langfuse/frontend-large-feature-architecture
| Architect large or state-heavy Langfuse frontend features. Use for virtualized lists, large tables, controllers, Zustand stores, row selection, high-frequency state, rendering performance, or before adding useEffect, useMemo, useCallback, or form defaults derived from loaded data.
npx skills add https://github.com/langfuse/langfuse --skill frontend-large-feature-architecture
Use this skill when building, changing, or refactoring a large frontend
surface.
In this skill, "controller" means a component or hook that owns feature logic:
data fetching, view state, table/list state, effects, actions, and expensive
rendering. The problem is not the name; it is one place owning too many
changing responsibilities.
When a feature grows, split rendering from logic. Most components should be
view-only. Data preparation should be pure. Complex user actions should live in
external async functions or local-store actions. Effects are integration
boundaries, not the normal way to derive state.
For the full rules, read
references/big-feature-rules.md.
UI is a pure function of state. Do not use useEffect to derive, prepare, or
sync data:
const title = data?.name ?? ""),never mirrored into state by an effect. Client state that refers to server
data stores only the user's intent (an id) and derives the effective value
by merging with query data in render.
data-preparer/controller fetches and renders a loading state (prefer a
skeleton; <Spinner /> as minimal fallback); the inner component receives
the loaded value as an initialValue prop and seeds useState(initialValue).
Do not render UI before its data is ready.
still loading, the form lives deeper in the tree.
query client — not inside components, not behind effects.
useCallback/useMemo are premature optimization; fix re-render problems bysplitting components, not memoizing.
useEffect is legitimate only for DOM/browser API integration: no (orminimal) dependencies, one concern, and a cleanup function.
For the golden example and full reasoning, read
references/react-without-useeffect.md.
router/filter hooks.
store. Create it with lazy useState, not useMemo.
frequently changing state directly in provider values.
effects, subscriptions, data loading, and workflows outside expensive views.
src/components/* exports should stay context-free or receiveexplicit props. Put view-scoped Zustand consumers in src/features/*.
README.md owner map.state boundaries, action workflows, data-preparation seams, and render
boundaries. Hundreds of legacy cases remain — do not preserve a legacy shape
just to keep a change small.
Prefer a local vanilla Zustand store for large or high-frequency feature state.
The store is created by the page/view and destroyed on unmount.
Use selectors that return primitives or stable references. If a component needs
multiple values, use shallow selector helpers or split subscriptions so one
changing field does not rerender unrelated UI.
Complex user workflows should live in actions/*.ts files or store actions.
The component wires hooks and passes dependencies; the action owns the workflow.
For a complete effect audit or component-splitting recipes, use
../refactor-react-effects/SKILL.md.
references/big-feature-rules.md.
loaded data, initialValue props, forms, actions outside React, and why not
to memoize — read
references/react-without-useeffect.md.
read references/virtualized-lists.md.
references/local-feature-state.md.
README.md owner maps, readreferences/feature-readmes.md.
pattern, read
references/controller-migration.md.
This is the default reference for traces/observations tables, sessions,
experiments, prompts, evals, datasets, and other controller-heavy surfaces.
Take langfuse/frontend-large-feature-architecture from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.