Repository navigation
fix(sidebar): restore the folder geometry decided in #3905 - #18362
Conversation
5dd0f95 to
1ad854c
Compare
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: |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Before merge, check that @CrysisDeu has restated the decision on this PR. The new What this change shipsInventory (8 items) — 8 justifiedIntent: put the sidebar folder glyph back on its connector line, restoring the folder geometry the maintainer set in #3905. This is a FIX that reverts #14717 (see #18359).
[FIRST-PRINCIPLES-REVIEWED] da83400 Description read[DESCRIPTION-READ] c7dcecf2b9ef90292abcf6b5f158006de4b846ffdc7ebb722ea4c14eb9786b8d 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 that @CrysisDeu restated the decision in a comment on this PR before merge. Rule 4 does not let an agent-written [DESIGN-REVIEWED] da83400 Description read[DESCRIPTION-READ] c6af322ca2a9277dd3e2a47d6b1a2527bcd0231c0d4df84a3faed97b0d2d5b9d 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 |
UX Review (Opus 5.5) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS At the default roster width the team crewmate count now disappears without a trace; in shot-08 "Docs and support" has no count at all. Watch
Suggestions
[UX-REVIEWED] da83400 Description read[DESCRIPTION-READ] 879e33f9aaebfc8cf908c883b11712e8646429fc1ea3d660f3f3b51f50047346 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 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Review detailsNo findings that block the merge; two advisory layout regressions. FINDING — website/src/pages/ChatSidebar.tsx:783 — In board view, session rows inside a folder now start 8px to the right of the folder's name. FINDING — website/src/pages/members/TeamGroupHeader.tsx:70 — The name is now [OPUS-REVIEWED] da83400 Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
1ad854c to
1eff447
Compare
1eff447 to
0eb4974
Compare
0eb4974 to
975f7d6
Compare
975f7d6 to
5c36791
Compare
buluoray
left a comment
There was a problem hiding this comment.
Request changes on 5c36791bfc. The list-view restore itself is right: the header pad, glyph, body inset and ROW_BOX_CLS match #3905, the restored tests pin the three guides, and the list body's 4px pad still clears FolderRail's -5..3 hit strip. Two behaviour regressions sit outside the list body, plus one merge precondition on the decision entry.
1. Board view: the folder rail now covers the first 3px of every board session row. BOARD_FOLDER_BODY_CLS = 'relative border-l border-border ml-4' (ChatSidebar.tsx:779) has no left pad, while .folder-rail is position:absolute; left:-5px; width:8px; z-index:1 (index.css:3001). Rows start at x=0 of the body's padding box, so the rail spans 0..3 of each row and sits above it. A click there runs toggleColumnCollapse (ChatSidebar.tsx:3298) and collapses the folder instead of opening the session. On main the body's pl-[5px] kept the rail clear. #3905's ml-4 predates FolderRail (#15442), so a byte-for-byte restore of that value collides with it. The description's "still never overlaps a row" holds for FOLDER_BODY_CLS only. Suggested fix: ml-3 pl-1 keeps board rows at the same x as ml-4 (12 + 1 + 4 = 16 + 1) and gives the rail a 4px pad, the same clearance the list body has. A test asserting the board body carries a pad of at least 4px would pin it. (I derived this from the CSS and DOM order; I did not click it in a browser.)
2. Crewmates: long team names now truncate at every roster width, not only under pressure. shrink-0 max-w-[40%] on the name (TeamGroupHeader.tsx) caps it at 40% of the header even when the flex-1 rule beside it is filling empty space. At the default 264px roster that is roughly 87px, so a name of about 15+ characters gets an ellipsis next to a long empty line. That reverses the component's own contract ("the rule gives way first ... and only then the name") and affects every user with a longer team name, every time the roster renders. On main the name could take the full width. The squeeze being fixed is real (it comes from this PR's 4px wider ROW_BOX_CLS), but the cap should only bite under pressure: for example keep shrink-0 and replace the percentage cap with max-w-[calc(100%-<icon + gaps + count + Open team>px)], or keep the name flexible and let the rule (basis 0) and the count absorb the squeeze. A capture at the default and a wider roster width with a long team name would show it; the shot in the description only uses short names.
3. Before merge: the maintainer restatement the decision entry cites. docs/decisions/2026-10-09-folder-glyph-sits-on-its-connector-line.md says "the maintainer restates the decision in a comment there" (this PR), and README rule 4 requires that comment. It is not on the PR yet. Rule 2 makes the entry immutable once merged, so if it lands first the ledger carries a permanent entry whose evidence is missing, and fixing it would take a superseding entry plus an override. @CrysisDeu, a one-line comment here clears it.
