Repository navigation
feat(ui): top bar search, readout and feedback pills wear the glass - #15523
Conversation
88a1f70 to
014d44d
Compare
014d44d to
5f61bc7
Compare
5f61bc7 to
069c7c2
Compare
First Principles Review (Fable 5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS The option-A material pick lives only in an unlinked crewmate DM thread — a maintainer should confirm that choice is theirs before merge. What this change shipsInventory (8 items) — 8 justifiedIntent: make the top bar's three pills wear the same Liquid Glass material as the rest of the shell — an ADDITION (restyle), maintainer-picked from three mocks.
[FIRST-PRINCIPLES-REVIEWED] ba9167d Description read[DESCRIPTION-READ] 70b9ed92d05c7191220927339b6c0ede05c641c9a24527656bec06eb167b4602 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Fable 5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing to check. [DESIGN-REVIEWED] ba9167d Description read[DESCRIPTION-READ] 0de3bbea4751522f2705bd6eeebaa392fc457cd4591cd403a86461158f8ce247 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
UX Review (Fable 5) — ✅ PASSUX-level review of UX-Verdict: PASS Verify live that the search trigger's hover is perceptible: the blind reader saw no difference between the rest and hovered full-bar frames (shot-01 vs shot-02). [UX-REVIEWED] ba9167d Description read[DESCRIPTION-READ] 3bb75d3a5fa757645b263c1d037c434e7d9345897ea5bf8114802cf06796ef5e That sha256 names everything this verdict read: the PR description, plus the 12 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
The three pills in the top bar -- the search / command-palette trigger in the centre track, the readout capsule (connection dot, metrics, credits) and the Request a Feature pill -- were bordered or bg-card boxes beside a composer dock, a sidebar search field and a crewmate header that are all Liquid Glass. They are now `components/Glass.tsx` panes (chip recipe, 12px radius, the `rounded-xl` the update pill already wears), rendered as the control itself: the trigger is `<Glass as="button">`, the capsule hosts its segments directly so the collapse rungs still see them, and the pill's two buttons sit transparent on the pane. The topbar's own blurred strip is unchanged (option A of the three mocks the maintainer chose from). Consequences the change has to carry: - The capsule's terminal rungs in index.css were `> *:not(:first-child)`; the Glass host's effect layers are its first children, so that would have kept a layer and hidden the connection dot. Both rungs now hide every non-layer child that has a non-layer sibling before it, still structural (registered extension segments cannot be enumerated), and the test pins the selector. - Offline is a tint step, `glass-danger`, added beside `glass-accent` / `glass-warn` in all three solidifying blocks (the mirror test keeps them in step), never a flat `bg-danger-subtle` fill. - The kiro-light inset ring on `.tb-capsule` / `.feedback-pill` is gone: the glass draws its own lit edge on the white chrome, and the hook test and the theming contract shrink to the user bubble. `.feedback-pill` no longer has a rule, so the class is dropped; `.tb-capsule` stays as the rung hook. - The motion layout wrapper around the capsule is now a plain flex div outside the pane, so the width tween is unchanged. Evidence harness: scripts/capture-topbar-glass.mjs (dark / light / kiro-light, hover, the solid fallback with the Translucent panels switch off, the offline tint, and a 180px-container probe of the terminal rung).
069c7c2 to
ba9167d
Compare
Opus 5 Review — ✅ no blocking findingsReviewed Review detailsNothing blocks — two advisory findings, both in the new capture harness. FINDING — website/scripts/capture-topbar-glass.mjs:170 — FINDING — website/scripts/capture-topbar-glass.mjs:115 — with [OPUS-REVIEWED] ba9167d Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
buluoray
left a comment
There was a problem hiding this comment.
LGTM. I checked this at head ba9167d3cf.
The search trigger, readout capsule and feedback pill now wear the shared liquid-glass material. Because liquid-glass.tsx inserts its effect layers before {children}, the old :first-child collapse rungs would have matched a layer instead of the first segment; both terminal rungs now use .tb-capsule > :not([data-liquid-glass-layer]) ~ :not([data-liquid-glass-layer]) (index.css:2161, :2174), which keeps the first real segment and hides the rest. The new glass-danger tint is built only from theme variables (--danger into --glass-tint) and is mirrored in all three solid-fallback blocks plus a prefers-contrast outline. Focus rings come from the existing .liquid-glass:focus-visible rules; dropping overflow-hidden from FeedbackPill is safe because InstantTip portals out. The removed .feedback-pill class has no remaining users.
Mutation: putting the 200px rung back to .tb-capsule > *:not(:first-child) reddens App.test.tsx "shifts the collapse rungs…". The three changed test files pass 139/139. No assertion was weakened; the kiro-light ring assertion left together with the rule it pinned. All five review bots pass at this SHA with no CONCERNS or BLOCK.
