rnd — React Native Directory CLI

Query reactnative.directory and audit React Native projects from the terminal. Designed for AI agents (Claude Code, Codex, Cursor, Windsurf) and humans alike.

Quick start

Pick your favorite package manager:

# Homebrew (macOS / Linux)
brew install Quegenx/tap/rnd

# npm / npx / bunx (any platform with Node)
npx react-native-directory-cli

# Cargo (any platform with Rust)
cargo install react-native-directory-cli

# Or install the AI-agent skill for Claude Code / Codex / Cursor / 40+ agents
npx skills add Quegenx/react-native-directory-cli

All of the above install the rnd binary. Run rnd --help to get started.

Then try:

$ rnd alternatives @react-native-async-storage/async-storage --limit 3
Candidates matching @react-native-async-storage/async-storage:

[topic] expo-sqlite                ★48.7k   241.9k/wk   [ios,and]
[topic] react-native-mmkv-storage  ★1.7k    10.3k/wk    [ios,and]
[topic] realm                      ★6.0k    44.5k/wk    [ios,and]

What it does

• Search 2,400+ React Native and Expo packages
• Audit a project's package.json + source files for unmaintained / archived / missing-new-arch deps
• Find alternatives with a 4-tier algorithm (directory-curated → topics → name tokens → description keywords)
• Discover quietly-great packages that don't dominate download charts
• Facts only, no opinions — the CLI emits ground truth; the consuming agent (or you) reasons about it

Install

The CLI

Pick whichever you already use:

All paths install the rnd binary. Verify:

rnd --version

If rnd: command not found after cargo install, add ~/.cargo/bin to your $PATH:

The AI-agent skill

Teaches Claude Code, Codex, Cursor, Windsurf, and 40+ other agents how to use rnd. Install with the Vercel skills CLI:

npx skills add Quegenx/react-native-directory-cli

This installs the skill to your project's .agents/skills/ directory. It'll trigger automatically when you discuss RN/Expo packages, audit dependencies, or ask about alternatives.

Works on macOS, Linux, and Windows (Node.js required).

Both at once

# CLI
cargo install --git https://github.com/Quegenx/react-native-directory-cli rnd

# Skill
npx skills add Quegenx/react-native-directory-cli

Commands

Global flags: --json (force JSON) · --pretty (force pretty) · --refresh (bypass cache) · --quiet (suppress non-essential output)

Auto-JSON for AI agents

rnd detects CLAUDECODE, CODEX, and CODEX_CI environment variables and auto-emits JSON. Pipe the output through jq or parse it directly — no extra flags needed.

# Inside Claude Code — returns JSON automatically
rnd analyze . | jq '.package_findings[] | select(.flags | index("unmaintained"))'

rnd analyze — the killer feature

Point it at any React Native project:

$ rnd analyze .
Scanned 335 source files, 58 deps, 4 flagged

⚠ react-native-easing-gradient — used in 1 file
    - unmaintained
    already in deps that could cover this: expo-linear-gradient
    repo: https://github.com/phamfoo/react-native-easing-gradient
    stats: ★258  12.6k/wk  score 0  pushed 2024-02-27
    src/components/shell/layout/.../animated-header-panel-scroll-view.tsx:11

For each flagged dep it emits:

• flags — archived, unmaintained, no new-arch support, no TypeScript types
• alternatives_in_project — packages already in your package.json that could replace the flagged one (the highest-value signal for agents)
• github_url, license, last_pushed_at — activity data
• used_in — exact file:line locations where the dep is imported

rnd alternatives — 4-tier matching

Candidates are tagged by how they matched:

Hard filters always apply:

• If target has native code, candidates must too
• If target is pure-JS, native alternatives are allowed (you might wrap them yourself)
• Archived/unmaintained candidates are excluded

Design principle

The CLI emits facts only:

• Directory metadata (stars, downloads, topics, score, license, activity dates)
• Flags from the directory's own maintenance data
• File:line locations of imports in your codebase

It does not:

• Hardcode opinions ("use MMKV instead of AsyncStorage")
• Compute arbitrary scores with made-up weights
• Suggest swaps based on training data that rots

Agents (and humans) reason better than any hardcoded registry. The CLI gives you ground truth; you decide what to do.

AI agent skill

A Claude Code skill ships in .claude/skills/rnd-cli/. It triggers on RN/Expo package discussions, audit requests, "is X maintained?" questions, and migration planning. Agents auto-reach for rnd without explicit prompting.

Built with

• clap 4 — CLI framework
• reqwest + tokio — API client
• regex + walkdir — source scanning
• owo-colors + indicatif — pretty output

Cache

The directory dump (~2.4 MB, 2,419 packages) is cached for 24 hours at your OS cache location:

• macOS: ~/Library/Caches/dev.rnd.rnd/libraries.json
• Linux: ~/.cache/rnd/libraries.json
• Windows: %LOCALAPPDATA%\rnd\rnd\cache\libraries.json

Bypass with --refresh, inspect with rnd cache info, clear with rnd cache clear.

Repository layout

Everything in the repo has a purpose — here's a quick map for the curious:

react-native-directory-cli/
├── src/                        Rust source (main binary + modules)
├── Cargo.toml / Cargo.lock     Rust package manifest + lockfile
├── README.md                   You are here
├── dist-workspace.toml         cargo-dist config for releases
├── .github/workflows/
│   └── release.yml             CI: builds binaries for macOS/Linux/Windows on tag push
├── .claude/skills/rnd-cli/
│   └── SKILL.md                Skill for local Claude Code discovery
└── skills/rnd-cli/
    ├── SKILL.md                Same skill, at the Vercel skills CLI convention path
    ├── metadata.json           Skill metadata
    └── README.md               Skill-specific docs

The two SKILL.md files are intentionally duplicated — each sits where its ecosystem expects to find it. They should be kept in sync.

License

MIT.