Verified on the head: CI 89 passed, 9 skipped, 0 failed. Auto-merge is off. I read the source and CSS for both findings; I did not run the tests locally.
folder header pad to 3px to tighten the per-level indent. The maintainer saw the result on main and asked for the #3905 geometry back: the glyph on the x of the border-l line that runs down under it, the folder header on the same 14px pad as a session row, and a folder name on one edge with the agent label, title and subtitle of the sessions inside it. A plain revert conflicts in five places, so this restores the values on top of main: header px-3.5, 14px glyph, gap-[5px], body ml-3 pl-1, ROW_BOX_CLS pl-3.5, divider ml-[14px], board body ml-4, and drops the per-body row-pad override (FOLDER_ROW_PAD_CLS) that only existed to re-pad rows inside a 3px header. Measured on the built SPA: glyph box x == connector x at depth 1, 2 and 3 (259/278/297), 0px apart; root lane content 259 == root glyph, depth 1 content 278 == depth 1 glyph, the same numbers #3905 recorded. The geometry is recorded under docs/decisions/ so the recorded-decisions-are-not-regressions rule blocks the next change. Refs #18359 Reverts #14717
5c36791 to
da83400
Compare
|
Addressed on
|
|
Decided: the folder glyph sits on its connector line and folder rows share the session rows' 14px pad; #14717's indent is reverted. |
buluoray
left a comment
There was a problem hiding this comment.
Approving at da83400. This supersedes my CHANGES_REQUESTED on 5c36791: all three items are resolved.
- Board view rail no longer covers the rows.
BOARD_FOLDER_BODY_CLSis nowrelative border-l border-border ml-3 pl-1(ChatSidebar.tsx:783): 12 + 1px border + 4px pad lands the rows at the same x as #3905, and the 4px pad keeps the.folder-railhit zone off the row edge. - Crewmates team names read whole.
TeamGroupHeadercaps the name atmax-w-[calc(100%-100px)](the icon, the gaps and "Open team") instead of a flat 40%, and the capture script now fails if a team name truncates without width pressure at either roster width. - The decision entry has its source. CrysisDeu restated the decision on this PR (#18362 (comment)), which the new
docs/decisions/2026-10-09-folder-glyph-sits-on-its-connector-line.mdentry references.
CI on this head: 89 success, 9 skipped, 0 failures. I read the diff and the capture script; I did not run the tests locally or click through the board view in a browser, so item 1 rests on the class geometry and the CSS order.


Problem / Motivation
Goal: The sidebar folder glyph sits on its connector line again, with the folder geometry the maintainer set and measured in #3905.
#14717 outdented the sidebar folder glyph off its connector line (3px left of it) and cut the folder header pad from 14px to 3px to tighten the per-level indent. The maintainer saw this on
mainand asked for the #3905 geometry back (#18359).Why it matters
The glyph on the line is the picture the tree view draws: a glyph hanging in the gutter next to a line that does not reach it reads as two unrelated marks, at every depth. The maintainer decided this geometry once; without a recorded decision the next indent change can move it again.
Not a goal
paddingLeft6, 11px glyph) or the Crewmates team header.8 + depth*12pad on the hidden-folders reveal row.What changed (motivation → approach → change)
ROW_BOX_CLS, the folder header pad, the glyph size and gap, and the list and board folder body insets, and addedFOLDER_ROW_PAD_CLSso rows inside a 3px header could be re-padded per body. The glyph box then sat 3px left of the connector at every depth.git revertof fix(sidebar): tighten the folder indent so nested titles get more room #14717 conflicts in five hunks ofChatSidebar.tsx, so the values are restored on top of currentmain, and the geometry is recorded underdocs/decisions/so therecorded-decisions-are-not-regressionsrule blocks the next pull request that moves it.pl-[3px] pr-2.5px-3.5gap-[4px]gap-[5px]ml-1 pl-[3px]+FOLDER_ROW_PAD_CLSml-3 pl-1ml-2 pl-[5px]+FOLDER_ROW_PAD_CLSml-3 pl-1(rows at the x of #3905'sml-4; the 4px pad clearsFolderRail)ROW_BOX_CLSpl-2.5pl-3.5MembersPagegrouped-row indentpl-6pl-7(tracks the row pad)FOLDER_ROW_PAD_CLSis removed: it only existed to re-pad rows inside a 3px header. Thedata-*hooks it targeted stay (used by tests and by the capture harness below).pl-3.5(the content column) instead of fix(sidebar): tighten the folder indent so nested titles get more room #14717'spl-2.5.FOLDER_BODY_CLSkeepsrelativeforFolderRail(feat(sidebar): collapse a folder by clicking its connector line #15442): its hit strip spans -5..3 of the body's padding box and the body pad is now 4px, so it still never overlaps a row.BOARD_FOLDER_BODY_CLSisml-3 pl-1, not fix(sidebar): put folder rows and session rows on one left edge #3905's byte-for-byteml-4: fix(sidebar): put folder rows and session rows on one left edge #3905 predatesFolderRail, and a body with no left pad puts the rail's strip over the first 3px of every board row, where a click folds the folder instead of opening the session.ml-3+ 1px border +pl-1lands the rows at the same x (17) and gives the rail the list body's 4px clearance. Pinned by a new test and by the harness'selementFromPointprobe (below).docs/decisions/2026-10-09-folder-glyph-sits-on-its-connector-line.md+ README row. @CrysisDeu: perdocs/decisions/README.mdrule 4, please restate the decision in a comment on this PR (one line is enough, e.g. "Decided: the folder glyph sits on its connector line and folder rows share the session rows' 14px pad; fix(sidebar): tighten the folder indent so nested titles get more room #14717's indent is reverted"). The entry cites this PR as the on-record artifact.TeamGroupHeader.tsx: the team header sharesROW_BOX_CLS, so its 4px wider pad squeezed a header that had zero slack at the default roster width, and flex shrank the name and the count together ("Inta…" / "1 crewma…"): flex shrinks every shrinkable item in proportion to its basis, so even a sub-pixel squeeze of the name becomes an ellipsis. The name is nowshrink-0 max-w-[calc(100%-100px)]and the rulemin-w-0, which is the order the component's own comment promised (rule, then count, then name). The cap is the header's width less what must stay beside the name (icon 13 + four gaps 32 + "Open team" ~55), so it bites only under real width pressure: at any roster width a name may take every pixel the count and the rule can give up, and a long name truncates only once the header is narrower than it. (An earlier revision usedmax-w-[40%], which clipped names of ~15+ characters at every width beside an empty rule.) Defect in code this PR changes, so in scope; the header's geometry (pad, icon, gap) is untouched.website/scripts/capture-folder-indent-surfaces.mjs: a capture harness for the surfaces beyond the folder header (list rows, flat view, board, Crewmates), with aMEASUREprobe. It also hit-tests the first three pixels of a board folder row (elementFromPointmust land in the row, never on.folder-rail) and checks every team name at the 264 and 400 roster widths (an ellipsis is allowed only when the name is wider than the header's budget). It produced the evidence below; run against head5c36791bfcboth probes fail, against this head both pass.Backwards compatibility
Compatible: Tailwind class values and one exported constant (
FOLDER_ROW_PAD_CLS, no reader outsideChatSidebar.tsxand its tests) change; no API, schema or stored data changes.Removes nothing:
FOLDER_ROW_PAD_CLShad no reader outside the file that defined it; thedata-*hooks it targeted stay in the DOM.Tests
ChatSidebar.folderAlignment.test.tsxandplaywright/sidebar-folder-alignment.spec.tsback to their fix(sidebar): put folder rows and session rows on one left edge #3905 versions byte-for-byte: they pinpx-3.5on the header,ml-3 pl-1on the body,pl-3.5on the row and the three alignment guides at every depth.ChatSidebar.boardFolderCreateInTab,ChatSidebar.flatFolderFilter,ChatSidebar.flatView,ChatSidebar.statusMarkerandlistShellParityreversed, so each asserts the restored class.ChatSidebar.boardFolderCollapse.test.tsx: the board folder body carriesrelative border-l pl-1 ml-3and notml-4, so the rail's hit strip never covers a row.DECISIONS_BASE_REF=origin/main scripts/check_decisions_history.py: 6 entries intact.scripts/docs-lint.sh: passed.Manual verification
Built SPA,
capture-folder-glyph.mjs MEASURE=1andcapture-folder-indent-surfaces.mjs(itsMEASURElines) againstorigin/mainand this branch; x in CSS px.Folder header (as recorded in #3905):
The other surfaces, each on the same content column as the rows beside it:
Glyph box x equals connector x at every depth (0px apart); every header number matches what #3905 recorded.
Screenshots / video
Folder header, list view (before left, after right):
List view rows: pinned divider, row divider, dormant toggle, empty-folder "New chat in Archive", hidden-folders reveal at the root and inside "Kiro", nested folder body:
Flat view: date headers and the pinned divider:
The same flat view at 4x (first date header, pinned row, pinned divider, first automatic row): the 4px shift of header text, row text and divider.
Board view: folder bodies nested inside a column:
Crewmates page grouped by team (two teams and a "No team" group), the Meet CrewMates first-run flow dismissed: team header rows and their indented crewmate rows:
Team header name cap, two long team names ("Docs and support" 99px, "Platform reliability engineering" 175px), head
5c36791bfc(left,max-w-[40%]) vs this head (right). Default roster 264: the 40% cap clipped "Docs and support" to 80px beside an empty rule; now it reads whole, and only the 175px name truncates, because the header has 100px for a name there:Wide roster 400: the 40% cap still clipped the 175px name to 134px; now every name reads whole:
Board rail clearance has no visible frame (the rail is transparent at rest); the harness probe is the evidence: on
5c36791bfcelementFromPointat x+0.5/1.5/2.5 of a folder row returns.folder-railthree times, on this head it returns the row three times.Related Issues
Closes #18359. Reverts the geometry of #14717.
Pattern harvest
Rule candidate: review-prompt
Pattern: a layout value the maintainer measured and recorded in a PR is moved by a later PR without that record being read; the decision file now exists so
recorded-decisions-are-not-regressionscan block it.Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.