Snapshot
How your agent reads a page and finds the right controls.
A Snapshot turns a web page into a compact view your agent can read. It includes page text and gives links, buttons, and fields short references such as @1 and @2.
For example, a login form might appear like this:
@1 [heading] "Log in"
@2 [input type="email"] placeholder="Email"
@3 [input type="password"] placeholder="Password"
@4 [button] "Continue"The agent can use those references to choose a field or button without guessing where it is on screen.
References change with the page
A reference belongs to the current Snapshot. After navigation, a form submission, or another page update, the agent takes a new Snapshot before using a reference again. @4 on the next page could be a different button.
You usually do not need to read Snapshots yourself. If the agent seems to be working from an old view of a page, ask it to take a fresh Snapshot and check again.
Snapshots are part of ego-browser. The agent reads them inside its Space, where you can review the pages it visited.