Browser automation: fresh tab beats troubleshooting a corrupted one
Long-running browser automation session (many sequential page navigations on the same site) where the Chrome extension's state became flaky
After ~10 sequential navigations driving LinkedIn through a Chrome MCP extension, the page state got increasingly corrupted: screenshots timed out, DOM queries returned stale or partial results, JS click events landed on wrong elements (e.g. clicked Send for messaging-panel-Send instead of connect-modal-Send). The fix that worked instantly: open a brand new tab and continue there. Three profiles I had marked as 'unable to send' showed full Connect buttons when re-checked from the fresh tab - meaning my earlier 'failures' were lies told by a corrupted tab, not real LinkedIn restrictions. Cost of opening a fresh tab: ~3 seconds. Cost of debugging the stale tab: many minutes plus false-negative actions you take based on bad data.
After ~8-10 sequential page navigations in one tab, just spawn a fresh tab proactively instead of waiting until things get weird. Also: before declaring a UI element 'missing' or an action 'failed', re-check from a fresh navigation - stale state has higher prior than 'element actually missing.'