[jeroendee]

Summary

• Fill the bulk tag-removal row’s hover highlight through the right edge when no scrollbar is needed.
• Show the pointer cursor over the clickable blank space beside session tags.

Two scoped CSS corrections, responsive regressions, and matching docs. Existing click/selection handlers, independent tag buttons, bounded scrolling, batch scope and save semantics are unchanged. This is a core presentation fix, not a setting or override.

Proof

Revision: 745c4f5ea20ab51549187e7c7e03b90d7f4989e5, targeting this fork’s main. Real isolated PocketApp routes/assets in Linux/Chromium, with synthetic sessions and a scripted model; no live data or mocked HTTP behavior.

Scenario / expectation	Observed result	Evidence
Bulk removal hover reaches the right edge, which remains clickable	Before: a 10px reserved gutter. After: less than 1px gap; clicking 2px inside the right border queues removal via the native label	Native UI regression at 1280/390/320px; screenshots/demo below
Clickable blank tag-row space uses the pointer cursor and opens the intended session	Before: computed cursor auto. After: pointer, followed by real opening of the correct conversation at all 3 widths	Extended native opening regression; demo assertions
Existing interactions remain independent	Ctrl/batch selection, tag-chip filtering/selection, touch hold, suppressed hold-click and keyboard focus return pass	All 5 tests in test/session-tags-polish-ui.test.ts

Validation: npm run format, npm run lint, npm run check, git diff --check, and the complete npm test run pass: 440 tests, 440 passed, 0 failed/skipped, 651683ms. All 4 committed files match the SHA-256 manifest of that fully tested precommit candidate. The targeted 5-test UI run also passed. Both reported defects were reproduced by the same assertions before the CSS repair.

The screenshots and 18-second native demo are attached below. The demo verifies a 0 CSS px right-edge gap, native opening, 2-session selection, right-edge removal and cancellation preserving both stored tag lists, with zero browser errors. It uses 90 real CDP screenshots sampled at approximately 5fps, duplicated to encode 10fps at 1280×800. No application behavior is simulated. A browser screenshot does not include the system cursor: the pointer value is machine-asserted, not claimed to be visible in the recording. Types, lint and all 5 targeted regressions also pass again on the exact commit.

Limits: local Chromium integration and mobile emulation, not physical-device, other-engine, screen-reader or production acceptance. No new API or persistence behavior. This PR does not update upstream PR TannerMidd#118, change its restored description, merge anything, or update/restart the running installation.

native-demo.mp4