Repository navigation
feat(dashboard): add a setting that puts unfiled sessions above folders - #22596
Conversation
bolichen97
left a comment
There was a problem hiding this comment.
Reviewed head 5745ce3. No blocking or should-fix findings from my read.
What I checked:
- Backend: the new
DashboardConfig.unfiled_sessions_above_foldersfield defaults to false, the loader reads it through_safe_bool, the PATCH allowlist row is a strict bool, and the change-card catalog lists it as a path that cannot widen agent permissions. The new test module round-trips both values, refuses non-booleans with the file bytes unchanged, and loads an absent or hand-edited value as off. - Tree lane: with the setting off, the
root-groupdroppable renders as before through the extractedrenderRootBucketRows. With it on, the bucket renders first at natural height and a newroot-lane-taildroppable fills the rest. Both registerfolderId: null, and no code outsidelistLanes.tsxkeys on theroot-groupid, so the drop handler and collision logic need no change. - Board lane: the unfiled cards only move position; the folder DndContext, the column drop and the unfile strip are untouched, and flat view stays inert.
- Menu row: gated by
folderRowsDrawn, usesmenuitemcheckboxwitharia-checked, and toggles from the shown value so two clicks during one save send opposite values. - Specs and the feature map are updated in the same commit. No
.github/file is touched. The linked issue is T2 and the change is additive and off by default.
Remaining before approval: CI and the review lanes are still queued on this head, so PR Readiness has not passed yet. I will approve once readiness passes on this same head.
Merge note: another open PR changes the same sidebar files and the same config section, so whichever merges second needs a rebase.
Design Review (Opus 5.5, fork) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: the setting is off by default and only adds things, the screenshots and recording show this revision, and both null-folder drop targets carry the same Notes
[DESIGN-REVIEWED] 88aa4d6 Description read[DESCRIPTION-READ] bfc85c6ad51a82fa9bd8a0a49c17fd233692c30fcb0876ac376bd65bd0d84f5c 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, fork) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The screenshots show every control and state the diff adds: the menu row checked and unchecked, the tree and board with the setting on, the refused-save notice and both drop targets. A recording covers the toggle movement. No first-time reader has seen them, because this fork lane has no blind read; pushing the branch to this repository runs one. Notes
[UX-REVIEWED] 88aa4d6 Description read[DESCRIPTION-READ] 9edc25c180f83ab82674f6694b6b6b3f70bd18412faf0d109ded3b8226fd5d71 That sha256 names everything this verdict read: the PR description, plus the 11 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, fork) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks: the setting is off by default, comes from the linked #17073, and no existing mechanism puts unfiled sessions above the folders. What this change shipsInventory (7 items) — 7 justifiedIntent: let a person keep the sessions they still have to sort above their folders in the chat sidebar. This is an ADDITION.
[FIRST-PRINCIPLES-REVIEWED] 88aa4d6 Description read[DESCRIPTION-READ] b6f44f0f599e058c5ea7bea18c1f433f32006f996698d00c0b5f42dff44f3a7b That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
Opus 5.5 Review (fork) — ✅ no blocking findingsReviewed |
dodgeblaster
left a comment
There was a problem hiding this comment.
Stewarding review for #17073. I read the diff at head 5745ce31, the issue, the AI-review verdicts, and docs/decisions/. The implementation is well-built and complete; the one thing keeping it from readiness is author-supplied visual evidence that a fork reviewer cannot add, so I'm requesting changes to flag exactly that.
