Search a webpage and extract results

Mandatory prerequisite: First fetch and fully execute the official Lexmount Setup Prompt returned by https://browser.lexmount.com/api/agent-setup/browser-cli, completing installation, authentication, and environment validation. Continue only after every check passes. If the URL is unreachable or validation fails, stop immediately and report it; never skip the setup or pretend it succeeded.
Task: Use browser-cli only to open https://en.wikipedia.org/, search for "browser testing frameworks," extract the first three content-page results as structured JSON, and make the result directly openable from the final Codex response.
Flow: Create a temporary Session, locate the Wikipedia search box, submit the exact query, wait for the Search results page, and extract the first three results from the main content list. For each result, record its rank, title, absolute URL, and optional excerpt. Exclude navigation links, search suggestions, account actions, and links outside the main result list. Include the query, requested URL, final URL, and results in valid JSON, verify the result, and close the Session.
Acceptance: Success requires a non-empty, parseable JSON result containing exactly three results. Every result must have a unique rank, a non-empty title, and an absolute en.wikipedia.org URL. The generated JSON must be available as a clickable file in the final Codex response. If login or CAPTCHA is unexpectedly required, hand control to the user only through Remote View without reading or recording anything the user enters.
Cleanup: On success, failure, timeout, or cancellation, close only the Session created for this task. If closing fails, report "partial success." Never close a pre-existing Session.
Return only the status, final URL, result count, cleanup result, and error summary, then provide the generated JSON file as the final item so it can be opened directly in Codex. Do not expose API keys, Project IDs, Session IDs, connection URLs, or login credentials.