VS Code 1.126: Control Whether Your Agent Skips Approval Prompts, Plus Session-Level Token Usage

TL;DR: VS Code 1.126 adds per-session token usage with cache hit rate, a bypass-approval toggle for agents, and deprecates the built-in chat Edit mode.

VS Code 1.126 shipped June 22 with several quality-of-life improvements for AI-assisted development.

Agent bypass-approval control. A new setting lets you decide whether an agent can skip tool approval prompts entirely. If you've been finding the per-action confirmation loop too disruptive, you can now unlock bypass — or lock it down explicitly if you want the prompts always.

Session-level token usage. Chat sessions now surface total token consumption for the session, including cache hit rate. Useful for understanding how much prompt caching is actually saving you, and for keeping AI-assisted work within cost targets when using metered models.

Unified model customization picker. Context window and reasoning effort controls are merged into a single model picker — less clutter when swapping configurations mid-session.

Edit mode deprecated. The built-in chat Edit mode is deprecated in favor of the Agents window. If you were still using Edit mode for targeted file rewrites, now is the time to migrate your workflow to Agents.

PR activity in chat. CI failure and comment indicators (via codicons) now appear directly in the chat session list so you can see PR health without leaving the editor.

Why it matters: The bypass-approval toggle and session token reporting are the two changes most worth configuring now — together they give you better control over both agentic autonomy and cost visibility in a single release.

Sources: VS Code 1.126 release notes, VS Code on X