What I verified (all good)
- Both views covered, as #17073 asks ("the sidebar list view and the board view columns").
listLanes.tsxdraws the unfiled bucket above the folders and a newroot-lane-taildroppable below them;boardLane.tsxgatesungroupedRowsabove the folder blocks withunfiledFirst = unfiledAboveFolders && !flatView. Flat/conductor lanes are correctly inert. - Off by default, off-path unchanged.
unfiled_sessions_above_folders: bool = False; the loader reads it through_safe_booland the sidebar reads=== true, so an absent or hand-edited value draws today's layout. TherenderRootBucketRowsextraction preserves the old render. - Both drop targets register
folderId: null, so ungrouping still works whether the bucket is above or the tail is below — and nothing outsidelistLanes.tsxkeys on the reusedroot-groupid. - Backend wiring correct: PATCH allowlist row (
bool) incore_config_schema.py, and the key is inchange_card_catalog._CANNOT_WIDEN_SETTINGS(a view preference must not widen agent permissions). - i18n complete: a genuine translation in every shipped locale (bn/de/es/fr/hi/it/ja/ko/pt/ru/zh-CN), the source
en.manual.json, and the regenerateden-XA.json. - Docs + feature map updated (
config.md,history.md,feature-map/README.md,config-baseline.json). - Tests are thorough and ran green in CI's backend suite: the new
test_config_patch_unfiled_above_folders.pyround-trips both values, refuses non-booleans with the file bytes unchanged, and loads absent/non-boolean as off; the three frontend files cover the tree, the board, off-is-identical, collision resolution, the double-click-during-save, refused-save recovery, and both drop targets. - No recorded decision in
docs/decisions/pins the unfiled-below-folders placement, so this opt-in setting conflicts with none. (The issue remains a maintainer product call on the setting's name and placement, per the T2/needs-human triage — this review does not pre-empt that.)
Why I'm requesting changes (not a code defect)
The only blocking readiness item is the UX Review BLOCK, and it is purely an evidence gap — the reviewer can't evaluate the feature because the required captures are missing. These have to be added to the PR body by you (the author); a fork reviewer can't push them. Please attach:
- A short recording of toggling "Unfiled sessions above folders" on, then off, in the folder tree — this is the claim ("rows use their existing movement animations") that currently has no evidence.
- A board-view still with the setting on — all three current images show only the tree, so the board half (which the issue specifically asks for) is unverified visually.
- The open menu with the new row checked (its
border-accent bg-accent+ Check-icon state); attachment-02 shows the unchecked state only. - A still of the error notice on the shared folder notice after a refused save.
Tip from the prepare-pr skill: gh pr edit <n> --body-file <body> --attach <png> mints permanent user-attachments URLs; after attaching, rewrite the Screenshots section to reference the fresh URLs.
One low-severity thing to consider (not blocking)
The UX reviewer's "watch": the checkbox sits directly under the Custom / By name / By date Folder order radios with no separator, so it can read like a fourth sort choice rather than a layout toggle. A thin separator or a small indent would set it apart. Your call — it's low severity and fades after first use.
Once the captures are attached and the UX lane clears on the same head, PR Readiness should pass and this is in good shape for a maintainer to take. I'm leaving this on Gary's list either way; as a fork PR it's review-only from me (I won't merge).
|
Updated the top level writeup with the requested screenshots and recording |
5745ce3 to
7af351a
Compare
bolichen97
left a comment
There was a problem hiding this comment.
Reviewed the new head 7af351a against the previous one (range diff over the rebase onto current main). Still no blocking or should fix findings from my read.
What changed since my last review and what I checked:
- The rebase follows main moving the cannot widen settings list into
change_card_catalog_kinds.py. The new key sits in that frozenset in sorted order, and nothing else in the backend diff changed. - The menu row now renders its own separator above it, inside the same
folderRowsDrawnguard, so it reads as a layout toggle rather than a fourth Folder order choice, and a lane that hides the row shows no doubled separator. This addresses the low severity watch item from the UX review. - The description now carries the recording of the toggle in the folder tree, a board view still with the setting on, the open menu with the row checked, and the error notice after a refused save. These are the four evidence gaps the UX review named on the previous head.
- The rest of the range diff is the feature map row picking up a neighbouring row that landed on main.
Remaining before this can merge: CI and the AI review lanes have not run on this head yet, so PR Readiness is still red and the UX review has to clear on this head. A maintainer also needs to take it, since there is an open change request from the previous head and the change adds a new user facing setting.
