The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Apr 18, 2026 0:27am

PR Summary

Medium Risk
Changes modify keyboard/focus behavior in a Radix dropdown and memoization/selection logic in React Query hooks; regressions could impact chat input UX and list rendering, but scope is localized and does not touch auth or data writes.

Overview
Fixes the chat “plus menu” resource dropdown so typing in the search input no longer loses focus, by keeping the default menu tree always mounted (toggled via hidden) and rendering filtered results outside Radix’s Collection while adding activeIndex-based keyboard navigation and scroll syncing.

Stabilizes render identity in several places to reduce unnecessary re-renders during typing: useContextManagement now avoids emitting new empty-array references, WorkspaceHeader is wrapped in memo, and React Query hooks (useWorkflowMap, useFolderMap, useWorkspacesQuery) hoist select functions to module scope.

Docs UI tweaks: video lightboxes now open at the clicked video’s current playback time (startTime + useLayoutEffect seek), AnimatedBlocks is removed, and small cleanup refactors switch to cn() and simplify language selection derivation.

^{Reviewed by Cursor Bugbot for commit 06c8066. Configure here.}

Greptile Summary

This PR fixes the Radix FocusScope bug that stole keyboard focus from the @-mention search input whenever filtered menu items mounted or unmounted during typing. The fix always-mounts the default Radix tree under the HTML hidden attribute and renders filtered results as plain <button role="menuitem"> elements that bypass Radix's Collection — preventing FocusScope from triggering on type-driven DOM mutations. Alongside this, the PR stabilizes render identity by hoisting inline select arrow functions in three React Query hooks to module scope (enabling structural sharing), applying a functional-updater bail-out in clearContexts, and wrapping WorkspaceHeader in React.memo. The docs side adds a lightbox video seek-to-position feature and cleans up a state-mirroring anti-pattern in the language dropdown.

Confidence Score: 5/5

Safe to merge — no P0 or P1 issues found across any changed file.

All findings are P2 or lower. The architectural fix is sound (always-mounted hidden tree + plain buttons avoids Collection mutations), the React Query select hoisting is a textbook optimization, the functional-updater bail-out is correct, and the lightbox seek uses useLayoutEffect at the right time. No regressions introduced.

No files require special attention.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User types in search input] --> B{filteredItems !== null?}
    B -- No/empty query --> C[div NOT hidden\nRadix DropdownMenuItems visible\nParticipate in Collection]
    B -- Has query --> D[div hidden=true\nRadix items hidden, still in DOM\nDO NOT cause Collection mutations]
    D --> E[Plain buttons rendered\nrole=menuitem, no Collection registration\nNo FocusScope trigger]
    E --> F[DOM focus stays on search input\nactiveIndex moves visual highlight]
    F --> G{Keyboard nav}
    G -- ArrowUp/Down --> H[setActiveIndex, scrollIntoView]
    G -- Enter/Tab --> I[handleSelect → setOpen false]
    I --> J[onCloseAutoFocus\nfocuses textarea]
    C --> K{Keyboard nav}
    K -- ArrowDown --> L[getVisibleMenuItems 0 .focus\nDOM focus moves into Radix tree]

_{Reviews (2): Last reviewed commit: "fix(chat): apply same setState bail-out ..." | Re-trigger Greptile}

Thanks @greptile-apps — looked into both and want to leave my reasoning on the record.

1. clearContexts bail-out — applied in 06c8066. Agreed it's worth the one-line consistency fix: clearContexts is part of the hook's public API and callers can't reasonably check emptiness before invoking it, so making it safe against an already-empty list matches the invariant we just established in the message effect.

2. handleSelect useCallback — intentionally declining. The reasoning:

• WorkflowFolderTreeItems (the only child that receives handleSelect directly as a prop) is a plain function component, not React.memo'd — see add-resource-dropdown.tsx:190. It re-renders whenever PlusMenuDropdown renders regardless of callback identity, so useCallback here provides zero runtime benefit today.
• The other handleSelect usages in this file are inline arrow wrappers inside DropdownMenuItem onClick handlers, which similarly don't benefit from stability.
• The doOpen comparison isn't quite apples-to-apples: doOpen is memoized because it's consumed by useImperativeHandle's dependency array — an identity-sensitive consumer. handleSelect has no such consumer.
• This project's you-might-not-need-a-callback cleanup pass (sim/.cursor/commands/you-might-not-need-a-callback.md) explicitly flags "useCallback + React.memo unnecessarily — only optimize when you've measured a problem", which is what we'd be doing here for a memoization that doesn't exist yet.

Happy to revisit if/when WorkflowFolderTreeItems gets wrapped in memo — that PR is the right place to memoize its callers.

Both points are well-reasoned — thanks for the detailed writeup.

On handleSelect: you're right that I overcounted the similarity with doOpen. The useImperativeHandle dependency is genuinely identity-sensitive in a way that handleSelect's callers aren't, and without a memo boundary to cross, the useCallback is dead weight. The cursor rules doc is a good tie-breaker too. Agreed — defer until WorkflowFolderTreeItems is memoized.

Tip: You can customize Greptile's behavior for this repo with .greptile/rules and .greptile/config.

@greptile

@cursor review