Skip to main content
Version: 1.3.0

AI Assistant Setup

This page explains how users and agent builders should expose react-mnemonic to coding assistants.

Canonical Source​

The canonical AI-oriented source is this docs section:

  • /docs/ai
  • /docs/ai/invariants
  • /docs/ai/decision-matrix
  • /docs/ai/recipes
  • /docs/ai/anti-patterns

Everything else is a companion surface:

Generated Instruction Packs​

The repo also ships first-party instruction packs generated from the canonical AI docs:

  • AGENTS.md
  • CLAUDE.md
  • .claude/rules/react-mnemonic-persistence.md
  • .claude/rules/durable-state-checklist.md
  • .cursor/rules/react-mnemonic-persistence.mdc
  • .cursor/rules/react-mnemonic-docs.mdc
  • .github/copilot-instructions.md
  • .github/instructions/react-mnemonic-persistence.instructions.md
  • .github/instructions/react-mnemonic-docs.instructions.md

Those files are projections over the same persistence contract rather than independent sources of truth.

Lowest-Friction Retrieval​

When an assistant has only HTTP or search access, give it these URLs first:

  1. /llms.txt
  2. /docs/ai
  3. /llms-full.txt
  4. /ai-contract.json

That ordering keeps the first context window compact while still leaving a long-form export available when the task is more complex.

DeepWiki​

The repo ships .devin/wiki.json to steer DeepWiki toward the files that matter most for correct persistence behavior:

  • canonical AI docs
  • README.md
  • the public export surface in src/index.ts
  • the hook, provider, and type definitions that define runtime semantics
  • the deeper guides for migrations, SSR, clearable values, and custom storage

If you re-index the repository, keep that file up to date with any future source-of-truth moves.

Validated MCP-Friendly Path​

The lowest-friction MCP setup is exposing this repository through a filesystem-capable MCP server or any equivalent local-docs MCP layer.

Mount these paths:

  • website/docs/ai/index.md
  • website/docs/ai/invariants.md
  • website/docs/ai/decision-matrix.md
  • website/docs/ai/recipes.md
  • website/docs/ai/anti-patterns.md
  • website/static/llms.txt
  • website/static/llms-full.txt
  • website/static/ai-contract.json
  • src/Mnemonic/use.ts
  • src/Mnemonic/provider.tsx
  • src/Mnemonic/types.ts

Why this path is validated:

  • the website build regenerates llms.txt, llms-full.txt, and ai-contract.json before Docusaurus runs
  • the generation step fails fast if any canonical AI source file is missing
  • the docs build already fails on broken links, so the published AI entry points stay connected to the site navigation

This gives an MCP client both the compact docs surfaces and the source files that define the contract underneath them.

Hosted Retrieval Path​

If your assistant can fetch remote files, use the published GitHub Pages URLs:

  • https://thirtytwobits.github.io/react-mnemonic/llms.txt
  • https://thirtytwobits.github.io/react-mnemonic/llms-full.txt
  • https://thirtytwobits.github.io/react-mnemonic/ai-contract.json
  • https://thirtytwobits.github.io/react-mnemonic/docs/1.3.0

Use the hosted path when you want versioned, read-only retrieval without mounting the repository locally.

Versioned Docs Behavior​

The site now uses Docusaurus versioned docs:

  • the latest released docs live at /docs
  • unreleased work from main lives at /docs/next
  • released snapshots live at /docs/<version>

Example released path:

  • https://thirtytwobits.github.io/react-mnemonic/docs/1.2.0-beta1

Current AI retrieval surfaces remain latest-only:

  • /llms.txt
  • /llms-full.txt
  • /ai-contract.json

That means version-specific AI retrieval should currently use the versioned docs pages themselves rather than the root llms exports. Version-specific llms surfaces can be added later if they become necessary.

Maintenance​

Keep the surfaces aligned this way:

  • edit the canonical prose in website/docs/ai/*
  • regenerate companion assets and instruction packs via npm run docs:ai
  • cut new released doc snapshots via npm run docs:version -- <version> when needed
  • use npm run ai:check in CI or before commits to catch drift in generated AI artifacts and instruction packs
  • keep context7.json aligned with the same high-signal docs and public API entry points
  • keep the DeepWiki priorities in .devin/wiki.json aligned with the same sources

The goal is simple: agents should load one canonical contract and then choose the right persistence behavior without inventing missing semantics.