Repository navigation
Conversation
UX Review (Opus 5.5) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS The blind reader didn't know that clicking the one-line banner jumps back to the prompt; they expected a click to expand it. Watch
[UX-REVIEWED] 37ca306 Description read[DESCRIPTION-READ] 29af9d72030b8114d49c28c45b1b2f92ac63c367102a08f7f9fe06ba1e0b0995 That sha256 names everything this verdict read: the PR description, plus the 8 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 |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Confirm #16340 carries the maintainer's direction. This PR is the only place the ledger entry's "Decided by" line is recorded, and no base RFC covers the banner. What this change shipsInventory (9 items) — 9 justifiedIntent: let a user read a prompt taller than the viewport again by putting the pinned banner back to one clickable line. This is a FIX (a rollback).
[FIRST-PRINCIPLES-REVIEWED] 37ca306 Description read[DESCRIPTION-READ] c21f4cd91830b987017498a8253cdda0598476c7cc1a703e0750084b2f394fb3 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check the tall-prompt case in the attached recording: the full row stays reachable by scrolling until the one-line card takes over, with no clipped or hidden lines. [DESIGN-REVIEWED] 37ca306 Description read[DESCRIPTION-READ] df704221fede26893a164b76dc1efcbc2f73e072a0c1089e1cd200f0926aa914 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 6.1 Review — ✅ no blocking findingsGPT 6.1 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: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
9566112 to
37faed1
Compare
Dispositions for R1 (
|
| Lane | Finding | Disposition |
|---|---|---|
| UX Review | restored card never shown; no hand-off recording | Closed. PR body now carries the collapsed card (48px one-line banner), the expanded card (368px) and a 14-frame hand-off recording (handoff.gif, 40px scroll steps through the row→card swap). |
| First Principles item 1 / Design | decision record not on head | Closed. docs/decisions/2026-10-02-pinned-prompt-is-one-line-to-jump-back.md + README row are the second commit (37faed1dc8). |
| First Principles item 4 / Design / Opus / Frontend Tests 1+2 | two later tests still pin the data-message-* reservation |
Fixed in the revert commit: the reserved inputs are removed from MarkdownRenderer.surface.test.tsx (ATTRS row) and MarkdownRenderer.domGolden.test.tsx (raw-div input); snapshots untouched. |
| GPT F1 | expanded card no longer capped at the transcript floor (#15828) | Rebutted — declared rollback. #15828 exists only to cap the fold's card; the maintainer's decision (#16340, the decisions entry on head) is to roll back to the pre-#11462 banner. The expanded card on a short pane is the pre-#11462 behaviour this PR restores on purpose (Not a goal, bullet 1; FP inventory item 6 "declared rollback of #15828"). |
| GPT F2 / Opus | an editing row can be hidden by the pin (#13630's data-message-editing guard) |
Rebutted — declared rollback. Pre-#11462 behaviour (the row hide itself dates to #950/#1131); #13630 added the guard as part of the fold chain. Restoring it alone would be new scope on a revert PR (FP inventory item 7, declared ride-along). If a guard is wanted it is a follow-up issue, not this PR. |
| Opus | wheel over the pinned card scrolls nothing (forwarder removed) | Rebutted — declared rollback. The forwarder was #11462's, needed because the fold held the card over unread lines; with the one-line banner there is nothing under it to read (FP inventory item 5, declared). |
No new scope; the diff is the three reverts plus the two sibling-test subtractions and the decision record.
