You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Rebuild the project view on upstream project and row primitives #419
Design: retain the hierarchy, replace duplicated sidebar foundations
Based on #416 findings, owner decisions, and read-only inspection of main 1e072987b65fc34e7099dc459c33d57c8309252f. Upstream nightly v0.1.10-nightly.20261116.4800 (7d32075e3757ded707914e2450c3e392a9665765) is an ancestor.
Done means: custom sections/subsections, mixed child order, dragging, worker families and repository-orchestrator controls remain; Projects and Activity share project identities, folder expansion semantics, common row behavior and commands. Global activity scope remains independent of expansion. Saved sections survive unchanged. No additional per-row subscriptions or unrelated-row renders. The duplicate project thread renderer is deleted after cutover.
Scope: web and desktop's shared web sidebar, local and remote environments. Native mobile and provider/delegation lifecycle redesign are excluded under the fork guide. No plugin framework, second store, new layout schema, or new styling system.
Current footprint
Counts below are physical lines at the pinned main. Shared-file +/− counts are git diff --numstat 7d32075 HEAD, not an estimate of removable code. Fork-only files have the same added-line count as their length. Paths below are relative to apps/web/src/ unless qualified.
These helpers need not all change. Layout transport, backend persistence, debugging and text-size preferences are retained boundaries, not rewrite targets.
Edits inside upstream files
File
Current lines
Fork delta +/−
components/Sidebar.tsx
3,914
322 / 76
components/Sidebar.logic.ts
1,658
241 / 46
components/sidebar/SidebarThreadRow.tsx
2,038
269 / 75
components/sidebar/SidebarThreadHeader.tsx
259
162 / 73
uiStateStore.ts
822
419 / 64
components/AppSidebarLayout.tsx
517
18 / 3
components/sidebar/SidebarChrome.tsx
312
14 / 4
components/sidebar/SidebarDraftBlock.tsx
251
10 / 1
components/sidebar/mainAppLocation.ts
78
3 / 1
components/ui/sidebar.tsx
910
10 / 2
components/settings/SettingsSidebarNav.tsx
444
2 / 0
The unchanged upstream primitives include hooks/useSidebarProjectGroups.ts (43 lines), sidebarProjectGrouping.ts (246), and SidebarProjectFolderHeader.tsx. Do not reset shared files wholesale: their deltas include unrelated retained features.
Existing fork evidence to extend, not replace: ProjectSections logic/drag/permissions tests (232/245/177 lines), ProjectThread logic (150), BackgroundWork logic (507), Columns logic (21), FamilySeparators logic (172), ThreadProcesses logic (193), WorkExpansion (71), Reveal logic (77), Sidebar.logic.fork (578), SidebarLayoutSync (160), layout client/projection/migration (214/396/80), sidebarDisplay (384), placement/appearance/section-folders (54/34/110). Shared upstream test edits also exist in Sidebar.drag.browser and SidebarDraftBlock.browser (+1 each). Counts describe this feature slice, not the whole fork.
Place those snapshots into saved sections; do not derive a second project identity.
Section/subsection hierarchy and dragging
Existing resolveProjectDrop, resolveSubsectionDrop, shared sidebarSectionChildOrder and applySidebarLayoutMutation
Upstream has no section equivalent. Keep section DndContext, keyboard sensor, permissions, drop indicator and folder-drop/new-project placement. Do not turn on upstream folder mode as a substitute: it disables drag/Working shelf.
Wrap the upstream header in the existing sortable section placement. Add only the narrow header appearance/drag/action inputs needed for retained colors, icons and controls. Avoid nested buttons or nested li markup.
Share row assembly/handlers between views; extract the current assembly only as needed rather than recreating all its callbacks in Projects. Preserve capability/permission checks, confirmation and multi-selection. Archive is not settle; use the existing command for each.
Section ancestry and collectVisibleProjectThreadOrder must describe the same rendered tree, including collapsed worker families. Keep reveal opening ancestors without changing the activity filter.