Merge note: other open PRs change the same sidebar files and the same dashboard config section, so whichever merges second will need a rebase.
|
HOLD at 7af351a. The code looks sound: the setting is off by default, the PATCH row accepts only a boolean, the key is in What holds it:
— crew-pr-reviewer (r121) |
|
@bolichen97 the failed CIs all pass locally and seem to have timed out, they're not touched by this PR at all. I don't have permission to re-run them without uploading another rebase. Would appreciate if you could help get this in before any other conflicts as it's been bouncing on rebases for a long time at this point |
dd52c24 to
31a798f
Compare
bolichen97
left a comment
There was a problem hiding this comment.
Reviewed 31a798f against my last review at 7af351a, with a range diff over the rebase. Still no blocking findings from my read.
What changed:
- A refused save of the new setting now gets a notice of its own ("Couldn't save this setting. Nothing changed. Try again." with the gateway's text under it), instead of sharing the folder notice. The next toggle or a dismiss clears it.
useUnfiledAboveFoldersreturns the error and a dismiss, and the config spec says the same. - The rest of the range diff is the rebase: the feature map and history spec rows follow neighbouring rows that landed on main.
Two things still stand between this head and a merge:
- @dodgeblaster's change request from 5745ce3 is still open. It asked for author-supplied visual evidence, and the description has carried the recordings since 7af351a. @dodgeblaster, could you take another look when you have a moment?
- CI and the AI review lanes have not run on this head. Only the fork pre-checks have reported, so PR Readiness is red until a maintainer releases the fork runs.
31a798f to
ba723f5
Compare
|
Thanks for this PR. I read the full diff at ba723f5. The change reads well:
No regex is added. Two things before this can move:
|
|
Correction to my comment above: please ignore point 2. The description already has the board view with the setting on, the menu with the setting checked, the error notice, and the toggle recording, which covers the evidence the open change request asked for. Only the rebase ask stands. |
ba723f5 to
646caec
Compare
|
@dodgeblaster could you re-review? Your change request at 5745ce3 asked only for visual evidence, and the description now has all four items: the toggle recording in the folder tree, a board view with the setting on, the menu with the setting checked, and the error notice after a refused save. The branch is rebased onto main at 646caec. Every non-shard check on this head has passed, apart from the readiness signal, which only reports the open change request. I read the full diff earlier, and it is clean: no regex, off by default, and both unfiled drop targets keep |
|
Re-checked at 570ab45 against my review at 31a798f. Since then the branch was rebased, and the unfiled root bucket now folds Slack code-channel crews under their conductor, as rows inside a folder already do. There's no regex and no new refusal, and the setting stays off by default. @dodgeblaster, your change request from 5745ce3 asked for author-supplied visual evidence. The description has carried the recordings since 7af351a. Could you take another look when you have a moment? Fork CI hasn't run on this head yet. I'll approve once it has and it's green. |
Code change approver recommendation: A person should decideWhy: the decider estimates a 12% chance that this change introduces a defect, within the 12% limit for approving on its own, but it changes 1406 lines, more than the 400 it may approve on its own. This PR adds a gateway setting, off by default, An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit |
bolichen97
left a comment
There was a problem hiding this comment.
Re-approving 88aa4d6280, @gsboylan. This is a rebase of 69ba740d75, which I approved earlier, with conflict fixes only. git range-diff of the two commits shows no change to your own lines. The only differences are main's lines next to them: command_bar_enabled now sits beside your new key in config-baseline.json, the config dataclass and change_card_catalog_kinds.py, where show_unread_dot used to be, and main reworded the folders row of the feature map and a history.md row. The setting, its placement under Folder order, and its tests are unchanged.
I approved the 17 held fork runs on this head.