|
Re-triggering CI: the 01:51Z force-push to 37faed1 registered no pull_request workflow runs (only pull_request_target lanes). Closing and reopening to fire the lanes on the current head; no code change. |
37faed1 to
3e19670
Compare
3e19670 to
746fbba
Compare
…chain) Reverts, newest first: - 5534a7a fix(chat): keep the pinned prompt card above the composer dock (#15828) - 1a3732a fix(chat): keep message actions visible under a pinned long prompt (#13630) - c250d3e fix(chat): hand the pinned prompt over at the pin line, not past it (#11462) - 65e23bb fix(chat): keep a long prompt readable instead of pinning it (#16321), merged while this revert was in review: its guard is written against the fold geometry (`bubbleBottom`, `foldY`, `restingH`) and is subsumed by the bottom-edge hand-off this revert restores. the moment its top crosses the fold line, the card stands in for it, and the card shrinks line by line as the reader scrolls. #13630 re-based the fold on the bubble and re-showed the action strip beneath the card; #15828 capped the card at the transcript floor so it stops at the composer dock. For a prompt taller than the viewport the three together make the prompt unreadable: the row is hidden as soon as its top passes the fold, the card shows only the first `maxH` pixels, and scrolling down only makes the card shorter -- the lines below the cap can never be reached (before #15828 they sat under the composer instead). The maintainer's call is that the whole fold / stand-in / hand-off geometry is over-engineering: the banner exists to keep one line of the latest prompt at the top as a click target to jump back, not to be a seamless continuation of the bubble. Roll back to the pre-#11462 behaviour -- the row stays visible and scrolls normally, the card appears once the row's bottom edge has passed the band. Conflict resolution, no behaviour beyond the revert: - `components/markdown/sanitize.ts`: the `data-message-*` attribute reservation #13630 added to MarkdownRenderer had moved here in #14580; removed here instead. - `pages/chat/PinnedPrompt.tsx`: #13318's `max-w-full` (Content Width setting) is kept; #12891's line-height fix lived inside the wheel forwarder #11462 added and goes with it. - `.github/coverage-baselines/frontend.txt`: current baseline kept. - `test/MarkdownRenderer.surface.test.tsx`, `test/MarkdownRenderer.domGolden.test.tsx`: two later tests encoded the `data-message-*` reservation as a dropped attribute; the reserved inputs are removed so the snapshots stay as is. - `test/PinnedPrompt.images.test.tsx` (new): the fold's tests carried this file's coverage; without them it reads 78.4%. The thumbnail / failed-image fallback and the jump click were untested pre-#11462 and are covered here. Refs #16340
746fbba to
37ca306
Compare
Dispositions for R3 (
|
| Lane | Finding | Disposition |
|---|---|---|
| Coverage Gate | src/pages/chat/PinnedPrompt.tsx 78.4% (76/97), below the 80% floor |
Fixed. The fold's tests had been carrying this file. New website/src/test/PinnedPrompt.images.test.tsx covers pre-#11462 behaviour that was never tested: one thumbnail per image, dropping only the thumbnail whose load failed, the ImageOff glyph once every image has failed (collapsed and expanded card), and the jump-back click. Folded into the revert commit; no baseline row added. |
| UX Review (CONCERNS) | wheel/touch over the restored card no longer scrolls the chat | Declared rollback. The forwarder was #11462's, needed because the fold held the card over unread lines; the one-line banner covers nothing the reader has not reached. Advisory, not re-added. |
| UX Review (CONCERNS) | expanded card lost the "more below" fade | Declared rollback. pinned-scroll-more arrived with #11462; the pre-#11462 expanded card never had it. Advisory, not re-added. |
| UX Review (CONCERNS) | the banner's click-to-jump is not discoverable | Noted as follow-up, out of scope. Pre-existing to the whole chain; the reviewer's own note says no fix is needed in this rollback. |
R4 head also carries the rebase onto main past #16321 (listed in the commit and the description).
|
Read-only review: the revert of the pinned-prompt fold chain (#11462, #13630, #15828, #16321) is clean; head files carry no leftover data-pinned-standin, stripUncovered, card-height or scrollTranscriptBy code, and CSS, the data-message-* sanitizer reservation and matching tests were removed together. Not approving from my side because open #16372 (buluoray) repairs the #16321 gate this removes and #16021 builds on the fold code; both collide in usePinnedPrompt.ts and PinnedPrompt.tsx. Maintainers should pick one direction first, then close or rebase the others. |
bolichen97
left a comment
There was a problem hiding this comment.
[operator: bolichen97]
Review of #16342 at head 37ca306 (main checked: b00e744, 2026-10-16 02:07 PDT).
Recommendation: close as superseded (or hold until a maintainer posts a design decision on #16340). The revert itself was clean on its own base, but its premise has been overtaken on main and it no longer merges.
