browser-use

# Using the built-in browser `curl` gets you the signed-out version of everything that matters, because the session lives in a browser. agentglass has one, in a pane, already signed in to whatever the person using it is signed in to. `agentglass-browser` drives it. ```bash agentglass-browser open https://github.com/notifications agentglass-browser read # title, url, and the visible text agentglass-browser click "a.notification-list-item-link" agentglass-browser type "#search" "billing bug" --submit agentglass-browser wait ".results" # until it appears, or it fails agentglass-browser shot /tmp/page.png # png on disk; read it if you can see images agentglass-browser text ".invoice-total" # one element, instead of the whole page agentglass-browser back # and `forward` agentglass-browser scroll --to bottom # or --by -400, or --selector "#footer" agentglass-browser press Escape # Enter, Tab, Escape, arrows, PageUp/Down, Home, End ``` Every command exits non-zero and prints one line to stderr when it did not do the thing — a selector that matched nothing, a page that never loaded, a window that is not open. Branch on that rather than on the text. ## How to work with it **Read before you click.** `read` gives you the page as text. Decide from that what to click, rather than guessing a selector from the URL. **Then read narrowly.** Once you know where the answer is, `text ".selector"` cost...