AI-assisted, sent by CommunityCrew
A new session starts in no folder, and the sidebar draws the unfiled sessions after every folder in the tree lane and in each board column. With many folders a session still to be sorted sits off screen and is easy to lose, and neither flat view, which hides the folders, nor pinning, which orders only within the unfiled bucket, reaches it. This commit adds dashboard.unfiled_sessions_above_folders, a gateway config boolean that is off by default, written through the config PATCH allowlist like folder_sort and session_sort, and classified in the change-card catalog as unable to widen what an agent may do. A checkbox row under Folder order in the sidebar's sort-and-filter menu toggles it. The row is drawn only in the tree lane and the non-flat board, where a folder can be drawn, and it keeps the menu open like the other checkbox rows there. With the setting on, the tree lane draws the unfiled bucket first, at natural height and only when it holds a row, then the folders, then a lane tail that fills the remaining height. The bucket and the tail are both null-folder drop targets, so dropping a session on the unfiled rows or on the empty space below the folders removes it from its folder, and the "Drop here to remove from folder" hints stay below the folders so nothing above the pointer moves when a drag starts. Each board column draws its unfiled cards before its folder blocks. With the setting off, the tree lane and the board columns render as before. The unfiled rows stay inside the bucket's render prop in both layouts, so their paint-order stamps follow the drawn order and are unchanged with the setting off. The config spec documents the key, both layouts, and the drop semantics, the history spec records the new owners, and the feature map lists the setting under Session folders. New tests pin both layouts, the drop targets, the menu row, and the config round trip. The MCP chat_folder_tree listing is unchanged. Refs kirodotdev#17073
bolichen97
left a comment
There was a problem hiding this comment.
Re-approving 5b9a9c191b, @gsboylan. This is a rebase of 88aa4d6280, which I approved earlier, with conflict fixes only. git range-diff shows your own lines unchanged, including your edit to the feature map's folders row. The differences are main's lines around them: the unread-folders-first option from #25781 (its locale keys, useFolderSort's renamed fields and the sort menu's props), the per-filter pause from #35906 (pausedFilters), and useNativeSessionDrag in the facade composition test. The setting, its placement under Folder order, and its tests are unchanged.
I approved the 17 held fork runs on this head.
AI-assisted, sent by CommunityCrew
|
HOLD; the "unread first" row has changed styling in a rebase and now it doesn't match. Will update to match that row and adjust the screenshots |
Problem / Motivation
Goal: Let a person keep the sessions filed in no folder above their folders in the chat sidebar, in the folder tree and in each board column.
A new session starts in no folder. Agents spin sessions off, and the composer opens them, unfiled. The sidebar draws the unfiled sessions after every folder, in the folder tree and in each board column. With many folders, a session still to be sorted is below the last folder and off screen. Flat view does not help, because it hides the folders. Pinning does not help either, because it orders only inside the unfiled group, which still sits below the folders.
Why it matters
The unfiled sessions are the ones a person still has to sort, so they are the ones they look for first. With a long folder list they scroll past every folder to find a session they just opened, or lose track of it. The setting is off by default, so anyone who relies on the unfiled group below the folders, and on it as the target that removes a session from its folder, keeps that layout.
Not a goal
layoutId. The folder blocks move in one frame, as they already do when the Folder order radios beside the new row reorder them, because folder blocks carry no layout animation today. Nothing disappears or is replaced, so this is not the hard swapwebsite/AGENTS.mdnames.chat_folder_treelisting. Its unfiled group stays last.dashboard.folder_sortnordashboard.session_sorthas one.What changed (motivation → approach → change)
New sessions start unfiled, so with many folders they land below the last folder, off screen. A new sidebar setting puts unfiled sessions above the folders instead. It is one gateway config boolean,
dashboard.unfiled_sessions_above_folders, off by default. It is stored on the gateway likedashboard.session_sort, so the choice follows the person to every browser and the desktop app. A checkbox row, Unfiled sessions above folders, under Folder order in the sidebar's sort-and-filter menu turns it on and off. The row shows only in the folder tree and the non-flat board.With the setting on, a drop on the unfiled rows or on the empty space below the folders still removes a session from its folder. The "Drop here to remove from folder" hint stays below the folders, so nothing above the pointer moves when a drag starts. One known trade: the first folder now sits right below the last unfiled row, so a manual reorder that overshoots the last row files the session into that folder. The move undo bar reverses it. With the setting off, the folder tree and the board columns render exactly as before.
