GitHub Copilot app walkthrough shows diff, terminal and browser panels for reviewing agent code

GitHub has published a beginner-focused walkthrough for its Copilot app, showing how developers can inspect, execute and verify agent-generated code inside a single window instead of juggling separate tools. According to the GitHub Blog, the app pairs a diff view, a terminal and a browser preview as built-in panels that sit alongside the coding agent.

Reviewing changes in the diff panel

The first panel covers inspection. The diff view presents a before-and-after comparison of the agent's edits, marking added lines in green and removed lines in red. From there, a reviewer can accept the changes, leave comments, or ask Copilot to revise specific parts. The post stresses that the human stays in charge: the agent proposes edits, but the reviewer decides what actually lands.

Running commands from the terminal panel

Reading the code only goes so far, the GitHub Blog argues, so the app also embeds a terminal where project commands can run directly inside the session. Commands can be typed by hand or configured as scripts that sit behind a Run button. The walkthrough uses a website as its example: you add a dev server script that opens the client folder and executes npm run dev, then press Run to start the server. Several terminal windows can stay open at once, so a dev server and other commands can run in parallel while you keep working.

Previewing results in the browser panel

For projects with a user interface, the browser panel renders the running application so a new feature can be tested the way an end user would experience it. GitHub also highlights its Pick & Polish tool, which lets you select an element on the page and have the agent adjust it. After each iteration, the dev server script can be re-run to confirm the fix actually took effect before you move on.

Finishing the review cycle

Once the diff has been checked, the project runs cleanly, and the preview behaves as expected, the change can be accepted and a pull request created without leaving the app. The GitHub Blog boils the workflow down to three questions a reviewer should answer before accepting any agent's work: what changed, does it run, and does it work in practice.

Why it matters

Agentic coding tools are changing what code review means: the reviewer is increasingly evaluating a batch of machine-written changes rather than a colleague's incremental patch. That shift makes verification, not just reading, the real bottleneck. Until now, gathering that evidence meant assembling an editor, a terminal and a browser yourself, which is precisely the friction GitHub is targeting by folding all three into one session. For teams adopting agent-driven development, the post doubles as a usable checklist — inspect the diff, run the code, exercise the feature — before anything merges. It also signals where the Copilot app itself is heading: less a chat sidebar, more a self-contained workspace where supervised agents do the heavy lifting. As with any vendor walkthrough the framing is promotional, but the underlying discipline — prove the change works before you accept it — is sound advice no matter whose agent wrote the code.