The Visual Studio Code February 2026 release makes agents practical for longer-running and more complex tasks. This gives you more control over how they run, new ways to extend what they can do, and smarter session management that keeps context intact.

Program your agents

  • Automate with hooks: Run your code at key agent lifecycle events to enforce policies, auto-lint, or block commands before they execute.
  • Fork a conversation: Branch from any checkpoint to explore an alternative path without losing your starting point.
  • Auto-approve from chat: Toggle global auto approval with /autoApprove or /yolo directly in the chat input and pair it with terminal sandboxing to let agents run within boundaries.
  • Queue and steer from chat: Send follow-up messages while the agent is working to redirect its approach without waiting for it to finish.

Extend your agents

  • Agent plugins: Install prepackaged bundles of skills, tools, hooks, and MCP servers from the Extensions view. This is currently available as an /experimental feature.
  • Use skills as slash commands: Trigger agent skills from chat, including ones contributed by your extensions.
  • Agentic browser tools: Let agents drive the integrated browser to navigate, click, screenshot, and verify their own changes. This is currently available as an /experimental feature.
  • Create customizations from chat: Generate reusable prompts, skills, agents, and hooks directly from a conversation with /create-* commands.

In addition, Copilot CLI is now included in VS Code with native support including diff tabs, trusted folder sync, and right-click to send code snippets.

Give agents the right context

  • Share agent memory: Agents share and store knowledge across the Copilot coding agent, Copilot CLI, and code review so context builds up over time.
  • Plan memory: Plans persist across turns and through compaction, so the agent builds on existing work instead of starting over.
  • Built-in Explore subagent: Delegates fast, parallelized codebase research to lightweight models so the Plan agent references specific files and code paths.
  • Context compaction: Copilot automatically compacts conversation history when the context window fills up. You can now trigger compaction manually with /compact, and guide what to keep (e.g., /compact forget about all variants, except the rust version).
  • Handle large outputs: Large tool output is written to disk instead of stuffed into context, so important details aren’t lost during compaction.

Also, long-distance next edit suggestions predict edits anywhere in your file.

  • Kitty graphics protocol for high-fidelity images in the terminal.
  • Redesigned model picker with search, sections, and rich hover details.
  • Chat accessibility improvements for screen readers, keyboard navigation, and notification signals.
  • AI coauthor attribution for commits with AI-generated code.

Happy coding!

Join the discussion within GitHub Community.