Claude Code New Updates: Better Desktop App, Simple Mode, In-built Worktrees, New Hooks & MORE!

Claude Code: Recent Updates & Deep Dive

Key Concepts:

• Claude Code: An AI-powered coding assistant developed by Anthropic.
• CLI: Command Line Interface – the text-based way to interact with Claude Code.
• MCP (Multi-Code Provider): Tools allowing Claude Code to interact with external services (Google Drive, Jira, Slack, etc.).
• Worktree Isolation: Running multiple Claude Code sessions on the same repository without code conflicts.
• Automemory: Automatic recording and recall of learnings and patterns across sessions.
• Opus 4.6: The latest Claude model, featuring a significantly expanded context window.
• LSP (Language Server Protocol): A standard protocol for communication between IDEs and language servers, enabling features like autocompletion and error checking.
• DV Server: Development Server - used to preview running applications.

1. CLI Updates & Enhancements

Recent updates to the Claude Code CLI focus on stability, control, and security. The most significant addition is Git Worktree Isolation enabled via the -w or --worktree flag. This allows multiple Claude Code sessions on the same repository to operate independently, preventing code edit conflicts. Subagents now also support this isolation through isolation: worktree in agent definitions, replacing the need for external tools like Ralphie.

Background Agent Control has been improved with a Control + F key binding (double-press within 3 seconds) to kill all background agents. Agent definitions can now specify background: true to ensure agents continue running even when the main thread is cancelled (via Escape key). Previously, background agents were silently ignored when the main thread was idle, a problem now resolved.

A new Config Change Hook Event has been introduced, primarily for enterprise security. This allows blocking configuration changes that violate security policies using exit code 2 or a JSON decision. Policy settings themselves cannot be blocked.

Simple Mode has been upgraded to include the file edit tool alongside the bash tool, enhancing its usability while maintaining its lightweight nature (disabling MCP tools, attachments, hooks, and claw.md loading).

LSP Server Startup Timeout configuration is now available, allowing users to specify a timeout (in milliseconds) for LSP server initialization. This addresses issues with slower servers like the Java language server timing out and failing to provide type information.

2. Model Upgrade: Sonnet 4.6 & Opus 4.6

Sonnet 4.6 has officially replaced Sonnet 4.5 as the default model, offering a 1 million token context window. Initial access issues following an auto-update have been resolved. Numerous stability improvements have been implemented, including fixes for sessions becoming invisible with symlinks, data loss on SSH disconnect (now flushing session data during graceful shutdown), excessive claw.json.backup file creation, and unbounded memory usage for shell commands with large output. Startup performance has also been improved by removing eager loading of session history.

Opus 4.6, released in February, boasts a 200,000 token context window (with 1 million tokens in beta) and a maximum output of 128,000 tokens. Its key feature is adaptive thinking, allowing Claude to dynamically adjust its approach to problem-solving. Anthropic reports Opus 4.6 can sustain work for up to 14 hours at a 50% success rate, enabling it to handle significantly larger and more complex projects. The value proposition of Opus 4.6 is particularly strong for users on the $200 max plan.

3. Claude Code Desktop: A Developing IDE

The Claude Code Desktop is evolving into a more comprehensive development environment. Key additions include:

• App Preview: A built-in DV server allows real-time previewing of running applications directly within the desktop interface. It automatically detects and iterates on console log errors.
• Code Review Panel: A dedicated panel for reviewing changes made by Claude, with inline comments and diff viewing, resembling a mini pull request experience.
• GitHub Integration: The desktop can monitor pull requests, automatically attempt to fix CI failures, and even automerge PRs upon successful completion of checks.
• Session Mobility: Seamless transition of sessions between the CLI, desktop, and web interfaces, maintaining full context. Use of the /desktop command facilitates movement from CLI to Desktop.

4. Automemory: Persistent Learning

Automemory automatically records and recalls learnings across sessions. This data is stored in ~/.cloud/projects/<project_name>/memory. Claude remembers project-specific testing patterns and coding styles, applying this knowledge to future sessions. A 200-line limit is imposed on the memory file, and the feature can be disabled via an environment variable. This functionality replicates what users previously achieved with custom claw.md files, but in an automated and integrated manner.

5. Security & Tooling Improvements

Claude Code Security (currently in limited research preview) offers automated security analysis using semantic reasoning, identifying complex vulnerabilities that rule-based scanners (like Snike or SonarQube) might miss. All suggestions require human approval.

MCP Server Performance has been significantly improved with lazy loading, reducing context usage by up to 95%. MCP Tool Search allows Claude to efficiently access all configured MCP servers without excessive context consumption.

IDE Integration:

• VS Code: Plan previews now auto-update, commenting is enabled only when a plan is ready, and the preview remains open upon rejection. The permission destination chooser now persists choices across sessions.
• JetBrains: An official Claude Code plugin is available, featuring a diff viewer and LSP integration for real-time code understanding (type information, errors, navigation).

6. Pricing & Availability

Claude Code is now included with every Team Plan standard seat, making it more accessible to teams.

Notable Quote:

“Nobody else has this combination of terminal power, desktop features, multi-agent coordination, and tool integration.” – Speaker, highlighting Claude Code’s unique capabilities.

Conclusion:

Claude Code is rapidly maturing into a powerful and feature-rich AI coding assistant. The combination of the desktop app, worktree isolation, model upgrades (Opus 4.6), automemory, and improved tooling positions it as a leading solution for AI-assisted development. While token costs remain a consideration, Claude Code offers a compelling value proposition for developers seeking to enhance their productivity and code quality. The ongoing development and integration with existing tools suggest a promising future for the platform.