Visual Studio Code 1.110: Agent‑Powered Chat, Browser Automation, and a Whole Lot More
When VS Code rolled out version 1.110 last month, the update didn’t just tweak the UI—it rewrote how you can get work done with AI built into the editor. The new release turns agents from a novelty into a practical tool that can browse your app, run terminal commands, and even remember long‑running plans without losing context.
Agents Get Control Panels and Debugging
The Agent Debug panel lets you watch every tool call in real time. After a recent project where an agent repeatedly opened the wrong file before correcting itself, the new logs revealed that the “openFile” skill was being triggered twice by a misconfigured hook. Seeing the event sequence made it trivial to adjust the hook and prevent another accidental edit.
Auto‑Approve: Convenience or Risk?
A big addition is the /autoApprove slash command, which skips all tool confirmation prompts. This can shave minutes off a multi‑step build script, but in one scenario an agent ran a destructive npm uninstall without waiting for a manual check. The release warns that enabling global auto‑approve should only happen when you’re confident every step is safe.
Background and Claude Agents Refreshed
Background agents now support slash commands like /compact to manually trim conversation history, freeing up the model’s context window. Claude agents get a new /getDiagnostics tool that pulls editor warnings straight into the chat, letting the agent suggest fixes before you touch code.
Browser Automation Inside VS Code
If you’ve ever had to switch windows to test a front‑end change, the integrated browser tools are a game changer. By enabling workbench.browser.enableChatTools, an agent can open a page, click elements, and even take screenshots—all inside the editor. In practice, a developer used these tools to confirm that a CSS tweak applied correctly without leaving VS Code.
Customizing Agents on the Fly
The /create‑skill command lets you package a multi‑turn workflow into a reusable skill. After debugging a complex API call over several turns, a user ran /create-skill and got a ready‑to‑use package that now automates the same sequence with a single agent message.
Context Compaction Made Simple
Long conversations can bloat an agent’s context window. VS Code 1.110 compacts automatically when the limit is hit, but you can also trigger it manually by typing /compact or selecting “Compact Conversation” from the chat bar. One user found that compacting after a week‑long refactor session kept the model focused on the latest changes without losing earlier decisions.
Forking Sessions to Explore Alternatives
You can now fork a chat with /fork, creating an independent copy of the conversation history. This is handy when experimenting with different code styles: keep the original plan intact while you try a more concise solution in the forked session.
Accessibility Improvements for Screen Readers
The new accessibility skill ensures that any custom agent or tool respects screen‑reader guidelines. Users who rely on Alt+N and Alt+P to navigate question carousels will notice fewer interruptions when an agent prompts for input, as notifications are now properly announced.
Terminal Sandbox Preview
If you need isolation, the terminal sandbox lets agents run shell commands in a controlled environment. Enabling chat.tools.terminal.sandbox.enabled keeps file system access limited to a whitelisted set of folders and blocks network requests to untrusted domains.
With all these changes, VS Code 1.110 turns the editor into a full‑stack assistant that can reason, browse, edit, and even debug—all without leaving your IDE.
