CLUX — CLI UX Evaluator

Get an AI-powered UX score for any command-line tool.

Paste your CLI's --help output or man page. CLUX evaluates it across 8 heuristic dimensions and gives you a CLUX Score — a single number that shows how usable your CLI actually is, and what to fix first.

→ Try it live ← (update with your Vercel URL)

---

What gets evaluated

Dimension	What it measures
📖 Learnability	Discoverability, help quality, naming clarity
⚠️ Error Tolerance	Error message quality, recovery guidance
⚡ Efficiency	Shortcuts, --json output, pipe-friendliness
🛡️ Safety	Protection from destructive ops, --dry-run
🐚 UNIX Compliance	Exit codes, stdin/stdout, composability
✨ Pleasantness	Consistent naming, output formatting
🔒 Security	Credential handling, secrets exposure
♿ Accessibility	Color-safe output, --no-color flag

Weights shift based on whether your CLI is Human-first or Scripting-first — because a confirmation prompt is good UX for humans and a pipeline-breaking bug for scripts.

Based on Professor Daniel Jackson's usability principles + the UNIX philosophy.In t

---

Self-host in 2 minutes

git clone https://github.com/amylily1011/clux
cd clux
npm install
cp .env.example .env.local
# Edit .env.local — required: SUPABASE_URL, SUPABASE_ANON_KEY, and one AI provider key
npm run dev

Open http://localhost:3000.

.env.local requires:

• SUPABASE_URL + SUPABASE_ANON_KEY — for share links (create a free project at supabase.com)
• One AI provider key — Anthropic is the default (ANTHROPIC_API_KEY); Google (GOOGLE_GENERATIVE_AI_API_KEY) and Kimi (MOONSHOT_API_KEY) are also supported. Set AI_PROVIDER to match.

Deploy to Vercel

---

How it works

1. Choose whether your CLI targets humans or scripts — this shifts the evaluation weights
2. Paste your CLI's help text, man page, or error output — or just type a bare command name (evaluated from AI training knowledge)
3. Optionally paste your org's CLI conventions to get a compliance check alongside the UX score
4. The AI evaluates it against 8 dimensions using calibrated rubrics
5. Get a radar chart, a CLUX Score, and per-dimension feedback with actionable fixes
6. Past evaluations are saved locally in your browser; share any result via URL — no account needed

---

Roadmap

• CI/CD API — fail builds when CLUX score drops below threshold
• PDF reports
• Score tracking over time

Found a bug or have a feature idea? Open an issue.

---

Stack

• Next.js — full-stack framework
• Vercel AI SDK — AI evaluation engine (Anthropic/Google/Kimi)
• Supabase — share link persistence
• Recharts — radar chart
• shadcn/ui — UI components
• Vercel — deployment