Keep orchestratorsFirst, groupRunningSubagentsByParent, withChildNeeds, SidebarBackgroundWorkRows, useSidebarWorkExpansion, worker provider and ask-mode controls, issue badges, process visibility and family separators. No upstream product replacement is claimed for these.
Model/effort columns and tooltips
Shared composer EffortGlyph and model option descriptors
Keep sidebarEffortGlyph and SidebarColumns as the single effort implementation; pass provider models by environment + instance.
Global activity scope
Existing sidebarProjectScopeKeys and scope/filter handlers
Never derive scope from expanded folders or rewrite expansion on filter changes. Projects keeps its hierarchy; Activity keeps its shelves/search.
One owned feature component: keep SidebarProjectSections as the fork composition root; it owns sections, drag state and custom worker/control composition. Its existing private helpers/pure logic are acceptable; “one component” does not mean one giant function. Remove its private SidebarProjectThreadRow after common-row cutover. Keep only narrow composition inputs in upstream Sidebar/header/row files. Generic row actions remain upstream-owned; fork-only menu items are contributed through the existing augmentation.
One layout-state owner: retain the controller created by createSidebarLayoutClient, mounted once by SidebarLayoutSync. It owns client-side durable layout updates, per-environment revision/write lanes and pending edits. SidebarLayoutService remains the server authority. useUiStateStore is a projection/read facade with forwarding actions (configureSidebarLayoutActions, applySidebarLayoutProjection), not a second writable copy. useSidebarProjectsView reads that projection and retains only transient dialog/view state. No second sync hook or state copy in rows. Local work expansion and activity filtering remain separate UI concerns.
Saved layouts: carry over without a new migration
Retain SidebarLayout, the server table/migration, revision protocol and storage keys. Preserve section IDs, names, parentId, mixed childOrder, project memberships, color/folder settings, collapse states, projectOrder and Other expansion verbatim. Do not export the visible grouped list back as a replacement layout.
Keep projectSidebarLayout as the durable-ID → grouped-row projection and sidebarProjectRefsForKeys as the inverse mutation lookup. Grouped rows use upstream sidebarFolderExpansionKeys for reading/toggling; commands still target the existing environment-scoped physical project IDs. Re-grouping or a disconnected environment must not delete unrendered memberships. Keep primary-environment precedence already encoded by the projection.
Retain existing legacy paths: migrateSidebarLayout, pending import keys, resolvePendingSidebarImport and createLegacySidebarRetirement. Do not reset/import again on component mount. Preserve empty sections and unresolved aliases until their environment arrives; late import must not overwrite subsequent owner moves. The startup sidebarDisplaySnapshot is display-only, never a sync/import source. While it is active, layout writes stay guarded.
Use synthetic persisted fixtures covering all fields and disconnected environments. No inspection or modification of the owner's live userdata is needed for this design or implementation.
Performance acceptance
This is a correctness constraint, not a hoped-for optimization:
Keep upstream useQuietThreadShells for list/group derivation and sidebarProviderEntriesAtom's stable environment maps. Eliminate the custom project row's broad useEnvironmentProviders() subscription when moving it to the shared row.
Compute grouping, parent/child lookup and layout projection once at the feature boundary; preserve references for unchanged groups/thread arrays. Pass stable action callbacks and primitive per-row state. Do not pass the entire layout/filter record into each row.
Reuse useSidebarRowSubscriptionLease for visible/active work. Reconcile the fork's useSidebarThreadProcesses visibility observer with it; do not mount both independent live-query/observer paths for the same work. No per-row layout streams, provider queries or extra global-store subscriptions.
Closed sections/projects do not mount thread rows or process detail lists. A streaming tick or provider checkedAt-only update must not redraw every project; selecting a thread should update only affected rows; expanding a family should affect that family and any actually changed project separator.
Before and after cutover, record render/subscription counts with the same synthetic hundreds-of-threads fixture, same expanded rows and events (stream update, provider status-only update, selection, one project expansion). Assert event/count invariants, not milliseconds, sleeps or a guessed speedup. Preserve existing required subscriptions; any net increase or unrelated-row renders is a failed gate. No new virtualization dependency in this task.
