Claude Code v2.1.210 Changelog

This update is all about making Claude Code—the command-line tool that helps you write and manage code—much more reliable, secure, and user-friendly. It focuses on fixing annoying bugs where the tool would freeze, crash, or get confused, while also making sure background tasks (called "agents") work smoothly without messing up your main project. Whether you are using it to write quick scripts or manage complex codebases, this release makes the whole experience feel much more stable.

What's new

• Live timer for long tasks: When Claude is running a tool that takes a while, you will now see a live, ticking timer. This prevents the tool from looking like it froze or got stuck.
• New permission warnings: When you start up, you will get a warning if you are using old permission rules like Write, NotebookEdit, or Glob. The system now wants you to use Edit or Read instead.
• Better screen reader support: If you use a screen reader, it will now read aloud when you cycle through permission modes using Shift+Tab.
• Waiting agent counter: The footer at the bottom of your screen now shows exactly how many background agents are waiting for your input, and it flashes a color briefly when that number changes.

Improvements

• Smarter timeout messages: When a terminal command takes too long and is sent to the background, the tool now explains this clearly to Claude. This helps the AI understand that the command timed out rather than assuming you asked to run it in the background.
• Better security against "prompt injection": The system is now much safer against indirect prompt injection—a security trick where malicious text hidden inside a file tries to hijack the AI's instructions.
• Smarter auto-permissions: The system that decides what Claude is allowed to do now defaults to using the smarter Claude 3.5 Sonnet model for external sessions, making it more accurate.
• Color-blind friendly charts: The built-in data visualization tool now uses advanced color science to check chart colors, making sure they are easy to read for color-blind users.
• Clearer memory errors: If Claude's memory file (MEMORY.md) gets too large, it will now show a clear error instead of secretly cutting off your text.

Fixes

• Isolated code safety: Fixed a bug where subagents working in an isolated "worktree" (a separate folder for testing code) could accidentally make changes to your main Git repository.
• No accidental "ultracode" triggers: The high-effort ultracode mode will no longer trigger on automated inputs like webhooks (automated web messages) or pull request comments. It now only triggers when a real human types it.
• No more stray characters: Fixed an issue where weird characters (like È or É) would randomly appear around pasted text when opening external text editors.
• Smoother session connecting: Fixed a bug where trying to attach to an active session with claude attach would fail with "job not found" errors. It now waits patiently for the background service to start up.
• Crash fixes: Fixed several crashes, including when a tool returned a giant number (a "bigint") instead of a visual element, and when background services disconnected.
• No more fake "user edits": Fixed an issue where approving a plan without making any changes would wrongly label it as "(edited by user)" and overwrite your files with outdated data.
• Cleaned up leftover files: Fixed a bug on Windows and network drives where temporary files were left behind if saving the plugin cache failed.
• Git lock cleanup: If a background session is forced to close, the system will now automatically clean up and unlock any locked Git files left behind.
• Temporary Fable disable: The Fable advisor is temporarily turned off while the team fixes a server-side issue causing it to fail.

Breaking changes

No breaking changes in this release.