Summary

Backlog tracker for the v0.1.34 prod UX review (Playwright, run on 2026-05-02).

Source report: docs/reports/mm-web-prod-v0.1.34-playwright-review.md

Each child issue is independently mergeable. This issue stays open until everything below is either shipped or explicitly declined.

Status

Remaining: #697 only (waiting on external contributor response for CLA + ruff format).

Verification adjustments to the original report

The review was reverified against source on 2026-05-02. Two adjustments worth noting:

• P0 "Cross-tab stale content leaks into the accessibility tree" — reframed and downgraded to P2 (Tab hidden attribute not re-applied on deactivation in activateTab() #699). Modern browsers exclude display: none from the a11y tree, so screen readers are not actually misled. The underlying DOM-state hygiene gap (the hidden attribute not being re-applied) is real and worth fixing, but the original "screen readers receive misleading page context" framing is incorrect.
• P0 "Duplicate chunks" — confirmed at the data layer via SQL (matched (source_file, start_line, end_line, content_hash, namespace) rows ~1 second apart, concentrated in claude-plans namespace). Indexing creates duplicate chunks within a single run (no UNIQUE on content_hash) #691 already tracks this — the fix should land at the insert layer (UNIQUE constraint or INSERT OR IGNORE), not at the presentation layer the original review suggested.

Items intentionally NOT filed as issues

• Long path / namespace truncation (P2 in review) — current right-side text-overflow: ellipsis is in place across .source-group-dir, .source-name, .result-filename, etc. The review's recommendation (middle-ellipsis, basename-first) is a UI preference, not a correctness bug. Reopen if user feedback shows scanability is actually painful.
• Prod includes dev/harness scripts (P2 in review, reclassified P3) — settings-harness.js and settings-hooks-watchdog.js are loaded unconditionally from index.html in prod, but their UI targets are all data-ui-tier="dev" so they do not activate. The cost is a few KB of payload and parser time. Will revisit if/when prod payload becomes a measured concern.

Re-verification scheduled

A remote agent is scheduled for 2026-05-30 (4 weeks after the review) to re-check each finding against the then-current release and post a status comment here. (Routine: trig_018GFJqCAhYux4h1SLpjbKFQ.)