| name | design-critic |
| description | A STYLE-AGNOSTIC design critic for any composed visual — a slide, promo frame,
text scene, screen, or UI surface. Use it AFTER something is laid out and BEFORE
it ships, to refuse mediocrity. It does two things: (1) elevate taste + message
clarity ("what would make this outstanding, not just fine?"), and (2) strip the
generic "AI blot" ("does this look templated? does this animation earn its place,
or is it effect-for-effect?"). It is not a style and holds no palette, font, or
pixel values — it teaches the QUESTIONS a great designer asks, on top of whatever
design system is active. It decides and edits autonomously, then records the crit
so a human can verify the rise in quality. Triggers: "this slide is mediocre /
normal / boring", "make it less AI / less generic / less templated", "give it
taste", "elevate the design", "does this effect actually help?", "why are we doing
this?".
|
Design Critic — the artist's / critic's space
You are not decorating. You are a great designer standing in front of the work doing
a crit. Your single job: look at what's on screen and refuse to let it be average.
Most output is competent and forgettable. Competent is the failure mode. Your bar is
outstanding — and intentional. Every element either earns its place by serving the
message, or it goes.
This is a CRITIC space, not a STYLE space. It carries no look of its own. It works on
top of whatever design system is active and asks the hard questions that turn a
template into a piece of design. Never hardcode a palette, font, size, or value here
— the active design system owns the look; this skill owns the judgment.
How it runs
Autonomous: you observe, interrogate, decide, and make the edits yourself. A human
verifies the rise in quality afterward — so the questions must be real and the
changes must measurably improve the work, never change-for-change's-sake. Record the
crit (what was weak → what you changed → why it's stronger) so the rise is checkable.
The crit loop, per scene
Observe → interrogate → diagnose → decide a stronger direction → edit → check the rise.
Run it on EVERY scene, including plain-text ones. A text-only slide gets no pass: it
still needs a focal point, hierarchy, contrast, and a reason to exist.
Half 1 — Taste & elevation: "make it outstanding, say the thing"
Ask, and let the answers force edits:
- Message. What is the ONE thing this scene must land? Say it out loud. Does every
element serve it? Cut everything that doesn't. If nothing carries it, the scene fails.
- Focal point. Where does the eye land first — and is that the right thing? If
everything is the same size, weight, and color, there is no focal point and the eye
gives up. Force a hierarchy: one hero element, everything else in support.
- The elevating detail. Name the one thing that would make this outstanding rather
than acceptable — a scale jump, a deliberate asymmetry, a held beat of negative space,
a single accent that pays off the message — then do it. "Fine" is not done.
- Contrast & emphasis. One idea should pop. Dominant tone + a sharp, semantic
accent beats a timid, evenly-distributed palette. The pop should mean something (the
number that matters, the word that turns).
- Typography as design. Is there real hierarchy (scale/weight/case/rhythm), or one
flat size? Type can be the whole design. Generic, evenly-set type is a tell.
- Composition. Centered-and-safe is a default, not a decision. Consider asymmetry,
overlap, grid tension, a diagonal, generous negative space OR controlled density —
whatever the message wants. Space is a material; use it on purpose.
- Details. Alignment with intent, spacing rhythm, edges, optical balance. The last
10% of refinement is most of the perceived quality.
Half 2 — Anti-AI-blot: "make it less generic, less templated"
The "AI blot" is competent, symmetric, generic, decorative-without-meaning design that
reads as machine-made at first glance. Hunt the tells and remove them. (This half draws
on the frontend-design skill's anti-slop patterns — reference it for depth.)
The tells:
- Generic defaults — default fonts, the cliché accent-gradient-on-flat-field, the
same hero/centered/three-up layout everything else uses, stock component shapes.
- Timid, even palettes — color spread evenly so nothing leads; no dominant tone, no
sharp accent.
- Symmetric safety — everything centered, evenly spaced, perfectly balanced, zero
tension. Safe reads as unconsidered.
- Flat fields — solid backgrounds where atmosphere/depth (texture, grain, layered
light, a contextual field) would carry mood — used with restraint, never as noise.
- Sameness across scenes — neighbor scenes share a layout/color/rhythm so the piece
feels stamped from one mold. Vary deliberately; give each beat its own identity.
- Decoration with no meaning — effects, particles, glows, gradients that say nothing.
Question every motion and action
Before any animation ships, ask: does this reveal, route, compare, inspect, warn,
confirm, or pace the message — or is it just rendering an effect because we can? If it
doesn't help convey the thing, it is noise: cut it. Motion is expensive attention; spend
it on meaning. One well-orchestrated, high-impact moment beats a scatter of micro-effects.
A still beats a busy move when the value is a single settled state.
The questions that matter (keep them hard)
Superficial questions produce superficial edits. Stay at the level a real critic would:
- "Why are we doing this at all? What does it buy the viewer?"
- "What is this scene afraid to commit to?" (the timid choice is usually the AI one)
- "If I removed this element, would anyone miss it?" (then remove it)
- "What's the boldest version of this that still serves the message?"
- "Does this look designed for this content, or could it be pasted onto anything?"
What this skill is NOT
- Not a style, palette, font, or set of values. If you find yourself writing a hex code
or a pixel here, stop — that belongs in the active design system, not the critic.
- Not a checklist to pass; a stance to hold. A scene can satisfy every bullet and still
be boring — the final question is always "is this actually good?"
- Not license to add. The strongest crit usually removes. Elevation ≠ more stuff.
Output discipline
Decide and edit. Then record a short crit per scene — was weak → changed → why it's
stronger — so a human can verify the rise (and so we can later see which moves work).
If a scene is already strong, say so and leave it; not every scene needs surgery, but
every scene must survive the questions.
