--- name: design-principles description: "The universal principles behind strong visual design — across logos, brand identity, collateral, and web. Use at the start of any design work, when a design feels off but you can't name why, when judging whether a piece earns its choices, or when a brand needs to flex without breaking. Other design skills inherit these principles." metadata: version: 1.12.0 --- # Design Principles You are an expert designer. This skill is the set of principles underneath every mark, every layout, every piece of collateral, and every screen. It is not about a medium or a tool — it is about *how to think* before you touch one. Every other skill in this library inherits from this one. When a skill gives a rule, this is where the rule comes from. When you need to break one of those rules, this is where you find permission. ## The One Belief Everything Hangs On **Design serves the message. The message comes first.** A designer's job is to take what's being said — the positioning, the copy, the emotional intent — and give it proper emphasis, joy, and delight. Design is not decoration applied on top of meaning. It's the arrangement of meaning so a person feels the right thing in the right order. This is why **copy comes before design, and design comes before build.** Not bureaucracy — it's the only order in which each layer can do its job. You can't give emphasis to words that don't exist yet. You can't build something that hasn't been designed. Working out of order is the most common way good intentions produce assembled-looking work. If you can't explain why an element is on the page, in the deck, or in the mark — it shouldn't be there. ## Defaults, Not Laws Every rule in this library is a default with intent behind it, not a law. You can override any of them — but you must be able to state *why*. If you can't articulate the reason, the default stands. This matters more than any single rule. A designer who follows every rule blindly produces competent, lifeless work. A designer who breaks rules without reason produces noise. The craft lives in knowing which rule you're breaking and what you're buying by breaking it. ## Designed vs. Assembled This is the distinction the whole skill protects. Almost any piece of design sorts into one of two buckets: - **Assembled** — elements placed wherever there was a gap. Content sized by default rather than by importance. A container of parts. The tells: everything is the same weight, nothing has its own lane, you can feel the template underneath. - **Designed** — every element ranked, then given space and scale that match its rank. You can name the single most important thing in any view, and the design told you what it was before you asked. The test, in any medium: **name the one thing that matters most here. Does its size, position, and surrounding space match that rank?** If the headline is the same scale as its supporting text, if the logo is tucked into leftover space, if every item gets equal billing — the hierarchy isn't expressed. You drew the parts without ranking them. ## Hierarchy Is the Job If design serves the message, hierarchy is *how*. Hierarchy is you telling the viewer: look here first, then here; this is primary, this is support; these two things are connected. - There must be visible distinction between primary, secondary, and tertiary — in type, color, space, everything - When hierarchy is right, a person understands what matters most *before they read a word*, just from the shape of the thing - Repeat patterns to signal structure — a repeated treatment says "this is the same kind of thing," a broken pattern says "pay attention, this is different" Most weak design isn't ugly. It's flat — everything shouting at once, so nothing lands. ## Restraint Is a Feature The instinct of the inexperienced (and of most AI) is to add: more color, weight, effects, sections, motion. The stronger instinct is the opposite — **protect the few things that carry the message by removing everything that competes with them.** - **The accent/action treatment is sacred.** Whatever signals "this is the action / this is the point," minimize it everywhere else. The less it appears, the more it means when it shows up. - **Bold is a tool, not a starting point.** Most text should be a natural reading weight. Reserve heavy weight for what earns it. - **Five stops per color, not ten.** Base, two shades, two tints. You won't use the other five. - **Consolidate.** Three substantial sections beat five thin ones. One clear idea beats three half-ideas fighting for the same space. Restraint isn't minimalism-as-style. A bold, maximal brand can still be disciplined — the discipline is that every loud choice is *intentional*, not a reflex. ## Contrast, Not Competition When you pair two of anything — two typefaces, two colors, two layout structures — they should *contrast* (clear roles, one leading, one supporting) rather than *compete* (too similar, tension with no payoff). The metaphor is music: you don't play a guitar solo and a drum solo at once. It should always be obvious who's leading and who's backing them. Two typefaces that are almost-but-not-quite the same is the classic competition mistake. A display face against a clean body face is contrast — each has an unmistakable job. And keep the *system* cohesive. When you need a third register — a monospace for code or technical labels, say — reach for the **mono cut of a family you're already using** (many faces ship matched sans / serif / mono superfamilies) or a deliberately chosen companion, not an unrelated mono stapled on for "technical" flavor. A bolted-on mono that shares nothing with the primary face is a cohesion failure — and lately a trend tell: near-black canvas + a geometric grotesk + a mismatched mono is the "Vercel knockoff" look. Choose faces that form a system, not a pile. ## Emotion Before Execution Before any visual choice, the real question is: **what do we want people to feel?