Skill: state-management

Fork

State management patterns and implementation. Trigger when the user wants to manage global state, use Redux, Zustand, or other solutions.

Configuration

Property	Value
Context	fork
Allowed tools	Read, Write, Edit, Glob, Grep
Keywords	state, management

Detailed description

State Management (pointer)

Library APIs, install steps and idiomatic code drift on each release and are canonical at:

• Zustand — docs.pmnd.rs/zustand (recommended default; 1.2kb, hooks-first)
• Redux Toolkit — redux-toolkit.js.org (enterprise scale, devtools)
• Jotai — jotai.org (atomic model, 2kb)
• TanStack Query — tanstack.com/query (server state, NOT global client state)
• React 19 built-ins — react.dev (useReducer, useOptimistic, use(Context), Actions)

Decision tree (version-agnostic)

Need shared state?
├── No → useState/useReducer
└── Yes →
    ├── Server state (fetch + cache + invalidate) → TanStack Query / SWR
    ├── Forms → React Hook Form
    ├── Small client state (<5 stores) → Zustand
    ├── Large client state (>5 stores) → Redux Toolkit
    └── Theme/Auth singletons → Context + useReducer

Foundation discipline (keep across releases)

• Client vs server state: never put server state in a client store (Zustand/Redux). Use TanStack Query for fetched data; Zustand/Redux for UI state, filters, drafts.
• Domain separation: one store per domain, not one global store. Easier to test and lazy-load.
• Granular selectors: subscribe to slices, not whole stores — see [[dev-react-perf]] for re-render audit when a store change cascades.

See also

• dev-react-perf skill — re-render diagnosis when selectors are too broad
• dev-tdd — store reducers/selectors test well in isolation (no DOM needed)

Automatic triggering

This skill is automatically activated when:

• The matching keywords are detected in the conversation
• The task context matches the skill's domain

Triggering examples

• "I want to state..."
• "I want to management..."

Context fork

Fork means the skill runs in an isolated context:

• Does not pollute the main conversation
• Results are returned cleanly
• Ideal for autonomous tasks

---

See also

• Back to skills
• Architecture