Counts: 2 blocking, 1 should-fix, 2 nit.
Blocking
-
The bug it fixes is already guarded on main, and the direction is parked.
website/src/pages/chat/usePinnedPrompt.ts:81- Why: main carries #16321 (on main:
usePinnedPrompt.ts:296-305, "A prompt taller than the resting card is NOT pinned while any of it is still below the card's resting bottom") and its repair #16372 (merged 2026-10-03 10:02 PDT). The tall-prompt-unreadable symptom in #16340 is the case that guard covers. What is left is a product call (fold vs. one-line jump target). On #16340 the Captain asked for maintainer design sign-off (2026-10-08) and cruisercohen parked it on 2026-10-13 ("not on the current plan"). Merging would roll back shipped behaviour with no sign-off. Not verified: I did not re-run a browser repro of the tall prompt on main; this is from reading main's code and tests. - Fix: close this PR, pointing at #16321/#16372. If a maintainer later picks the revert direction on #16340, re-cut it from current main.
- Why: main carries #16321 (on main:
-
11 conflicts; main has built on the fold since this was cut.
website/src/pages/chat/PinnedPrompt.tsx:390- Why: merge-base is 5c7a6fd (2026-10-02 19:36 PDT). A trial merge against main conflicts in
PinnedPrompt.tsx,usePinnedPrompt.ts,UserMessage.tsx,ChatPage.tsx,ChatPane.tsx,index.css,sanitize.ts,page-layout.md,docs/decisions/README.md,usePinnedPrompt.test.tsx, plus modify/delete oncapture/pinned-prompt-handoff.tsxandUserMessage.pinnedStandinActions.test.tsx. Since then main landed fold-dependent work on these files: #16372, #16970 (thumbnail expands the card), #16842 (mask under the card), #15919 (keep the pin while reading the last reply), #19139 (Alt+Up/Down), and #21054 (moved the row markup intopages/chat/page/ChatTranscriptView.tsx, which now also carriesdata-pinned-standin, andPinnedPrompt.peek.test.tsxpins it). A rebase is no longer a mechanical revert: each of those needs its own keep/drop call. - Fix: do not rebase this branch. If the revert is approved, open a fresh one-commit revert from current main that lists those PRs too and states, per PR, what is kept.
- Why: merge-base is 5c7a6fd (2026-10-02 19:36 PDT). A trial merge against main conflicts in
Should-fix
- The decision record says a decision was made that was not.
docs/decisions/2026-10-02-pinned-prompt-is-one-line-to-jump-back.md:3- Why: "Decided by: ... (maintainer)" and the "Decision" section record the fold as rejected, but #16340 never got the sign-off the Captain asked for and is parked. The ledger exists so later PRs trust it; an unratified entry would block future fold work on a decision nobody took.
- Fix: drop this commit, or reword it as "Proposed" until a maintainer confirms on #16340.
Nit
-
Evidence list omits #16321.
docs/decisions/2026-10-02-pinned-prompt-is-one-line-to-jump-back.md:18 -
Jump button's purpose is only in
title.website/src/pages/chat/PinnedPrompt.tsx:390- Why: the accessible name is the prompt text;
titleis not reliably announced, so a screen-reader user does not hear that activating it jumps back (the UX lane hit the same confusion). Main has the same pattern (PinnedPrompt.tsx:754), so this is not introduced here. - Fix: add an sr-only "Jump to this turn" span referenced by
aria-describedby(or prefix the label), in whichever version ships.
- Why: the accessible name is the prompt text;
What checks out
- On its own base the revert is complete: no leftover
data-pinned-standin,computeLiveCardH,computePinnedCardMaxH,scrollTranscriptByorstripUncoveredin the head tree; CSS, sanitizer reservation, capture scripts and fold tests go together; pre-#11462 tests restored; newPinnedPrompt.images.test.tsxcovers the restored card. - Keyboard on the restored card: both controls are native
<button type="button">; the chevron hasaria-expanded+aria-label; keyboard focus opens the peek, pointer focus does not. - AI lanes on this head: Opus 5.5 no findings, GPT 6.1 no findings, First Principles PASS, Design PASS, UX CONCERNS (advisory). No human reviews in the REST list.
| // A prompt hands over to the banner only once it is entirely behind the band | ||
| // (bottom edge at or above the band's bottom), so a prompt taller than the | ||
| // band scrolls away line by line instead of collapsing the moment it is sent. | ||
| const handoffY = pinHandoffY(foldY, pinCollapsedHRef.current) |
There was a problem hiding this comment.