Non-blocking follow-ups:
- The PR body calls
ChatInput.liquidGlass.test.tsx"(unchanged)", but the diff edits it (adds--glass-tint-dangerto the:rootregex and a.glass-dangerassertion). - The per-block loop in that file (~line 276) pins
glass-warnin each fallback block but notglass-danger, so the@supports not (backdrop-filter)block'sglass-dangerline is untested. capture-topbar-glass.mjs: withGLASSunset, the solid fallback hides the glass layers, solayersShown === layers(line ~170) fails and the run throws before writingreport.json; withOFFLINE=1and noSTOP_FILE,readFileSync(undefined)is swallowed and the run times out after 120s (line ~115). Dev-script only.
Problem / Motivation
Goal: the three pills in the top bar -- the search / command-palette trigger, the readout capsule and the Request a Feature pill -- wear the same Liquid Glass as the rest of the shell.
They were bordered or
bg-cardboxes next to a composer dock, a sidebar search field and a crewmate DM header that are all Liquid Glass panes, so the top row read as a different, older material from everything under it. The maintainer picked this shape (option A, "swap the material of the controls only") from three mocks.Why it matters
The top bar is on every page; while its controls stay flat boxes the glass elsewhere reads as unfinished rather than as one system. It is also the one place where the kiro-light theme needed a special inset ring to keep a
bg-cardpill visible on the white chrome -- a glass pane carries its own edge and retires that exception.Not a goal
.topbar-glass): option B, a transparent bar with free-floating capsules, was not chosen.components/Glass.tsx,ui/liquid-glass.tsx) beyond one new tint step.What changed (motivation → approach → change)
Goal → approach: reuse the existing primitive as the control itself, the way the follow-up chips and the jump-to-bottom button already do, rather than wrapping each control in a glass box (a wrapper would break the capsule's structural collapse rungs and the trigger's
flex-1sizing).Change:
App.tsx: the search trigger is<Glass as="button" variant="chip" radius={12}>withglass-hover; the readout capsule is aGlasspane hosting its segments directly (the motionlayoutwrapper moves outside it as a plain flex div, so the width tween is unchanged); offline isglass-dangerinstead of abg-danger-subtlefill. 12px is therounded-xlthe update pill already wears.components/FeedbackPill.tsx: the pill is aGlasspane; its two buttons sit transparent on it. Thefeedback-pillclass had one consumer (the kiro-light ring below) and is dropped.index.css: the two terminal rungs were.tb-capsule > *:not(:first-child); a Glass host's first children are its effect layers, so that would have kept a layer and hidden the connection dot. Both rungs now hide every non-layer child that has a non-layer sibling before it -- still structural, since registered extension segments cannot be enumerated. Newglass-dangertint step besideglass-accent/glass-warn, mirrored in all three solidifying blocks. The kiro-light inset ring on.tb-capsule/.feedback-pillis removed.components/Glass.tsxheader andwebsite/docs/theming-contract.md: the surface list and the "card must carry its own edge" paragraph now name the three pills and shrink the kiro-light hook list to the user bubble.scripts/capture-topbar-glass.mjs: the evidence harness (below).Backwards compatibility
Compatible: no API, config or storage shape changes.
.tb-capsulekeeps its class and the rung container queries fire at the same widths; downstream top-bar extension segments (getCapsuleSegments) render exactly as before, as children of the capsule.Tests
src/test/App.test.tsx: the terminal-rung assertions now pin the layer-skipping selector, so a future:first-childregression that would hide the dot behind a glass layer fails the test.src/test/kiroLightShellHooks.test.ts: the hook table and the ring assertion shrink to the user bubble, so the removed rule cannot come back unnoticed and the dropped class does not read as load-bearing.src/test/ChatInput.liquidGlass.test.tsx(unchanged) pins that theglass-dangerline was added to the media block and its attribute mirror identically.Manual verification
Real gateway serving this branch's
website/dist, Chromium 1400x900, Translucent panels on and off, plus the harness's rung frame pinning.tb-rightto 180px (under the 200px terminal rung): all 5 glass layers still render and exactly the first segment (the dot) stays visible. The gateway was then stopped with the page open to photograph theglass-dangercapsule. The harness asserts each pill is a.liquid-glasshost, the trigger has no border, the capsule's layers precede its segments, and the transparency switch state.Screenshots / video
Each frame is the centre track (top) and the right group (bottom), 2x, Translucent panels on.
glass-hover)glass-dangercapsuleglass-hoverlifts the pane ~12/255 on dark (measured mean 33,31,38 -> 45,43,50) plustext-muted -> text-textWhole page, dark
Related Issues
N/A — maintainer request in the crewmate DM thread (follow-up to #14896 glass search field and #15167 glass identity pill).
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)