VS Code 1.135

[Agents] Agent host: The agent host lets you connect to the same agent session from multiple VS Code windows. It runs agent harnesses in a dedicated process based on the Agent Host Protocol (AHP). The agent host's Copilot agent is powered by the Copilot SDK, which aligns the agent's behavior and functionality with the Copilot CLI, the standalone GitHub Copilot app, and other Copilot products. We're actively developing the agent host. The following screenshot shows the Copilot harness selected for an agent host in an editor window: Learn more from our VS Code Agent Host documentation and the video below. If you have any feedback or requests, please let us know by filing an issue. See it in action in our Agent Host introduction YouTube video.

[Agents] Continue external agent sessions in VS Code: Setting: chat.agentSessions.showExternal The Sessions list in VS Code can now show recent Copilot or Claude agent sessions that you created in other applications. By default, VS Code shows up to two recently updated external sessions. Select a session from the Sessions list to view the conversation and continue it in VS Code with your Copilot subscription. When you open an external session, a banner at the top of the chat lets you configure how many external sessions appear in the Sessions list. You can also use the External submenu in the Sessions list filter to choose which external sessions are shown. Change this preference at any time with the chat.agentSessions.showExternal setting.

[Agents] Rubber Duck (Experimental): Rubber Duck is an experimental feature that allows you to get a second opinion from a complementary model on the agent's work. It helps surface missed details or edge cases. You can use Rubber Duck by invoking the /rubber-duck command in a Copilot agent host session. Learn more about Rubber Duck in GitHub Copilot CLI.

[Agents] Single-pane side layout is now the default: Setting: sessions.layout.singlePaneDetailPanel In the previous release, we introduced the single-pane layout. In this layout, session details and editors are located in a single side pane, with a shared tab bar next to chat. The single-pane layout is now enabled by default for the Agents window on desktop. This release also improves the layout: Diffs use a side-by-side view when space permits and switch to an inline view when the side pane becomes too narrow. Use Always Show Inline Diff from the editor title menu to keep diffs inline at every width. The action bar is less crowded. Editor-specific actions for diffs, view modes, code review, and attachments move to the editor title area, while the shared header focuses on showing or hiding Details. To use the classic layout, disable sessions.layout.singlePaneDetailPanel and reload the window.

[Agents] Simplified session controls and information: The session header is less crowded, so you can identify the active session more easily and focus on the conversation. The session title has a prominent position, while an overflow menu next to the title groups actions for creating a chat and pinning the session. Search remains available in the Sessions list. When a session contains multiple chats, chat tabs replace the single-chat header. Session information moves directly above the chat input, where it is easier to find while you work. Pills can show changes, pull requests, issues, browsers the agent is interacting with, and artifacts from the session. For example, the Changes pill shows live file and diff counts and opens the complete set of session changes. Right-click an individual pill to open a context menu and choose which pill types are visible. The Changes pill remains visible whene

[Chat] Editor-style sticky scroll for chat (Experimental): Settings: chat.stickyScroll.enabled , chat.experimental.stickyScroll.enabled Chat sticky scroll lets you keep the current prompt visible as you scroll through and review long responses. We've further refined the behavior to make it more consistent with sticky scroll in the editor. Enable both settings to try the redesigned experience.

[Chat] View detailed chat turn usage: To give you better insight into your chat usage, we've redesigned the chat response footer. When you hover over it, the footer shows a per-model breakdown of the input, cached input, and output tokens used in that chat turn.

[Chat] Sandboxing for the local agent harness: We previously rolled out sandboxing to 50% of users to validate it on a larger scale and in real-world use cases. While we didn't identify a specific blocking issue, continuing the rollout would likely require more support and follow-up work at a time when we want to preserve focus on higher-priority investments, especially in the area of the agent host and the Copilot harness. We've therefore returned the default rollout to 0% for now. Sandboxing for the local agent harness remains available as an opt-in feature through the UI, so users who want to try it can still enable it.