent tasks that need to run and complete autonomously over extended periods
• Repetitive routine work such as automatic PR fixes and dependency updates
• Overnight or background execution where humans cannot approve each step

Change 5/5: /ultrareview Command — Comprehensive Review with Cloud Parallel Multi-Agent

Overview of /ultrareview — Added in Claude Code v2.1.111

The /ultrareview command was added in Claude Code v2.1.111 (2026-04-16) Source.

From the official changelog: "Added /ultrareview for running comprehensive code review in the cloud using parallel multi-agent analysis and critique — invoke with no arguments to review your current branch, or /ultrareview <PR#> to fetch and review a specific GitHub PR"

How to Use /ultrareview — No Arguments for Current Branch, PR # to Auto-Fetch from GitHub

/ultrareview          # Review the current branch
/ultrareview 42       # Fetch and review GitHub PR #42

When run without arguments, it targets the current branch. When a PR number is specified, it automatically fetches the target PR from GitHub for review Source.

Difference from Diff Viewer Built-in Review — /ultrareview Is More Comprehensive, Suited for Large Changes

The desktop diff viewer has a built-in review feature via the "Review code" button. This focuses on "compilation errors, clear logic errors, security vulnerabilities, and obvious bugs," and does not cover style or pre-existing issues Source.

/ultrareview, by contrast, is oriented toward comprehensive review using "parallel multi-agent analysis in the cloud," making it better suited for scrutinizing larger codebases and complex changes Source.

How to Get the Most Out of Each Plan — Recommended Workflows for Pro / Max / Routines Users

Pro Plan — Maximize the Default Effort Level Upgrade to high and /resume Speed Improvements

• The default effort level was raised from medium to high (v2.1.117, 2026-04-22), so you can expect higher quality output from the same prompts Source
• Large session loading with /resume has been sped up by up to 67% (v2.1.116, 2026-04-20), making it easier to resume long-term projects Source
• Use the desktop app's integrated terminal and diff viewer to reduce the number of times you switch between browser and terminal

Max Plan — The Four-Feature Combination of Opus 4.7 + xhigh + Auto Mode + /ultrareview

• When tackling high-difficulty implementation tasks with the Opus 4.7 + xhigh mode combination, starting with /effort xhigh is officially recommended
• Auto mode is now easy to enable. Enable it in "Settings → Claude Code" and let it autonomously run repetitive tasks like CI fixes and dependency updates to reduce manual effort
• Integrating /ultrareview as a final check before submitting a PR can reduce reviewer burden and catch bugs early

For Those Who Want to Use Routines — Start with Small Scheduled Tasks and Have a Manual Fallback Ready

• Start by trying a small scheduled task at claude.ai/code/routines (e.g., post yesterday's diff summary to Slack at 9 AM)
• When automating PR reviews with GitHub triggers, always set filter conditions (such as is draft = false) to avoid unnecessary executions
• Since Routines are in research preview, we recommend having a manual fallback for critical workflows in case of specification changes

Sources (Primary Information)

The following primary sources were directly referenced in creating this article. Always check the links below for the latest accurate information.

• Claude Code Official Changelog — All changes in v2.1.111 through v2.1.117 (accessed: 2026-04-23)
• Use Claude Code Desktop — Anthropic Docs — Authoritative source for desktop app features
• Automate work with routines — Anthropic Docs — Official documentation for Routines
• Schedule recurring tasks in Claude Code Desktop — Anthropic Docs — Includes comparison with local scheduled tasks
• What's new in Claude Opus 4.7 — Anthropic Docs — Recommended use cases for the xhigh effort level