Program Your Agents for Control and Automation

The February release makes GitHub Copilot agents far more practical for complex, long-running tasks. Developers can now run custom code at key agent lifecycle events using hooks—useful for enforcing policies, auto-linting code, or blocking unsafe commands before execution. A new /autoApprove command (with a /yolo shorthand) enables auto-execution paired with terminal sandboxing for bounded autonomy.

Conversation forking lets you branch from any checkpoint to explore alternative approaches without losing context. Developers can also send follow-up messages while agents are working via the /queue feature, redirecting their approach without waiting for completion.

Extend Agents with Plugins and Browser Automation

A new Agent plugins system (currently experimental) lets developers install prepackaged bundles of skills, tools, hooks, and MCP servers directly from the Extensions view. Skills can now be triggered as slash commands in chat, and developers can generate reusable skills, custom agents, and hooks directly from conversations using /create-* commands.

Most remarkably, agents can now drive an integrated browser to autonomously navigate pages, click elements, screenshot, and verify changes—all available as an experimental feature. The Copilot CLI is now natively integrated into VS Code with diff tabs and trusted folder sync.

Better Context Management Across Tools

Agents now share memory across Copilot's coding agent, CLI, and code review, allowing context to persist and build over time. Plan memory persists across conversation turns and automatic compaction, so agents build incrementally rather than restarting. A new Explore subagent delegates fast, parallelized codebase research to lightweight models, ensuring the main agent has specific file references.

Context compaction now runs automatically when the context window fills, with manual triggering available via /compact and selective guidance (e.g., /compact forget about all variants, except the rust version). Large tool outputs are written to disk rather than stuffed into context, preventing information loss during compaction. The release also adds long-distance next edit suggestions that predict edits anywhere in your file.

Additional Improvements

Related VS Code updates include Kitty graphics protocol for terminal image fidelity, a redesigned model picker with search and hover details, improved chat accessibility for screen readers and keyboard navigation, and AI coauthor attribution for commits containing AI-generated code.