Agent Programming and Control

GitHub Copilot agents are now practical for longer-running, complex tasks with new automation capabilities. Developers can use hooks to run custom code at key agent lifecycle events—enforcing policies, auto-linting, or blocking commands before execution. The release adds fork-a-conversation functionality, allowing you to branch from any checkpoint and explore alternative paths without losing your starting point.

Chat-based controls give you finer-grained steering: toggle global auto-approval with /autoApprove or /yolo commands, and queue follow-up messages while agents are working to redirect their approach without waiting for completion.

Extending Agent Capabilities

Agents are now extensible through Agent plugins—prepackaged bundles of skills, tools, hooks, and MCP servers installable from the Extensions view (currently experimental). Agent skills can be triggered directly from chat as slash commands, and new agentic browser tools let agents drive the integrated browser to navigate, click, screenshot, and verify changes (experimental). Developers can also generate reusable prompts, skills, agents, and hooks directly from conversations using /create-* commands.

Additionally, Copilot CLI is now natively integrated into VS Code with diff tabs, trusted folder sync, and right-click code snippet sharing.

Context and Memory Management

A major improvement is shared agent memory—context now persists across the Copilot coding agent, Copilot CLI, and code review so knowledge accumulates over time. Plans persist across conversation turns and through compaction, preventing the agent from starting over. A new Explore subagent delegates parallelized codebase research to lightweight models, enabling the Plan agent to reference specific files and paths.

Context management is smarter: context compaction automatically triggers when the window fills, and you can manually invoke it with /compact. Large tool outputs are now written to disk instead of stuffed into context, preserving important details. Long-distance next edit suggestions also predict edits anywhere in your file.

Additional Improvements

The release includes related VS Code productivity updates: Kitty graphics protocol for high-fidelity terminal images, a redesigned model picker with search and rich hover details, improved chat accessibility for screen readers and keyboard navigation, and AI coauthor attribution for commits containing AI-generated code.