feat(sidebar): mark the row a create just made

[mindaugaskasp]

A new snippet or saved diff landed below every starred row, in a panel that is usually scrolled past its insertion point, and nothing marked it.

Spec: specs/2026-08-07-new-row-highlight/plan.md — from interaction proposal artifact 89f60c6d.

What it does

Four channels on one create event, because the sweep across all 14 themes shows no single one survives every palette:

channel	what it contributes
insertion	the row animates in — layout motion cannot fail a contrast check
wash	transient accent tint, peaks at 26% (the sepia minimum), decays to nothing
rail	3px --accent down the row's RIGHT edge, wipes in and stays
badge	a NEW keyline chip — says it in words, so hue cannot fail it

Plus the behavioural half: expand a collapsed section, scrollIntoView({ block: 'nearest' }), say so when a filter hides the row, and retire the mark on the row's primary action, the next create, or 60s.

The rail goes right because the left carries the pin bar — and starring a row you just made is the ordinary response to finding it, so the two states genuinely co-occur and must stack.

Why not the wash already in the repo

.key.added (TrustedKeyRow.css) works in a dialog whose rows have no hover background. A sidebar row has --bg-hover, so a resting tint competes with hover, not the panel. The tint needed to clear a theme's own hover step ranges 12% (dim) to 26% (sepia), and on light no tint ≤60% clears it at all.

Evidence — all 14 themes

Every number was re-derived from themes.css before the code was written, then confirmed against the live DOM by make theme-sweep. The two agreed exactly.

• accent edge on panel: min solar 3.21, max matrix 13.92 — clears the 3.0 non-text floor everywhere
• accent as a label: fails 4.5 on solar, meridian, sepia, bloom, light — so the label is --text (min sepia 8.55)
• --accent vs --favorite: 0.044 ΔE on dim and matrix — which is why words, not hue, carry the identification

Notable

• One key on uiStore, not one per store. "The row you just made" is a single row, so saving a diff then adding a snippet must leave one mark, not two in different sections.
• marks opt-out. add() is also the seam for seeding, bulk import and backup restore. Caught by reading the sweep's own PNGs, where the first-run seeded example wore a NEW badge.
• Two utils/ extractions. The size ratchets refused the growth at the right seam, so the pure logic came out instead of a number going up: utils/snippetState.js and utils/vaultEntries.js. snippetStore 575 → 496, vaultStore 408 → 366, both baselines retightened. Stores re-export their public names, so no caller changed.
• theme-sweep probes can now name the channel they gate. The accent keyline was collected and never enforced, which would have left the design's central claim unguarded.

Verification

• npm run check — green. 2650 passed | 2 skipped; coverage 95.1 st / 88.03 br / 95.7 fn / 96.1 li
• make e2e (Docker, full suite) — 395 passed, 2 skipped (the macOS-gated specs)
• make theme-sweep (Docker, all 14) — 504 measurements, none under floor
• e2e/new-row-marker.spec.mjs watched red → green; it asserts the rail's box and computed colour, the badge's keyline vs label, and that a starred new row keeps both marks

🤖 Generated with Claude Code