v3.0 — THE BRAIN · 2026-05-28

ux-skill v3.0 vs UI/UX Pro Max — engine vs atlas, scored honestly.

UI/UX Pro Max is the finished atlas of UI knowledge. ux-skill v3.0 is the engine that decides what UI should exist. Different categories. Here's where each wins, where the other doesn't, and why we picked the engine path.

The honest framing

Both projects appear in the same search results ("Claude Code UX skill", "AI design plugin", "Cursor design system"). They're not the same thing. ChatGPT (in a review we asked it to do honestly) summed up the distinction:

UI UX Pro Max = finished atlas of UI knowledge. UX Skill = engine that decides what UI should exist. Atlas beats engine in immediate usefulness. Engine beats atlas in future value if built correctly.

That's accurate. If you install both today and use them for production work, Pro Max has more "finished surfaces" you can pull from. We won't pretend otherwise.

What Pro Max does best

Breadth of finished skills. Banner design, slide design, expanded data sets — Pro Max v2.5 covers a wider net of "I need a thing, give me a thing."
Maturity signal. They've been at v2.x stable for months. We just hit v3.0 stable hours ago.
Immediate usability. If you want to install once and never think about how design happens, Pro Max ships that flow today.
Engineering completeness as a static system. Their atlas is internally consistent. It's done.

What v3.0 does that Pro Max doesn't

Deterministic 7-axis synthesizer. A brief → 7 axis values → a complete novel design language per call. Pro Max picks from a catalogue. v3.0 distills from one.
Triple-mode dispatch. strict_brand / brand_anchor / pure_synthesis — three different speeds of customization, auto-dispatched from what's in the brief.
Decisions ledger + recommender re-rank. Every call is logged locally. The next call reads that log and biases ranking by past wins. Real closed feedback loop. Pro Max generates once and forgets.
Axis interaction matrix. Documented conflict resolution between competing axes (density × formality for spacing, geometry × formality for radius, etc.). Pro Max's tokens are static.
Quality gate at 65. /ux-evolve auto-iterates polish until score ≥ 90 or plateau. Refuses to ship below 65 unless --force. Pro Max has no such gate.
Responsive-by-construction layout primitives. 7 primitives using auto-fit minmax(min(N, 100%), 1fr) + container queries. Zero size-based @media queries. Bulletproof at every viewport.
145 deterministic anti-pattern rules. Sub-50ms scan. Pro Max has rules too but ours are individually URL-addressable at /anti-patterns.html.
160 brand specs as training data. The same files Pro Max would treat as templates, we treat as vocabulary for synthesis. Same data, completely different role.
Zero LLM in the engine. Same brief → same output. Reproducible across machines. Pro Max's stance on LLM use varies by command.
18 MCP tools. Exposed over stdio for Claude Desktop, Cursor, generic agents. Pro Max doesn't ship an MCP server.

The scoring

Dimension	UI/UX Pro Max v2.5	ux-skill v3.0
Finished skill count	~25	23
Catalogue entries	~800 (estimated)	1,182
Brand specs	—	160
Anti-pattern rules	~50 (estimated)	145 (deterministic)
IDE support	Claude Code	17 IDEs
Languages (homepage / README)	1 (en)	17 locales
MCP server	No	18 tools
Generative engine	Pick from catalogue	7-axis synthesizer
Closed feedback loop	No	Decisions ledger drives re-rank
Auto-iteration / polish loop	No	`/ux-evolve` 5-round loop
Quality gate	No	65 (force-overridable)
Determinism guarantees	Mixed	Same brief → same output, locked
Telemetry	—	None
LLM dependency	Yes (varies)	None in the engine
Local stats dashboard	No	`uxskill stats --html`

Where the comparison is unfair (in their favor)

Pro Max has more public adoption / GitHub stars today.
Pro Max has a more polished documentation site (we've been heads-down on engine work).
Pro Max's "banner design" / "slide design" skills target use-cases we don't ship yet.

Where the comparison is unfair (in our favor)

We can synthesize a novel design language for a fictional brand. Pro Max needs the brand in its catalogue.
We can learn from your local history. Pro Max can't.
We work in 17 IDEs. Pro Max is Claude-Code-centric.
Our engine has zero ongoing costs. Pro Max has LLM call costs per command.

Which should you install?

Both. They're complementary, not exclusive. Use Pro Max when you want "here's a finished surface, ship it." Use ux-skill when you want "here's a constraint set, generate something that meets it AND learns from feedback."

If forced to pick one:

You ship one-off marketing pages with simple briefs → Pro Max
You ship products with recurring design decisions → ux-skill (the closed loop pays off after 20-30 decisions)
You work in something other than Claude Code → ux-skill (17 IDEs vs 1)
You can't have LLM costs in your generation path → ux-skill (offline + deterministic)
You need the longest catalogue today → Pro Max
You need the system to get smarter from your usage → ux-skill

The architectural argument

This blog post wasn't supposed to be marketing. We've been honest about where Pro Max wins. The real argument is structural:

An atlas reaches diminishing returns. After the 200th finished surface, the user can't find the one they want because there are too many. The catalogue becomes search-hostile.

An engine has no such ceiling. As you add brand specs, the synthesizer's vocabulary grows. As users log decisions, the recommender gets sharper for their bucket. Adoption compounds quality.

v3.0 bets on the engine path. If we're wrong, we'll still have a respectable catalogue. If we're right, the math wins.

Try them side-by-side

# Install ux-skill v3.0
pip install uxskill
uxskill init

# Generate a synthesized fintech dashboard
uxskill synthesize \
  --industry fintech-payments \
  --tone bold --tone serious

# Then install Pro Max via Claude Code marketplace and compare outputs
# for the same brief. The differences will be visible.