Anthropic Releases Major Claude Code 2.1.1 Update

This is a comprehensive update to Claude Code, with massive improvements across three key areas: CLI, feature flags, and prompt behavior.

🛠️ CLI Updates (Command Line Tool) — 109 Changes

This is the most significant part of the Claude Code update! The CLI is the part where developers interact with Claude Code in the terminal. Updates include:

📌 Feature Enhancements

• Skills Support Automatic Hot Reload

  • No more restarts needed! When you add new skills or modify existing ones in the skills folder (~/.claude/skills), Claude Code will automatically detect them.

• Support for Sub-Agent Execution of Skills/Commands

  • You can spawn a “sub-Claude” for a task that runs in its own independent context with Claude Code, such as running a script without interfering with the main task.

• Set Claude Code Language

  • You can set language: "japanese" to make Claude Code respond in Japanese, perfect for multilingual development.

• Terminal Compatibility Improvements

  • Now in mainstream terminals like iTerm2, WezTerm, and Kitty, Shift+Enter works with Claude Code without manual configuration!

• Slash Commands Are More User-Friendly

  • After typing /, Claude Code will auto-complete commands, so you don’t always need to start with / for it to work.

• Image Paste Support

  • You can directly paste screenshots with Ctrl+V into the command line with Claude Code—iTerm2 now supports this!

🔒 Security and Permissions

• Fixed Sensitive Information Leakage Issues

  • Previously, debug logs could leak API keys, OAuth tokens, etc. This has been fixed in Claude Code.

• New Environment Variable to Hide Account Information

  • If you’re recording a demo, you can set CLAUDE_CODE_HIDE_ACCOUNT_INFO so Claude Code won’t display your email and organization name.

• More Powerful Permission System

  • Claude Code now allows more granular control over which tools Claude can use, and you can define which skills/commands can access which files.

• Support for .gitignore to Control File Selection Behavior

  • Claude Code automatically recognizes which files to ignore, such as .env and node_modules files you don’t want to upload.

⚡ Performance and Reliability Improvements

• Faster startup times for Claude Code, especially when using Bun or macOS installer
• Optimized bash command parsing in Claude Code, fixing various edge cases
• Improved interaction experience with plugins and VSCode for Claude Code
• More stable paste, image processing, and large notebook file handling in Claude Code

🧠 Task and Skills Experience Optimization

• New /plan command lets you quickly enter “Plan Mode” in Claude Code
• Support for task backgrounding (Ctrl + B) allows you to send all running tasks to the background with Claude Code and continue with other work
• Task output management optimization: Claude Code stores results in files, so you don’t have to keep watching the terminal waiting for Claude’s output

🚩 Flags (Feature Toggles) Changes

These are additions and removals of internal experimental features or configuration options. While not directly user-facing, they can affect Claude Code behavior.

✅ 3 New Additions

• Support for searching unsupported models in Claude Code
• Guidance and upsell (upgrade recommendations) prompts on VSCode for Claude Code

❌ 8 Old Flags Removed

For example:

• tengu_bash_command_backgrounded
• tengu_spinner_words
• doorbell_bottle (unknown purpose, but appears to be an experimental feature)

💬 Prompt Updates — Smarter Claude Code!

This section adjusts how Claude Code “speaks” in the command line, focusing on making it more natural and logical.

🧠 Detailed Prompt Updates (10 Items)

1. No Longer Forced to Use Official Documentation to Answer Questions

   • Previously, Claude Code was required to call claude-code-guide to answer “How to use Claude Code.” Now the restriction is relaxed, allowing Claude Code to answer based on its own knowledge.

2. No Colon After Sentences

   • For example: Previously, Claude Code would say “Let me read the file:” then execute a tool, but when the tool might not display, the colon felt awkward. Now it’s changed to “Let me read the file.” which is more natural.

3. More Logical Questioning in Plan Mode

   • AskUserQuestion is only used to clarify requirements and collect information in Claude Code
   • No longer used to ask “Is the plan OK?”
   • “Should I execute?” type questions are asked by ExitPlanMode in Claude Code, making it clearer

4. Simplified Git Commit Messages

   • Commits no longer include “🤖 Generated with Claude Code”, but still retain Co-Authored-By: Claude Sonnet 4.5

5. Removed LSP Tools (code navigation, definition lookup, etc.)

   • Claude Code now uses search/read tools to achieve similar functionality, likely for compatibility and system simplification

6. Avoid Duplicate Calls to Already Loaded Skills

   • Claude Code prevents redundant skill invocations

7. Enhanced Task Execution Structure Requirements

   • TaskOutput must provide fields like block, timeout, etc., in Claude Code to avoid inconsistent behavior

8. Optimized Background Bash Execution Strategy

   • Claude Code doesn’t need immediate output monitoring; Claude will notify you when execution completes later

How to Upgrade

If you previously installed Claude Code CLI through the official recommended method, you can try the following command to upgrade directly:

claude upgrade

Conclusion

The Claude Code 2.1.1 update brings significant improvements across CLI functionality, security, performance, and user experience. With 109 CLI changes, enhanced permission systems, and smarter prompt behavior, this update makes Claude Code more powerful and user-friendly than ever before. Whether you’re using skills, working with sub-agents, or managing complex development tasks, these improvements will enhance your workflow with Claude Code.