Screenshots

The browser subagent can take a screenshot of the visible page or of a single element on it, for example to check how a page rendered. Ask for a screenshot in your prompt, or the subagent may take one on its own when the task calls for it.

To start the browser subagent, type /browser in the message box, select it from the menu, and describe the task. The subagent drives your own Google Chrome, which must already be running with remote debugging turned on. Chrome doesn’t need to be signed in. For setup steps, refer to the Browser command.

In your conversation, the subagent appears in a row that reads Starting or Started, followed by a name the agent gives it, such as “Web Browser Automation Agent”. Click that name, or the subagent in the Subagents section of the conversation overview, to open its conversation in the side panel. Each screenshot is a step labeled with a short summary of what the subagent captured. Expand the step to see the tool’s arguments, its result, and a thumbnail of the screenshot.

A screenshot isn’t an artifact on its own. To comment on one, ask the agent to save it to the conversation’s artifacts, then open the image from the agent’s reply. Drag to select a region of the image, type your comment, and click Add Comment. Comments aren’t sent right away: they’re held as pending comments and go to the agent with your next message, or when you submit them from the Review button in the artifact’s header.