redirect-to-page! now mutates tab state (switches to an existing tab or updates the active tab) when navigating, which is core navigation behavior and easy to regress. There are existing route handler tests, but none cover the new tab-related branching (e.g., switching to an existing tab vs updating the current tab, and the skip-auto-tab? escape hatch). Consider adding tests in frontend.handler.route-test (or similar) that assert the expected :tabs/* state changes for these navigation paths.

The auto-hide logic treats any focused INPUT/TEXTAREA/contentEditable element as “the editor”, so focusing search fields, settings inputs, etc. will start the hide timer and may hide tabs unexpectedly. Since the spec says “when editor has focus”, consider narrowing this predicate to Logseq’s editor element(s) (e.g., by checking against the actual editor input from state, or requiring a specific container/selector).

This change removes the header breadcrumb (block-breadcrumb) entirely, so the header no longer shows the block/page hierarchy context that previously appeared there. The PR description focuses on tabs and doesn’t mention removing breadcrumbs; if this is not intentional, consider restoring the breadcrumb (or moving it elsewhere) to avoid a UX regression.

The new settings row uses hard-coded English strings for the label and tooltip text. Elsewhere in this file, user-facing settings labels/hints are consistently translated via t keys. Consider adding i18n keys for this setting label + hint and using t so non-English locales get translations (with the existing fallback to :en).

This PR is scoped to tabs, but this hunk adds a new “Remove icon” button and changes the icon action behavior on pages. If this is intentional, it should be called out in the PR description; otherwise consider reverting/splitting into a separate PR to keep the tabs change focused and reduce review/rollback risk.

open-all-journals-in-tab! only creates a new Journals tab when there are already more than 1 tab. With exactly one non-Journals tab open, Cmd/Ctrl+clicking Journals (which calls this handler) will update the current tab instead of opening a new tab, violating the spec that Cmd/Ctrl+click opens in a new tab. Consider always creating a new tab here (unless an existing Journals tab already exists), regardless of current tab count.

close-all-tabs! triggers a redirect to :all-journals, but graph-switch immediately calls route-handler/redirect-to-home! right after. This creates back-to-back redirects during graph switches (potentially affecting history, causing UI flicker, and doing extra work). Consider splitting close-all-tabs! into a “reset tab state” variant that does not navigate, and use that during graph switching.

The document-level focusin/keydown/pointerdown handlers update typing? and schedule/clear timers even when the “Auto hide tabs when typing” setting is off. This can leave typing? stuck true while disabled, so enabling the setting later may hide tabs immediately without the intended 3s delay. Consider gating these handlers on the setting (or resetting typing? when auto-hide? is false) so state/timers aren’t mutated while the feature is disabled.

(tabs/tab-bar) is rendered unconditionally in the header. On native mobile platforms tabs are meant to be disabled (tabs-enabled?), so this will still mount the component + global event listeners and may reserve header space even when there are no tabs. Consider rendering the tab bar only when tabs are enabled (and/or when there is at least one tab) to avoid layout/regression on iOS/Android native.

The auto-uninstall notification message claims the plugin “has been automatically uninstalled” even on non-Electron platforms, where the code does not uninstall it (it only calls the IPC uninstall when Electron). Consider adjusting the message based on platform (e.g., “Please uninstall manually” on web) to avoid misleading users.