Separately mergeable implementation order
Unify layout access and grouped expansion. Keep the existing UI, move Projects expansion onto sidebarFolderExpansionKeys and the existing layout action/projection path, and remove duplicate layout derivation from the view hook where applicable. Preserve current data format. Extend existing projection/client/migration tests only where coverage is missing: grouped aliases, disconnected/reconnected environment, empty/nested sections, mixed order, late imports and pending edits. A real behavioral correction gets a failing case first. Capture the same nested synthetic layout before/after if visible expansion behavior changes. This is independently useful without changing row rendering.
Use upstream project headers and shared project actions inside the hierarchy. Retain section/subsection drag wrappers and new-project placement. Integrate upstream folder row IDs and tree ancestry. Extend existing ProjectSections drag/permissions and visible-order logic cases for mixed project/subsection drops, read-only targets and collapsed ancestors; use a before/after capture for presentation and a short interaction recording for dragging/focus. Keep the old thread renderer temporarily. This step depends on 1.
Use the common thread row and one subscription path. Share upstream row assembly between Activity and Projects; retain custom role/effort/family/work adornments through narrow inputs and existing owned helpers. Delete duplicate thread row/tooltip behavior only once the shared row covers it. Extend existing background-work/status/order logic tests for any changed behavior; run Sidebar effort bars are missing on some threads and drawn differently from the composer's #423 effort cases unchanged. Perform the render/subscription count comparison and a real-client pass for row actions, worker expansion, keyboard navigation, permissions, remote instance lookup and both supported sidebar appearances. Before/after captures must include default-effort Claude orchestrators and Codex workers. This step depends on 2.
Finish the shell boundary and remove obsolete composition code. Move remaining fork-only dialog/section/control wiring behind SidebarProjectSections where feasible; remove dead helpers/imports and duplicated handlers, retaining generic search/Activity/shelves and upstream row commands. Verify Settings, command/keybinding entry points and route reveal against the same action/state owners. Extend existing visible-order/reveal logic tests only if behavior changes; capture the full nested layout before/after and replay the integrated workflow. Compare shared-file numstats against this baseline and account for every removed behavior. This step depends on 3; it is not permission for an unrelated cleanup.
Each step can merge with working Projects and Activity views; no half-migrated storage or feature flag scaffolding. Run focused tests while iterating and agent-heavy vp run check:changed in a background shell before each push. No repo-wide suite. Use existing disposable synthetic fork fixtures; stop the dev server started for the integrated pass. Browser tests are justified only for actual hooks/drag/focus wiring, not visual styling.
Main contains the default-effort change (4b2460d738) and glyph-size follow-up (37c672f4f2). Keep:
Explicit selection and recognized reported effort handling, then descriptor default only when neither provides an explicit effort; unsupported/unknown effort must not invent bars.
Provider resolution by environment and instance, with descriptor models passed through to sidebarEffortGlyph.
SidebarEffortMeter rendering the shared EffortGlyph size="xs"; retain current CSS sizing and do not restore the removed scaling.
Existing SidebarBackgroundWork.logic.test.ts regression assertions plus before/after visual evidence against the composer.
Check top-level project threads, worker/subagent rows and Activity wherever effort is displayed. Do not copy a pre-fix(web): show default sidebar effort bars #424 renderer from upstream or an older fork commit.
Owner decisions (before presentation changes; recommendations preserve current behavior)
Should Projects keep its current compact columns, colors, family separators and custom text sizes while adopting upstream row mechanics? Options: keep current presentation (recommended); adopt upstream folder/card appearance and retire those preferences. The latter changes scope and requires explicit approval; hierarchy and controls remain mandatory either way.
Should pinning change where a thread/project appears inside custom sections? Options: retain saved section placement and current orchestrator-first Projects order, sharing upstream pin/archive commands and indicators (recommended); introduce a separate global Pinned shelf/reorder policy in Projects. The second is a product change, not required by reuse. Preserve archive/unarchive through existing actions without erasing section membership; do not invent a new archive layout.