Blocking. Main already carries #16321's tall-prompt guard (main usePinnedPrompt.ts:296-305) and its repair #16372, so the #16340 symptom is covered there. #16340 is parked (2026-10-13) with no maintainer design sign-off. Fix: close this PR; re-cut from main only if a maintainer picks the revert direction on #16340.
| // which would let the column refuse to shrink below the full prompt. | ||
| className={`min-w-0 min-h-0 flex-1 flex flex-col bg-transparent border-none p-0 m-0 text-left ${folding ? '' : 'cursor-pointer'}`} | ||
| onClick={onJump} | ||
| title={i18nT('pages.chat.pinnedPrompt.jump_to_this_turn')} |
There was a problem hiding this comment.
Blocking. 11 conflicts against main. Main has built on the fold since: #16372, #16970, #16842, #15919, #19139, #21054 (ChatTranscriptView.tsx now also carries data-pinned-standin). A rebase is no longer a mechanical revert. Fix: if approved, a fresh revert from main listing those PRs and what is kept from each.
Nit (a11y, pre-existing). The jump action lives only in title, which screen readers do not reliably announce. Fix: sr-only "Jump to this turn" via aria-describedby.
| @@ -0,0 +1,19 @@ | |||
| # The pinned prompt is one line to click and jump back, never a stand-in for the row | |||
|
|
|||
| Decided by: Zezhen Xu (maintainer, @CrysisDeu) | |||
There was a problem hiding this comment.
Should-fix. Records a decision as made, but #16340 never got the requested maintainer sign-off and is parked. Fix: drop this commit or mark it "Proposed" until a maintainer confirms.
| ## Evidence | ||
|
|
||
| - https://github.com/kirodotdev/KiroCrew/issues/16340 -- the issue recording the decision and the maintainer's direction. | ||
| - https://github.com/kirodotdev/KiroCrew/pull/16342 -- the revert of #15828, #13630 and #11462 and the pull request that adds this entry. |
There was a problem hiding this comment.
Nit. The PR also reverts #16321; list it with the chain here and in "Why".
|
Closing this as superseded. The parent issue #16340 was parked by the maintainer on 10-13, and main has kept building on the pinned-prompt fold since then, so reverting the fold no longer fits main. @bolichen97 also suggested closing it. The branch is kept, so this can be reopened if the direction changes. |
Problem / Motivation
Goal: A prompt taller than the viewport can be read in full again: the pinned-prompt banner goes back to pinning one line once the prompt has scrolled past, and never hides or shrinks the row it points at.
Since #11462 (2026-09-22) the pinned-prompt card is a progressive "fold": the moment a user prompt's row top crosses the fold line the whole row goes
visibility: hidden, the card stands in for it, and the card shrinks line by line as the reader scrolls. #13630 re-based the fold on the bubble's height and re-showed the action strip beneath the card; #15828 capped the card atmaxHso it stops at the composer dock.For a prompt taller than the viewport the three together make the prompt unreadable: the row is hidden as soon as its top passes the fold, the card shows only the first
maxHpixels, and scrolling down only makes the card shorter. The lines below the cap can never be reached (before #15828 they sat under the composer instead). The maintainer's screenshot: the card's bottom edge cuts through the middle of a line while the rest of the prompt is unreachable.Why it matters
The maintainer writes long prompts routinely; every one of them is unreadable after it scrolls past the fold. The banner exists for one thing -- keep one line of the latest prompt in view as a click target to jump back -- and the fold chain broke that one thing while adding ~23 files / ~6,900 lines of geometry, tests and capture scripts.