The gateway side is the
DashboardConfigfield inconfig/sections_dashboard.py, its read inconfig/loader_builders.py(any stored value other thantrueis off), the PATCH allowlist row indashboard/handlers/core_config_schema.py, thechange_card_catalog_kinds.pyclass that says the setting cannot widen what an agent may do, and the regeneratedconfig-baseline.json. The sidebar side ispages/chat-sidebar/folders.ts(the read and the optimistic save),listLanes.tsx,boardLane.tsx,SearchDock.tsx,scope.tsandpages/ChatSidebar.tsx. The label is ini18n/locales/en.manual.json, with a translation in each shipped locale and the regenerateden-XA.json.docs/system-specs/modules/config.md,docs/system-specs/modules/history.mdanddocs/feature-map/README.mddescribe it.Backwards compatibility
Compatible: the new key is optional and off by default, the loader and the sidebar read anything but
trueas off, and an older gateway without the field draws today's layout.Removes nothing: the change adds one config key, one allowlist row and one menu row, and every existing layout, route and key keeps working as before.
Tests
test/test_config_patch_unfiled_above_folders.py:trueandfalseround-trip through the config file andKiroCrewConfig.load(). A non-boolean value is refused with 400 and the file bytes stay the same. An absent or non-boolean stored key loads as off.website/src/test/ChatSidebar.unfiledAboveFolders.test.tsx: with the setting off, the folder tree's element signature is the same for an absent key and for a storedfalse,'true',1ornull. With it on: the order of unfiled rows, folders and lane end; the unfiled group is absent when nothing is unfiled and when a search matches no unfiled session; the menu row's checked state, its flip before the save lands, two clicks during one save, and a refused save put back with a notice; the row's presence per lane; where the drop hints appear during a session drag and a folder drag; both drop targets register no folder and unfile a dropped session; and which targetsidebarCollisionpicks above, on and below the folders.website/src/test/ChatSidebar.boardUnfiledAboveFolders.test.tsx: a board column with the setting off renders the same for an absent key and for every other stored value; with it on, the unfiled card is above the folder block; the column's unfile strip still unfiles; the menu row shows on the board and not on the flat board.website/src/test/ChatSidebar.ownerComposition.test.ts: the new hook's place in the sidebar's pinned hook order.Determinism: the backend module passed 20 of 20 runs at
-n0, 10 of 10 at-n 4 --dist loadgroupbesidetest/test_config_patch.pyandtest/test_config_baseline.py, and a shuffled order (seed 29953). The three frontend files passed 20 of 20 runs underKIROCREW_ACT_STRICT=1and a shuffled order (seed 19524).Manual verification
A gateway built from this branch, run on a scratch copy of a real data home (33 open sessions, 11 folders), was checked by hand: the setting moved the unfiled sessions in the folder tree and the board, and turning it off brought the old order back.
Screenshots / video
Existing UI unchanged without this feature enabled:

This PR adds a toggleable setting in the existing "Folder Order" section of the session filter:

When enabled, sessions are filed above folders in the list. Sessions moving up in the list use their existing movement animations when the toggle state changes.

Error notice when the write operation is denied:

Board view, setting enabled:

Dropdown menu with the setting checked:

(minor note - you can see that there's a doubled up separator between the 'Folder Order' and 'Tags' sections, I confirmed that's present in mainline already and not introduced by this change)
Drag onto an unfiled row:

Drag a filed session below the folders to unfile:

Checking/unchecking, movement:
Screen.Recording.2026-10-21.at.10.45.56.AM.mov
Related Issues
Closes #17073
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributed under the project's Apache-2.0 license, as the Licensing section of CONTRIBUTING.md describes.