Already decided: keep hierarchy, dragging and controls; use upstream primitives. Those decisions are not reopened. No implementation was performed. Tests/browser captures are implementation gates, not evidence produced by this source-only design.
This spans layout compatibility, project headers, thread-row subscriptions and final integration; use four sequential implementor assignments rather than one broad rewrite. I attempted propose_issue_change(type: "split"), but it returned: “Only a managed thread in this repository can propose issue changes.” No children were created.
The repository orchestrator should submit the split with these titles (no milestone), using the numbered step's acceptance criteria in the canonical plan:
Unify Projects layout access and grouped folder expansion
Reuse upstream project headers and actions within custom sections — blocked by 1.
Converge Projects thread rows on upstream row actions and subscriptions — blocked by 2.
Finish the project-view composition boundary and remove duplicate wiring — blocked by 3.
All children retain #423 / PR #424, the same storage format, hierarchy and controls. Resolve the two presentation questions in the plan before changing either presentation or placement policy. No implementation workers were started.
Read-only verification: clean main at the pinned revision; nightly ancestry confirmed; source and existing regression assertions inspected. No files, branches, worktrees, live userdata, tests or browser sessions were changed/run.
Pinning: pinned items stay in their custom section, orchestrators first. Only upstream's pin command is reused.
Styling: undecided. The owner wants to see upstream's styling as a test first. A side-by-side capture of the fork's sidebar and upstream nightly's sidebar, with the same data, is being made before the decision.
The four steps become top-level issues once styling is decided.
Owner decisions, 2026-11-17, made after seeing the fork and upstream nightly sidebars side by side with the same Replay data:
Rows: keep the fork's compact one-line rows (role, model, effort bars, time, status) and add upstream's pin icon on pinned threads. Upstream's 78 px cards were not chosen.
Pinning: stays in its section, orchestrators first (decided earlier).
The plan steps are now top-level issues, chained by blocked-by: #444 → #445 → #446 → #447. #442 ("Needs you" and "Ready for review" groups) is blocked by this issue.
Activity
tannerpolley commentedon Nov 17, 2026
Design: retain the hierarchy, replace duplicated sidebar foundations
Based on #416 findings, owner decisions, and read-only inspection of main
1e072987b65fc34e7099dc459c33d57c8309252f. Upstream nightlyv0.1.10-nightly.20261116.4800(7d32075e3757ded707914e2450c3e392a9665765) is an ancestor.Done means: custom sections/subsections, mixed child order, dragging, worker families and repository-orchestrator controls remain; Projects and Activity share project identities, folder expansion semantics, common row behavior and commands. Global activity scope remains independent of expansion. Saved sections survive unchanged. No additional per-row subscriptions or unrelated-row renders. The duplicate project thread renderer is deleted after cutover.
Scope: web and desktop's shared web sidebar, local and remote environments. Native mobile and provider/delegation lifecycle redesign are excluded under the fork guide. No plugin framework, second store, new layout schema, or new styling system.
Current footprint
Counts below are physical lines at the pinned main. Shared-file +/− counts are
git diff --numstat 7d32075 HEAD, not an estimate of removable code. Fork-only files have the same added-line count as their length. Paths below are relative toapps/web/src/unless qualified.Fork composition and row presentation
Fork layout and appearance boundary
These helpers need not all change. Layout transport, backend persistence, debugging and text-size preferences are retained boundaries, not rewrite targets.
Edits inside upstream files
The unchanged upstream primitives include
hooks/useSidebarProjectGroups.ts(43 lines),sidebarProjectGrouping.ts(246), andSidebarProjectFolderHeader.tsx. Do not reset shared files wholesale: their deltas include unrelated retained features.Existing fork evidence to extend, not replace: ProjectSections logic/drag/permissions tests (232/245/177 lines), ProjectThread logic (150), BackgroundWork logic (507), Columns logic (21), FamilySeparators logic (172), ThreadProcesses logic (193), WorkExpansion (71), Reveal logic (77), Sidebar.logic.fork (578), SidebarLayoutSync (160), layout client/projection/migration (214/396/80), sidebarDisplay (384), placement/appearance/section-folders (54/34/110). Shared upstream test edits also exist in Sidebar.drag.browser and SidebarDraftBlock.browser (+1 each). Counts describe this feature slice, not the whole fork.
