News
Visual Studio Code 1.112 Adds Integrated Browser Debugging, More Copilot CLI Control
Microsoft shipped Visual Studio Code 1.112, delivering a broad set of updates centered on agent autonomy, security, and developer workflow integration. The release -- following the new weekly cadence -- extends Copilot CLI with new permission tiers, introduces sandboxing for locally running Model Context Protocol servers, adds full debugging support to the integrated browser, and expands how agent customizations are discovered in monorepo setups.
The update continues a pattern from recent VS Code releases of deepening integration between the editor and GitHub Copilot's agentic features, giving developers more control over how much autonomy agents are allowed -- and new tools for diagnosing when they go wrong (see full March 18 release notes).
Copilot CLI Gets Autonomous Mode and Steering Controls
The most prominent agent-facing change in 1.112 is a restructured permissions model for Copilot CLI sessions. Three levels are now available: Default Permissions, which uses existing approval settings; Bypass Approvals, which auto-approves all tool calls without confirmation dialogs; and Autopilot, which "auto-approves all tool calls, auto-responds to questions, and continues working autonomously until the task is complete." Autopilot is enabled by default in VS Code Insiders builds and can be toggled via the chat.autopilot.enabled setting (see "Hands On with VS Code 1.111 Autopilot (Preview) for AI Prompts").
[Click on image for larger view.] New Autopilot Preview Input from VS Code 1.111 (source: Microsoft).
Also new in Copilot CLI sessions is message steering and queueing, a feature previously available only in local agent sessions in the Chat view. Users can now send follow-up messages while a request is still running, either to redirect the agent mid-task or to queue up the next instruction. A related improvement lets users preview uncommitted workspace changes directly in the Chat view before delegating a task to Copilot CLI, removing the need to switch to the Source Control panel to review those changes first.
[Click on image for larger view.] Preview Changes Before Delegating to Copilot CLI (source: Microsoft).
Terminal output from Copilot CLI now includes clickable file links for paths under the ~/.copilot/session-state/ directory, which the built-in link detector previously did not resolve. The feature handles both absolute and relative paths and is enabled by default via the github.copilot.chat.cli.terminalLinks.enabled setting.
New Agent Diagnostics and Richer Context Support
Version 1.112 adds a /troubleshoot skill in preview that reads JSONL debug log files from the active chat session and surfaces analysis directly in the conversation. The skill can explain why tools or subagents were used or skipped, why instructions or skills did not load, what caused slow response times, and whether network issues occurred. Using /troubleshoot requires enabling two settings: github.copilot.chat.agentDebugLog.enabled and github.copilot.chat.agentDebugLog.fileLogging.enabled. Microsoft's debug chat interactions documentation covers the full setup.
Also in preview, agent debug logs can now be exported and imported, allowing developers to share session diagnostics with teammates or analyze them offline. The release notes specify that importing files larger than 50 MB triggers a warning dialog. Agents also gain the ability to natively read image files and binary files from disk -- binary files are presented to the agent in hexdump format. When an agent produces image output, such as a screenshot from the integrated browser, those images can be browsed in a new image carousel view enabled via the experimental chat.imageCarousel.enabled setting.
A smaller quality-of-life addition: pasting a copied symbol -- such as a class name or method -- into the Chat input now automatically converts it to a #sym:Name reference, giving the agent immediate context about the symbol. Developers who want to paste the symbol as plain text can use Ctrl+Shift+V (Cmd+Shift+V on macOS).
MCP Sandboxing and Plugin Management Improvements
On the security front, locally configured stdio MCP servers can now be run in a sandboxed environment on macOS and Linux, restricting their access to the file system and network. Sandboxing is opt-in and is enabled by setting "sandboxEnabled": true in the relevant mcp.json file. When a sandboxed server requests access to an additional folder or domain, VS Code prompts for the user's approval and updates the sandbox configuration. The release notes are explicit that sandboxing "is not currently available on Windows," though remote scenarios such as WSL and SSH continue to work.
MCP servers and plugins can now be enabled or disabled without uninstalling them, both globally and per-workspace, via the Extensions view or the Chat: Open Customizations view. Plugins also gain support for automatic updates driven by the existing extensions.autoCheckUpdates setting; plugins sourced from npm and PyPI require explicit user approval before updating, given that "updating these plugins might cause new code to be run on your machine."
Monorepo Support for Agent Customizations
A new chat.useCustomizationsInParentRepositories setting enables VS Code to discover agent customization files from parent directories up to the repository root, addressing a gap for developers who open subfolders or packages within a monorepo. The feature applies to all customization types, including always-on instruction files such as copilot-instructions.md, AGENTS.md, and CLAUDE.md, as well as instruction files, prompt files, custom agents, skills, and hooks. Discovery activates only when the open workspace folder is not itself a Git repository, a parent folder contains the .git directory, and the parent repository is trusted via VS Code's workspace trust mechanism. Microsoft's customize AI documentation covers the full range of customization options.
Integrated Browser Now Supports Full Debugging
The integrated browser -- which lets developers open web applications directly inside VS Code -- gains a new editor-browser debug type in 1.112, enabling breakpoints, code stepping, and variable inspection without leaving the editor. This builds on functionality introduced earlier this month (see "VS Code 1.110 Ships with Agent Plugins, Browser Tools and Session Memory"). The release notes describe migration from existing configurations as straightforward: "Most options from existing msedge and chrome debug configurations are supported, which makes migration often as simple as changing the type of an existing configuration in launch.json."
[Click on image for larger view.] VS Code Insiders 1.110: Native Browser Integration in Chat. (source: Ramel).
Additional integrated browser improvements in this release include right-click context menus exposing options such as copy, paste, open in new tab, and inspect; an independent zoom level separate from VS Code's window zoom, controllable with Ctrl+=, Ctrl+-, and Ctrl+Numpad0; and per-website zoom memory. A default zoom level can be set via the workbench.browser.pageZoom setting.
Edit Mode Deprecation Timeline
The release notes confirm that Edit Mode, deprecated as of VS Code 1.110, remains temporarily re-enableable via the chat.editMode.hidden setting through version 1.125. Beginning with 1.125, "Edit Mode will be fully removed and can no longer be enabled via settings." Developers relying on Edit Mode workflows should plan their transition accordingly.
Other changes in 1.112 include an optional editor.find.closeOnResult setting that automatically closes the Find control after a match and returns focus to the editor (disabled by default), and an improved IME composition experience in the terminal that constrains the preview text to the visible terminal viewport rather than allowing it to overflow at the right edge.
About the Author
David Ramel is an editor and writer at Converge 360.