Claude Code v2.1.236 Changelog

This update is all about giving you more control over how the AI starts up and how different tasks talk to each other, while also making the app much more stable and less glitchy. It matters most to developers who use Claude Code heavily for long, automated work sessions.

What's new

• Default model setting: You can now use a system environment variable (a setting outside the app) called ANTHROPIC_DEFAULT_MODEL to choose which AI model new sessions start with. If you manually pick a different model later using the /model menu, your manual choice will save and override the default next time.
• Cross-session chat: If you have multiple Claude Code windows open, they can now talk to each other. One session can ask another to send a single notification when it finishes its work and goes idle. This is an opt-in feature, meaning it only happens if you specifically ask for it, and it doesn't constantly check in (which saves computer power).
• Smarter goal check-ins: If you give the AI a long-term goal (/goal) but it's stuck waiting on a background task, it will now automatically check in after 30 minutes (then every 1 or 2 hours) instead of just freezing until you come back.
• Screen reader support in VS Code: For users who are visually impaired, the VS Code extension can now read out loud what the AI is doing, including replies, errors, and permission requests.

Improvements

• Faster startup: The app launches a bit faster because it now counts your past sessions in the background instead of making you wait for it.
• Smarter "auto mode": When you let the AI run commands automatically, it used to follow special "Monitor" rules. Now, it treats those commands just like normal terminal commands, making it safer. It also got better at not being fooled by hidden Git settings that lie about whether your project folder is clean.
• Better auto mode on other platforms: If you use Claude through other cloud providers (like Bedrock or Vertex AI), the AI's automatic decision-making now uses the same smart scoring system as the main Claude app.
• Better usage tracking: The /usage menu now shows how many usage credits you've spent if you're on a Team or Enterprise plan, and it safely caps the display at 0% before you spend anything.
• Typo protection: If you type a slash command (like /help) and make a typo, the app will tell you it doesn't exist instead of guessing and running a totally different command.
• Faster offline detection: If you are using Remote Control and close your terminal, the system marks your session as offline within seconds.
• Message limits: If you send a rapid burst of messages to another session, it will now stop accepting them upfront if the inbox is full, rather than letting them get silently deleted.
• Cleaner UI: The text at the bottom of the screen (like your goal or background agent status) and the session title chip are now neatly aligned with the edges of the prompt box.
• Better model picker: The menu that lets you pick an AI model now only highlights the newest model, making it easier to spot new releases.

Fixes

• Crash when folders are deleted: Fixed an issue where deleting a folder you were working in broke basic features like copying text, background tasks, and logs.
• Full-screen crashes: Fixed a bug where full-screen mode would break forever after a single failed startup. It now safely falls back to the classic view instead of crashing every time you open it.
• Menu sizing: Fixed the model picker menu so it no longer grows taller than your actual window; it now scrolls if there are too many models.
• Message glitches: Fixed cross-session messages being rejected if the message text had a broken formatting tag.
• Subprocess errors: Fixed a crash that happened when a background program (like PowerShell on Windows) failed to start.
• Missing messages: Fixed a bug in full-screen mode where a newly sent message wouldn't show up until you resized the window.
• Visual bugs: Fixed a blank line that got stuck above the prompt, and fixed panes not redrawing correctly when you resized the window away and back in full-screen mode.
• Security prompt bug: Fixed an issue where a security approval prompt wouldn't show up at startup, but would secretly capture your first keystroke as a "yes."
• Jumping tab titles: Fixed terminal tab titles flickering every 960 milliseconds in tmux (a terminal multiplexer). It now only updates the title when the text actually changes.
• Unclear errors: Fixed a confusing error message that appeared when the list of cloud environments came back empty or broken.
• Auto-select timeout: Fixed a prompt that would automatically pick a fallback model after 60 seconds of silence when using Remote Control.
• Broken tips: Fixed an issue where helpful spinner tips never showed up because of a broken guest-pass file.
• Hot-reload errors: Fixed an error that popped up every time a "skill" changed if the folder you were working in had been deleted.
• Runner overlap: Fixed a bug where a background runner session would sometimes start up on another server before its cleanup process was totally finished.
• Mascot glitch: Fixed the Clawd mascot's eyes and feet rendering unevenly in the iTerm2 terminal at certain font sizes.
• Runaway text: Fixed an issue where automatic session recaps could go on forever. They are now capped at 400 characters and cut neatly at a word boundary.

Breaking changes

No breaking changes in this release.