The free repo scan

npx @getsystemix/cli audit is the free scan you run yourself, in your own repo. Point it at a codebase with no design system defined and it tells you what you already have, where it's inconsistent, and hands you the seed of the fix. It only reads — every file it would create is proposed for your approval.

(Want it done for you, with a scored report emailed back? That's the paid AI-Readiness Audit (€249) — same idea, our hands, in 24–48h.)

Install it

npx @getsystemix/cli audit

Zero setup: this vendors one read-only skill (design-audit) into .claude/skills/. No config, no MCP, no wizard.

Run it

Open Claude Code in the repo and run:

/design-audit            # audit the whole repo
/design-audit src/       # scope to a directory

What it does

Four read-only stages:

1. Find your de-facto system — scans CSS + TSX/JSX for colour, spacing, radius, and type literals, detects any token source you already have (globals.css custom properties, a Tailwind config), and clusters the raw values into an inferred palette and scale.
2. Grade the drift — every raw value is graded Critical (a token exists; reference it), Warning (recurs; should become a token), or Info (a plausibly intentional one-off).
3. Spot the slop — near-duplicate components (three Buttons, a Modal rebuilt inline), mixed component libraries, one-off screens, inconsistent prop patterns.
4. Score + propose the fix — a health score, plus a proposed design/tokens.css and design/guardrails.mdx starter seeded from what it found.

What you get

A report (design-audit.md) and two proposed starters:

design/
  tokens.css       ← the inferred palette + scale, as CSS custom properties
  guardrails.mdx   ← the rules to enforce next, pre-filled from the findings

Nothing is written without your approval — the starters carry your signature, not the tool's defaults. Once you approve guardrails.mdx, /drift-report keeps the code true to it on every change.