A design-judgment skill for AI coding agents.
Show it a screen you built. It hands back a working page where every design change it suggests sits behind its own toggle, with the reasoning next to each one. You keep the ones you agree with, kill the rest, and sharpen your own design eye in the process.
An independent project. Not affiliated with, endorsed by, or sponsored by Jony Ive, LoveFrom, or Apple Inc. Quotes belong to their speakers and are cited to their published sources.
The one rule everything else serves: Immersion must be earned by naturalness, not manufactured by spectacle.
Any competent model can now generate a visually impressive interface on demand. Gradients, glass, glow, parallax, and motion are free. What is no longer free is a human being successfully doing the thing they came to do.
So the failure this skill prevents is specific. It is not ugliness. It is spectacle that crowds out use: an interface that photographs beautifully, demos beautifully, and quietly fails the person holding it.
Jony Ive gives the clause that makes this enforceable rather than merely tasteful: "if something doesn't work, it's ugly." Not compromised, not flawed. Ugly. That removes the aesthetic defense entirely.
The main use. Point it at a page you built and watch a design philosophy get applied to it, one idea at a time, with the reasoning attached.
"here's my pricing page, run it through the Ive lens"
"https://mysite.com/pricing, what would Jony Ive change and why?"
"critique my dashboard before I ship it"
What comes back is not a list of tips. It is a working page showing your actual interface with each proposed change behind its own checkbox, and beside each checkbox, the principle that drives it and why he would have made that call.
Works on a component, a live URL, or a whole page. If all you have is a screenshot it says so honestly and gives you an annotated critique instead, rather than pretending a picture can be toggled.
It is not a hands-off design service. Told to "make it good", it produces the same interchangeable output as anything else, only with citations attached, which is worse than not using it at all. The point is not the fixes. It is that you watch a recorded human philosophy operate on your own work, one principle at a time, and form a view on each move. What accumulates is your judgment, not this document's. That is why the toggles exist and why they stay unbundled: bundle the changes and you get a verdict on a package instead of an argument you can evaluate.
npx skills add TheBaffledKing/jonyThat command finds which agents you already have installed and writes the skill into each
one's skills directory. Add -g to install it globally, across every project, instead of
only the current one.
npx skills add TheBaffledKing/jony -gTo clone it yourself, put it in your host's skills directory:
| Host | Clone into |
|---|---|
| Claude Code | ~/.claude/skills/jony |
| opencode | ~/.config/opencode/skills/jony |
| Cursor | ~/.cursor/skills/jony |
| Codex | ~/.codex/skills/jony |
| Kiro | ~/.kiro/skills/jony |
| Factory Droid | ~/.factory/skills/jony |
| Any host reading the open standard | ~/.agents/skills/jony |
git clone https://github.com/TheBaffledKing/jony.git ~/.claude/skills/jonyConfirmed working on Claude Code, opencode, Cursor, Codex, Kiro, and Factory Droid, plus any
host that implements the open Agent Skills standard. Windsurf is the exception: it does not
read the SKILL.md format, so you must paste SKILL.md into .windsurfrules or AGENTS.md
by hand, and the whole thing sits in context at once.
The skill routes on what you are actually asking for.
| You want | Mode | What you get |
|---|---|---|
| A verdict on something that exists | CRITIQUE | Findings with named mechanisms of harm, checked against each principle's documented limits |
| To make something better | LOOP | An Ive Cut. See below |
| Permission to ship | GATE | A 15-question checklist where a failure blocks and three questions have no aesthetic override |
Most design tools give you advice. This one gives you a page.
Above: the real artifact, produced against the sample card in examples/fixture/. The
working file is examples/ive-cut-example.html, with its
move list in examples/ive-cut-example-moves.md. Open
the HTML in any browser and click through it.
LOOP mode does not describe its suggestions. It builds a real, working preview of your actual interface, with every proposed change behind its own toggle:
- Current is the default on load. You see your own work first, never a version of it.
- One checkbox per change. Each changes exactly one thing, so you know precisely what it did.
- Masterpiece turns everything on at once, to be judged as a single object.
- Every toggle states its risk, because a change proposed without a cost is a sales pitch.
- The URL encodes your selection, so you can send back exactly what you mean.
Default output is a single self-contained HTML file. No build step, no dependencies, opens in any browser. If your project already has a preview convention (Next.js routes, Vite, Storybook) it matches yours instead. Nothing auto-applies, nothing ships: the preview is a question, and rejecting a suggestion costs one click.
Run against a real pre-launch waitlist page, the Cut proposed six moves. Two were taken and are in production today.
Move 1 of six. The contrast figures are measured rather than estimated, and the after state is live. The other four moves were declined, which is the intended shape: the tool proposes, the person decides.
It states where it is wrong. Every principle carries documented limits with named failure mechanisms, drawn from the strongest published critiques of the philosophy it teaches, including Don Norman and Bruce Tognazzini on Apple's usability regressions. A principle that has never met a serious objection produces obedience rather than judgment.
It teaches amendment, not obedience. references/amendments.md documents four cases where
Ive revised his own positions, with both sides quoted, plus three cases where a working
practitioner overruled the reference. He is a reference here, not an authority, and
questioning him is the intended use.
SKILL.md The router. Read first, under 500 lines by design
references/
principles.md 27 principles, warnings, and lenses with sources
limits.md 8 documented failure conditions, each with a mechanism
loop.md The working method, with a worked example
ive-cut.md The preview artifact: harness spec, HTML template, framework routes
gate.md The 15 pre-ship questions, with blocking rules
amendments.md When and how to overrule this skill
anti-slop.md Concrete tells of AI-default design, and what to do instead
examples/
ive-cut-example.html A real, working Ive Cut you can open and click through
ive-cut-example-moves.md Its paired move list, the other half of a LOOP delivery
fixture/index.html The sample card those moves answer
assets/
ive-cut-example.png The screenshot above
References are loaded on demand, not all at once. SKILL.md stays small so it costs almost
nothing to have installed.
Built on published primary material, including an official verbatim transcript of Ive in conversation with Patrick Collison at Stripe Sessions 2025. Where a claim could not be traced to a citable source, it was removed rather than softened. Critics are named inline, at the point where their argument is used.
Built on Jony Ive's published statements, Dieter Rams' Ten Principles, and the critical
literature that pushes back on both. Assembled and amended in practice while designing
CREAM, a portfolio tracker for everything you own. The principles here
were applied to real screens before they were written down, and three were overruled in the
process; those overrules and their reasons are in references/amendments.md.
Quotes belong to their speakers and are cited to their sources. The skill itself is MIT licensed.
MIT. See LICENSE.