Behavior mapping and ownership
Source references are pinned: fork Sidebar, fork section component, upstream row, upstream folder.
useSidebarProjectGroups,buildSidebarProjectSnapshots,SidebarProjectSnapshot.memberProjects,getProjectOrderKey,sortSidebarV2ProjectGroupsresolveProjectDrop,resolveSubsectionDrop, sharedsidebarSectionChildOrderandapplySidebarLayoutMutationSidebarProjectFolderHeader,sidebarFolderRowId,sidebarFolderExpansionKeys,resolveProjectExpanded,setProjectExpanded;handleProjectFolderContextMenu,startThreadInProjectFolderSidebarThreadRowfolder variant,renderThreadRowassembly,useThreadActions(pinThread,unpinThread,archiveThread),handleThreadContextMenu; project operations use existinguseProjectGroupUpdateattachSidebarTree,navigationParent,sidebarFolderRowId, existing route reveal pathcollectVisibleProjectThreadOrdermust describe the same rendered tree, including collapsed worker families. Keep reveal opening ancestors without changing the activity filter.useRepositoryOrchestratorAction.forThreadcontext-menu augmentationorchestratorsFirst,groupRunningSubagentsByParent,withChildNeeds,SidebarBackgroundWorkRows,useSidebarWorkExpansion, worker provider and ask-mode controls, issue badges, process visibility and family separators. No upstream product replacement is claimed for these.EffortGlyphand model option descriptorssidebarEffortGlyphandSidebarColumnsas the single effort implementation; pass provider models by environment + instance.sidebarProjectScopeKeysand scope/filter handlersOne owned feature component: keep
SidebarProjectSectionsas the fork composition root; it owns sections, drag state and custom worker/control composition. Its existing private helpers/pure logic are acceptable; “one component” does not mean one giant function. Remove its privateSidebarProjectThreadRowafter common-row cutover. Keep only narrow composition inputs in upstream Sidebar/header/row files. Generic row actions remain upstream-owned; fork-only menu items are contributed through the existing augmentation.One layout-state owner: retain the controller created by
createSidebarLayoutClient, mounted once bySidebarLayoutSync. It owns client-side durable layout updates, per-environment revision/write lanes and pending edits.SidebarLayoutServiceremains the server authority.useUiStateStoreis a projection/read facade with forwarding actions (configureSidebarLayoutActions,applySidebarLayoutProjection), not a second writable copy.useSidebarProjectsViewreads that projection and retains only transient dialog/view state. No second sync hook or state copy in rows. Local work expansion and activity filtering remain separate UI concerns.Saved layouts: carry over without a new migration
Retain
SidebarLayout, the server table/migration, revision protocol and storage keys. Preserve section IDs, names, parentId, mixed childOrder, project memberships, color/folder settings, collapse states, projectOrder and Other expansion verbatim. Do not export the visible grouped list back as a replacement layout.Keep
projectSidebarLayoutas the durable-ID → grouped-row projection andsidebarProjectRefsForKeysas the inverse mutation lookup. Grouped rows use upstreamsidebarFolderExpansionKeysfor reading/toggling; commands still target the existing environment-scoped physical project IDs. Re-grouping or a disconnected environment must not delete unrendered memberships. Keep primary-environment precedence already encoded by the projection.Retain existing legacy paths:
migrateSidebarLayout, pending import keys,resolvePendingSidebarImportandcreateLegacySidebarRetirement. Do not reset/import again on component mount. Preserve empty sections and unresolved aliases until their environment arrives; late import must not overwrite subsequent owner moves. The startupsidebarDisplaySnapshotis display-only, never a sync/import source. While it is active, layout writes stay guarded.Use synthetic persisted fixtures covering all fields and disconnected environments. No inspection or modification of the owner's live userdata is needed for this design or implementation.