Not a goal
max-w-full, the Content Width setting on the card) or fix(chat): insert file/folder mentions at caret and unstage edited-out chips #6511 / refactor: split MarkdownRenderer into owners under components/markdown #14580 / docs-only commits that landed on the same files.sanitize.tsbeyond removing thedata-message-*reservation that fix(chat): keep message actions visible under a pinned long prompt #13630 added (it had moved there in refactor: split MarkdownRenderer into owners under components/markdown #14580).What changed (motivation → approach → change)
Observed: a tall prompt is hidden the moment its top passes the fold and the card never shows more than its first
maxHpixels → root cause: the fold design itself (row hidden + card as stand-in + height tracks the row's remaining height) → change: revert the chain, newest first.Reverted, as one squashed commit:
fix(chat): keep the pinned prompt card above the composer dock (#15828)fix(chat): keep message actions visible under a pinned long prompt (#13630)fix(chat): hand the pinned prompt over at the pin line, not past it (#11462)fix(chat): keep a long prompt readable instead of pinning it (#16321)-- merged while this revert was in review. It patches the same symptom from inside the fold geometry (bubbleBottom,foldY,restingH); with the fold gone the bottom-edge hand-off already does what its guard did, so its hunks inusePinnedPrompt.ts, its test rewrite and itspage-layout.mdparagraph go with the chain.Conflict resolution, no behaviour beyond the revert:
website/src/components/markdown/sanitize.ts: thedata-message-*attribute reservation fix(chat): keep message actions visible under a pinned long prompt #13630 added toMarkdownRenderer.tsxhad moved here in refactor: split MarkdownRenderer into owners under components/markdown #14580; removed here.website/src/pages/chat/PinnedPrompt.tsx: fix(chat): apply the Content Width setting to the user's message bubble #13318'smax-w-fullis kept; fix(chat): convert a line-mode wheel with the paragraph line height #12891's line-height fix lived inside the wheel forwarder fix(chat): hand the pinned prompt over at the pin line, not past it #11462 added and goes with it..github/coverage-baselines/frontend.txt: the current baseline is kept (the pre-fix(chat): hand the pinned prompt over at the pin line, not past it #11462PinnedPrompt.tsx 3.5row is not re-added).The decision is recorded in
docs/decisions/2026-10-02-pinned-prompt-is-one-line-to-jump-back.md(second commit).Backwards compatibility
Compatible: UI-only; no API, config, or storage shape changes.
Reader:
website/src/index.css:[data-pinned-standin]-- chat -- the rule set is removed with the attribute it targeted; no other reader (grep on origin/main:data-pinned-standin,computeLiveCardH,computePinnedCardMaxH,scrollTranscriptBy,stripUncoveredall return nothing outside the reverted files).Tests
Removed with the reverted code:
PinnedPrompt.foldScroll,PinnedPrompt.foldContent,PinnedPrompt.ceiling,UserMessage.pinnedStandinActions,MarkdownRenderer.reservedDataHooks, plus the fold/ceiling cases inusePinnedPrompt.test.tsxandpinnedPrompt.test.ts. The pre-#11462 cases in those two files are restored verbatim. Two later tests outside the chain had pinned thedata-message-*reservation as a dropped attribute (MarkdownRenderer.surface.test.tsxATTRS row,MarkdownRenderer.domGolden.test.tsxraw-div input); the reserved inputs are removed so their snapshots stay as they are. Local gates:tsc -b,eslinton every changed file,scripts/docs-lint.sh-- all clean. Test suites run in CI.Manual verification
Captured with a Playwright harness against the built SPA from both trees (same 60-line prompt, 1280×860, same scroll offsets):
Screenshots / video
Left: main (fold chain). Right: this PR. Same prompt, same scroll position.
The restored card once the tall prompt has scrolled past the band (hand-off by the row's bottom edge; card is the one-line 48px banner, 368px expanded):
Recording of the hand-off (14 frames, 40px scroll steps; the row stays visible until its bottom reaches the band, then the one-line card takes over):
Related Issues
Closes #16340
Pattern harvest
Not generalizable: a product-shape decision (the banner is a jump target, not a continuation of the bubble), recorded in
docs/decisions/so the fold is not re-introduced.Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)