Desktop Sidebar Resize & Collapse (feedback #12)

Date: 2026-07-28

Goal

Properly implement algosu-feedback #12 ("Notion-style sidebar resize and collapse") in the real usage layout. A single-issue sprint.

Background — mis-resolution correction #12 was closed as "resolved" by Sprint 263 #510 (study-room / sidebar UX improvements), but the user pointed out it was not applied on the actual screen. Investigation showed #510 had implemented resize/collapse in StudySidebar.tsx, a dead-code file referenced only by a single JSDoc @related comment (never imported/rendered anywhere). The real desktop sidebar is AppLayout.tsx (fixed w-[220px]), which #510 never touched. → Reopened #12 and handled it properly as Sprint 265.

Additional user requirement: apply a size limit on drag resize (prevent over-expand/over-shrink).

Decisions

D1. Implementation site = the real AppLayout.tsx, remove dead-code StudySidebar.tsx (#524)

Implement resize/collapse in the actually-rendered desktop sidebar (AppLayout.tsx). Since StudySidebar.tsx — where #510 put the feature — is dead code with 0 imports, do not reuse it; delete it (preventing the same feature scattered across two places). Factor the resize logic into a reusable custom hook useSidebarResize.ts.

D2. Size limit = clamp min 180 / max 400 / default 220 (user requirement)

Apply a clamp to drag resize: min 180px / max 400px / default 220px. Clamp at both the pointer-move moment and the localStorage-load moment, so an out-of-range stored value is always corrected to a valid width. This constant pattern was ported into the real layout using clampWidth (MIN 180 / MAX 400 / DEFAULT 220) from the dead-code StudySidebar.tsx as the reference implementation.

D3. Resolve collapsed-header overlap = hide logo, reposition toggle (Critic P2)

The Critic caught a click-area overlap between the header logo link and the expand toggle (right-2, 24px) in the collapsed state (56px, P2) — clicking the logo fired the expand toggle instead of navigating to the dashboard. At 56px the logo and toggle cannot physically coexist, so I hid the logo link, centered the header, and dropped the toggle's absolute positioning when collapsed, eliminating the overlap at its source (consistent with the Critic's "hide/reposition one" recommendation).

Implementation

• #524 1d0958db (feat/sprint-265-sidebar-resize):
  • New frontend/.../useSidebarResize.ts — drag-resize hook (clamp 180~400, width save/load in localStorage)
  • AppLayout.tsx — integrated the sidebar drag handle / collapse toggle, wired the hardcoded md:ml-[220px] offset to the dynamic width
  • Removed StudySidebar.tsx (dead code)
  • Palette impl commits 49cea675·4f8f2b67 + Critic P2 direct fix by Oracle 017bf5f9

Verification (physical facts): Oracle direct gate — tsc delta 0 (globals.css TS2882 is an unchanged baseline, CI green) · ESLint 0 errors (warnings are the react/forbid-dom-props baseline; dynamic-width inline style is unavoidable) · jest 1860 pass/0 fail (incl. AppLayout.test.tsx 10 pass). Critic re-review CLEAN (P2 gone). Squash-merged 1d0958db, confirmed on origin/main.

Incidents

1. #12 mis-resolution / reopen: Sprint 263 #510 implemented #12 in dead code and closed it → the non-application was found via the user's report → reopened and handled properly this sprint. Separately, #3 (tab UUID)·#6 (AI-report padding) were initially misclassified as unresolved, but measurement showed they were already resolved by PR #505 03c13ef1 → corrected and closed.
2. Palette re-delegation stall (ACP background): after the Critic P2 finding, the Palette fix re-delegation stalled for 21 minutes with 0 commits / working-tree changes → rather than wait longer, Oracle fixed P2 directly (a small FE change).
3. Auto-verdict false positive (reconfirmed): the .verdict auto-parser labeled #524's first review "CLEAN-hint", but the codex verdict section in the log had 1 P2 → caught by Oracle's direct log read. The re-review had 0 findings = CLEAN confirmed.
4. Local main stale misjudgment: during the close procedure git merge --ff-only returned "Already up to date", but the gh-API measurement showed local lagging origin/main → stale remote-tracking ref. Reconciled with fetch + reset --hard origin/main.