Anthropic has shipped a big update to the Claude Code extension for Visual Studio Code. The chat interface is still the heart of it, but almost everything around it has been redesigned. Here is what is new.
Session history
The history panel now lets you search past sessions and switch between them. You can also pull in sessions from the Claude web app or the mobile app directly into VS Code. That is a genuinely useful addition if you start something on your phone or in the browser and want to pick it up in your editor. You can also rename sessions and delete ones you no longer need.
The plus menu
Clicking the plus button next to the input gives you three options: upload a file from your computer, add context (which means pulling in files already in your project), or let Claude browse the web.
The slash command menu
All the old slash commands now live in a proper menu. From there you can toggle extended thinking on or off, check your account and usage, switch to fast mode for Opus 4.6, switch models, and customise other settings. No more memorising commands.
Effort control
You can now set the effort level directly from the settings panel. The options are low, medium, high, and max. Useful when you want a quick answer and do not need Claude to think hard about it, or the opposite.
Edit modes
Claude can handle your code in three ways. Ask before edits means it checks with you before changing anything. Edit automatically means it just gets on with it. Plan mode means it maps out what it intends to do first, then waits for you to confirm. You switch between them from the bottom right of the chat window.
Implementation plans as documents
When Claude creates a plan, it no longer dumps it into the chat as a wall of text. It opens as a proper document in the main editor window. That means you can read it properly, scroll through it, and refer back to it without losing your place in the conversation. A small change that makes a big difference.
Branching
This is the standout feature. When you hover over any message you have sent, a small back arrow appears. Click it and you get three options: fork the conversation from that point, rewind the code to that point, or do both at once. In practice this means you can go back to any earlier message, try a different approach, and keep the original thread intact. Anyone who has ever painted themselves into a corner with an AI coding session will immediately see the value.