VS Code v1.110 - February 2026 Update
Introduction
Following on from the January release which brought Agent Skills GA and Copilot Memory, the February 2026 release of VS Code, v1.110, continues that same theme of deepening the agentic development experience. I use the Insiders build on my personal device in an attempt to keep up with what is coming to the stable release.
Session memory
Plans created by the Plan agent now persist to session memory and stay available across conversation turns. When you ask for refinements, the agent builds on the existing plan instead of starting from scratch.
The plan is also recalled after unrelated messages in the same session, so you can return to a plan without repeating context. During longer implementation work, the plan remains accessible in memory even when older conversation history is compacted to free up context space.
The concept of agent memory is probably the most interesting capability coming out of recent releases, whether persistent or session-scoped. Persistent memory carries context and guidance across sessions, while session memory ensures a plan stays consistent within a conversation without you needing to repeat yourself.
Native Browser Control for Agents
The integrated browser was introduced last month in v1.109. Some experimental tools have been added allowing agents to directly read and interact with that browser, take screenshots, click elements, and even run Playwright code. This is no longer just a passive viewer.
For the kind of work I do around infrastructure and DevOps, the immediate use case I can see is having an agent verify that a deployment is working by loading the application URL, checking for expected content, and reporting back. I have yet to test this but hopefully my hopes will be validated.
Agent Plugins
Where January gave us skills as part of VS Code, February takes the next step by allowing you to install bundles of chat customisations, skills, tools, commands, hooks, and more directly from the Extensions view.
The practical difference is discoverability. Skills required you to know they existed and drop them into the right folder. Plugins surface through the Extensions marketplace with copilot-plugins and awesome-copilot repos present by default, or through private sources, local paths, and Git repositories if your organisation wants to host its own internal plugins without publishing them publicly. The setting to enable them is chat.plugins.enabled and the feature is still experimental, but the direction of travel is clear.
I have been waiting for something like this. Packaging skills and related customisations as a plugin and pointing at a Git URL is a much better distribution story than copying files into one or more of the .agents, .copilot, or .claude folders.
Fork and Compact Chat Sessions
Two related quality-of-life improvements land in this release for longer chat conversations.
The /fork command creates a new session that inherits the current conversation history. This lets you explore an alternative approach without losing the original thread.
Context compaction addresses the other end of the same problem. Long sessions eat into the model’s context window. The /compact slash command summarises older messages and clears them, keeping the essentials while freeing space. There is also a context usage indicator so you can see how much of the window is currently being used before performance starts to suffer.
Taken together, these two features make longer sessions much more workable. One helps when you want to branch, the other helps when you want to stay in the same thread without carrying every previous turn in full.
Other Updates
A new Agent Debug panel provides real-time visibility into what an agent is actually doing, including tool calls, loaded customisations, and chat events. It is still in preview and limited to local chat sessions, but it should make troubleshooting much easier when a custom setup is not behaving the way you expect.
The integrated terminal also now supports the Kitty graphics protocol, allowing high-fidelity images to be rendered directly in the terminal for compatible tools and terminal setups. This one is not especially relevant to my own workflow, but it is a useful quality-of-life addition for people working with terminal-based visual output.
Key takeaways
- Session memory keeps plans consistent across conversation turns without needing to repeat context when you return to an earlier thread
- Agent Plugins move agent customisation into the Extensions model, making discovery and sharing considerably easier than copying files into folders manually
- Native browser control opens up agent-driven UI verification as a real workflow, not just a preview feature
- Fork and compact give you practical tools for managing long agent sessions without losing context or starting over
- Other additions include the Agent Debug panel for diagnosing custom agent behaviour, and Kitty graphics support in the terminal for compatible setups