agent-browser

Launch a controlled browser session, navigate to the URL, and wait for a specific selector or network-idle rather than an arbitrary sleep. Capture the screenshot immediately after the wait condition succeeds, before any interaction changes the state.

Walk through the reported flow one click at a time, waiting between actions and re-screenshotting after each state change. Record the console log filtered to errors, any non-2xx network responses, the viewport size, and exact reproduction steps.

Use a browser only when the content is JavaScript-rendered, login-gated, or requires interaction. If the page is reachable as static HTML, curl or an HTTP fetch is cheaper, faster, and more reliable.

The page may have changed or you may be querying before rendering completes. Take a screenshot to see the actual state, then adjust the selector or wait for a specific element instead of a fixed delay.

Prefer programmatic auth such as API tokens or magic links over UI login. If UI login is the only path, the user must provide credentials explicitly for that run, and credentials must never appear in logs, screenshots, or output.

Some sites detect headless Chrome and serve different content, and cross-origin iframes you do not own cannot be inspected. It is also unsuited to unattended large-scale scraping, which needs rate limits and robots.txt handling.