Claude Code v2.1.207 Changelog

This release is all about making Claude Code smoother, safer, and more reliable. If you use Claude to write code, run commands automatically, or connect to cloud platforms like AWS and Google Cloud, this update fixes several annoying bugs that caused screen freezes, connection drops, and security false alarms. It also changes how some settings are saved to keep your projects more secure.

What's new

• Auto Mode is now on by default: If you use Claude through cloud platforms like Amazon Bedrock, Google Vertex AI, or Foundry, "Auto Mode" (which lets Claude run commands and edit files automatically) is now enabled right away. You can still turn it off in your settings if you prefer.
• New default AI model: If you connect Claude to AWS or Google Cloud, it will now default to using the more powerful Claude Opus 4.8 model.

Improvements

• Cleaner chat history: If you paste the exact same block of text into the chat multiple times, Claude will now group them under a single expandable label instead of cluttering your screen.
• Better waiting screens: When a task is paused and waiting for your input, the screen will now clearly show the question it's asking and a simple timer (like "waiting 3m") so you know exactly how long it has been stalled.

Fixes

• No more laggy typing: Fixed a bug where your terminal would freeze or your typing would lag when Claude was writing out very long lists, tables, or code blocks.
• Fewer annoying permission prompts: Claude will no longer ask for permission to run simple folder-changing commands (cd) when the output is just being hidden.
• Smoother scrolling: Fixed a glitch where the chat window would suddenly jump up to the top of a long answer right when Claude finished typing it.
• Fewer false alarms: Fixed annoying, fake security warnings (prompt-injection warnings) that were triggered by normal, safe system updates.
• Windows freeze fix: On Windows, if Claude gets stuck trying to log into AWS (Amazon's cloud service), it will now give up after 60 seconds and let you know, instead of freezing forever.
• Fewer login prompts on AWS: Fixed an annoying bug where Claude would constantly ask you to log in to AWS for every single question you asked.
• Clearer research sources: When Claude is doing deep research, it will now show the actual website or server name it's gathering data from, instead of just labeling everything as "unknown."
• Stable remote control: If your internet drops while controlling Claude remotely, it will no longer lose track of what it was doing once the connection comes back.
• Better mobile and web tracking: If you are monitoring Claude from your phone or a web browser, you will now correctly see the progress of background tasks running on your computer.
• Restoring background work: Fixed an issue where reopening a background task that was working in a specific Git folder (a tool used to track code changes) would sometimes load as a blank screen.
• Background task names: Background tasks that were automatically named will now actually show their correct names in your dashboard.
• Crash loop fix: Fixed a severe bug where a corrupted message between AI "agents" working together would cause the program to crash repeatedly every second.
• Better file searching: Fixed a bug where typos in file-matching rules (like brackets) would completely break Claude's ability to read files or suggest code.
• Git cleanup: Fixed a bug where Claude left behind messy configuration files in Git, which was breaking other developer tools.
• Respecting custom setups: The automatic updater will no longer accidentally overwrite your custom shortcuts or startup scripts for Claude.
• Security consent fix: Fixed an issue where running Claude in the