Use with your AI agent

super-store ships an agent skill so coding agents write correct StoreValue code without guessing. The skill lives at skills/super-store/ in the repo and the docs themselves are LLM-friendly. Pick the section for your agent below.

Claude Code

The skill is a Claude Code skill: SKILL.md (loaded on demand) plus REFERENCE.md (the full API) and RECIPES.md (collaboration, persistence, undo, React patterns). Install it into your project:

npx degit mertdogar/super-store/skills/super-store .claude/skills/super-store

Commit .claude/skills/super-store/ so the whole team shares it. Claude Code discovers the skill from its YAML frontmatter and loads REFERENCE.md / RECIPES.md only when a task needs them.

Cursor

Cursor reads rules from .cursor/rules/*.mdc. Drop the condensed AGENTS.md there:

mkdir -p .cursor/rules
curl -fsSL https://raw.githubusercontent.com/mertdogar/super-store/main/skills/super-store/AGENTS.md \
  -o .cursor/rules/super-store.mdc

GitHub Copilot

Copilot reads .github/instructions/*.instructions.md. Use the same condensed AGENTS.md:

mkdir -p .github/instructions
curl -fsSL https://raw.githubusercontent.com/mertdogar/super-store/main/skills/super-store/AGENTS.md \
  -o .github/instructions/super-store.instructions.md

Any other agent

AGENTS.md is a single self-contained file most agents can read from the repo root. Place it where your agent looks for instructions:

curl -fsSL https://raw.githubusercontent.com/mertdogar/super-store/main/skills/super-store/AGENTS.md \
  -o AGENTS.md

Point an agent at these docs

Every docs page is published as plain Markdown — append .md to any page URL:

https://mertdogar.github.io/super-store/guide/ai-agents.md

For whole-site context, two manifests are published at the docs root:

File	URL	Use
llms.txt	https://mertdogar.github.io/super-store/llms.txt	A short index of pages with links — point an agent here to navigate.
llms-full.txt	https://mertdogar.github.io/super-store/llms-full.txt	The entire docs concatenated — paste into context for a one-shot answer.

Fetch the full corpus for an agent that takes a URL or a pasted blob:

curl -fsSL https://mertdogar.github.io/super-store/llms-full.txt

What the agent should know

The skill encodes the rules that keep StoreValue code correct. The essentials, in agent-ready form:

• Import from the right package — the primitive and the symbol from @super-store/store, the hooks from @super-store/react:

  import { StoreValue, STORE_ORIGIN } from "@super-store/store"
  import { useStore, useStoreSelector } from "@super-store/react"

• Write through set / update, never by mutating store.value. In-place mutation yields a stale snapshot and, in bound mode, silently fails to converge.

  const pos = new StoreValue({ x: 0, y: 0 })
  pos.set({ x: 10, y: 0 })        // diff-and-patch in one transaction
  pos.update({ x: 20 })           // object stores only: merge plain keys

• set returns true only when data actually changed; a structurally-identical set is a no-op (returns false, no emit).

• Pass the store's pre-bound subscribe / getSnapshot straight through — never wrap them:

  function Coords({ pos }: { pos: StoreValue<{ x: number; y: number }> }) {
    const { x, y } = useStore(pos)
    return <span>{x}, {y}</span>
  }

• For per-field CRDT merge on a sub-object, make it a nested StoreValue, not a plain object — plain nested objects are stored opaquely and replaced wholesale.

• Collaboration and persistence move Yjs update bytes through the sync surface — no yjs import in app code:

  store.onUpdate((update, { local }) => { if (local) bus.send({ update }) })
  bus.on("update", ({ update }) => store.applyUpdate(update))