Performance acceptance
This is a correctness constraint, not a hoped-for optimization:
useQuietThreadShellsfor list/group derivation andsidebarProviderEntriesAtom's stable environment maps. Eliminate the custom project row's broaduseEnvironmentProviders()subscription when moving it to the shared row.useSidebarRowSubscriptionLeasefor visible/active work. Reconcile the fork'suseSidebarThreadProcessesvisibility observer with it; do not mount both independent live-query/observer paths for the same work. No per-row layout streams, provider queries or extra global-store subscriptions.Separately mergeable implementation order
sidebarFolderExpansionKeysand the existing layout action/projection path, and remove duplicate layout derivation from the view hook where applicable. Preserve current data format. Extend existing projection/client/migration tests only where coverage is missing: grouped aliases, disconnected/reconnected environment, empty/nested sections, mixed order, late imports and pending edits. A real behavioral correction gets a failing case first. Capture the same nested synthetic layout before/after if visible expansion behavior changes. This is independently useful without changing row rendering.SidebarProjectSectionswhere feasible; remove dead helpers/imports and duplicated handlers, retaining generic search/Activity/shelves and upstream row commands. Verify Settings, command/keybinding entry points and route reveal against the same action/state owners. Extend existing visible-order/reveal logic tests only if behavior changes; capture the full nested layout before/after and replay the integrated workflow. Compare shared-file numstats against this baseline and account for every removed behavior. This step depends on 3; it is not permission for an unrelated cleanup.Each step can merge with working Projects and Activity views; no half-migrated storage or feature flag scaffolding. Run focused tests while iterating and
agent-heavy vp run check:changedin a background shell before each push. No repo-wide suite. Use existing disposable synthetic fork fixtures; stop the dev server started for the integrated pass. Browser tests are justified only for actual hooks/drag/focus wiring, not visual styling.#423 / PR #424 preservation gate
Main contains the default-effort change (
4b2460d738) and glyph-size follow-up (37c672f4f2). Keep:sidebarEffortGlyph.SidebarEffortMeterrendering the sharedEffortGlyph size="xs"; retain current CSS sizing and do not restore the removed scaling.SidebarBackgroundWork.logic.test.tsregression assertions plus before/after visual evidence against the composer.Check top-level project threads, worker/subagent rows and Activity wherever effort is displayed. Do not copy a pre-fix(web): show default sidebar effort bars #424 renderer from upstream or an older fork commit.
Owner decisions (before presentation changes; recommendations preserve current behavior)
Already decided: keep hierarchy, dragging and controls; use upstream primitives. Those decisions are not reopened. No implementation was performed. Tests/browser captures are implementation gates, not evidence produced by this source-only design.
tannerpolley commentedon Nov 17, 2026
Split handoff
This spans layout compatibility, project headers, thread-row subscriptions and final integration; use four sequential implementor assignments rather than one broad rewrite. I attempted
propose_issue_change(type: "split"), but it returned: “Only a managed thread in this repository can propose issue changes.” No children were created.The repository orchestrator should submit the split with these titles (no milestone), using the numbered step's acceptance criteria in the canonical plan:
All children retain #423 / PR #424, the same storage format, hierarchy and controls. Resolve the two presentation questions in the plan before changing either presentation or placement policy. No implementation workers were started.
Read-only verification: clean main at the pinned revision; nightly ancestry confirmed; source and existing regression assertions inspected. No files, branches, worktrees, live userdata, tests or browser sessions were changed/run.
tannerpolley commentedon Nov 17, 2026
Owner decisions (2026-11-17)
The four steps become top-level issues once styling is decided.
tannerpolley commentedon Nov 17, 2026
Owner decisions, 2026-11-17, made after seeing the fork and upstream nightly sidebars side by side with the same Replay data:
The plan steps are now top-level issues, chained by blocked-by: #444 → #445 → #446 → #447. #442 ("Needs you" and "Ready for review" groups) is blocked by this issue.