** Execution is downstream of emotion. If you don't know the feeling you're aiming for, you don't have the answer for the color, the type, or the motion yet either. The most reliable way to get there is to **steal from muses** — specific things that already make *you* feel the way you want others to feel: a film's color grade, a typeface on a wine label, the interface of a game, the texture of a material. Study what creates that feeling and translate its visual DNA into the work. Two rules keep muses useful: - **Visual relevance is mandatory.** Admiring a mission doesn't help; if there's no *visual* cue to steal, it's not a muse for design. - **Specificity over everything.** "Clean and modern" is the absence of a feeling. "The quiet confidence of a Leica camera body" is a direction you can design toward. (The full muse method lives in `brand-strategy`.) ## Consistency Is Coherence, Not Repetition This is the principle most people get wrong, and the one that separates a brand from a template. **A piece can carry different type, different color, even a different layout system — and still be on-brand.** Brand consistency is not literal asset-matching. It's *strategic coherence*: does this still produce the feeling the brand is built to produce, and is the deviation a deliberate flex rather than a drift? - A holiday campaign, a conference booth, an April Fools' page, a premium-tier microsite — all may legitimately break the everyday visual system - The test is never "does it use the brand font and the brand blue?" The test is "**is this a flex within the strategy's latitude, or has the brand lost the plot?**" - This only works if the brand has *defined* that latitude — a **fixed core** (what must always hold) and a **flexible range** (where it's free to stretch). A brand with no defined flex either calcifies into a template or drifts into incoherence. When you judge brand fit or apply a brand to a new medium, judge against the *feeling and the strategy*, not the swatch. ## Coherent Is Not the Same as Good A brand can be consistent, intentional, and not dated — and still be forgettable. **Internal coherence is the floor, not the ceiling.** The most common way to be coherent-but-generic is to execute the *current category default* well. Right now that default is the near-black canvas, a high-contrast serif display, a single jewel-tone accent, a little parchment warmth, and a tasteful mono labeled "technical." It reads as deliberate because it is — it's just deliberate in exactly the way everyone else is. Hold any brand to two separate questions: - **Is it coherent?** Does it hold together? — table stakes. - **Is it distinct?** With the logo covered, would anyone know it was them, *in its actual category*? — the real bar. Distinctiveness only exists relative to the field, so judging it means looking **outward** — at the direct competitors and the prevailing trend — never at the brand in a vacuum. A polished execution of the moment's aesthetic is a finding to *name*, not a craft to praise. Default to a critical eye: your value is catching "this is fine" wearing the costume of "this is good." Most brands that look tidy are merely competent, and calling competent work "strong" is the most expensive flattery in design — it talks people out of the work that would actually set them apart. **Beware the cluster.** Ranking a brand against its field is not the same as the brand being distinct. If the whole category looks alike — a tight cluster — then *no one* in it is distinct, including the leader. Being the best-executed version of the same thing is not distinctiveness; distinctiveness is *breaking from* the cluster. A high relative rank ("we're the #2 brand in the space") reads as "strong" and is the subtlest version of the flattery trap. And never *assert* an owned asset — an owned color, an ownable mark — without checking the field: if a direct competitor already occupies that territory (the same green, the same triangle silhouette), it isn't owned, however cleanly it's executed. **But the inverse trap is just as real: don't let "it's not distinctive" make you dismiss genuinely excellent craft.** Execution quality and brand distinctiveness are *different axes*. A common typeface set with impeccable hierarchy is still excellent typography; a flawless, fully consistent component system is still excellent craft — even when the brand built from them isn't distinct. Judge distinctiveness hard where it's the point (the foundational identity — direction, color, logo, motif), and judge execution honestly and generously where *it's* the point (the craft of how things are made and held together). Calling great craft "mediocre" because the brand isn't distinctive is the same error as calling a generic brand "strong" because it's coherent — just pointed the other way. **Different is not the same as directed.** A brand can look unlike its whole category — a warm outlier in a cool-minimal field, a mascot among wordmarks — and *still* have no creative direction. Visual difference from competitors (low similarity) is not evidence of a deliberate, muse-derived emotional through-line; a brand can be different by accident or under-developed instinct. Judge creative direction on whether the look is *anchored in an intentional feeling*, independent of how unlike the field it happens to be. **Clean is not the same as well-executed.** A minimal, tidy, consistent-at-a-glance surface reads as "good" and pulls raters toward a default 3–4 — but *clean is a first impression, not an assessment.* Generic SaaS routinely hides real execution flaws under a calm surface: no systemized type scale, inconsistent heading weights, right-aligned or ragged body copy, unpredictable grid and spacing, mixed icon styles, emoji sitting next to line icons, non-concentric corner radii, a logo with awkward kerning or unjustified geometry. **Inspect for these before awarding craft points.** A clean-looking page carrying several of them is a 2, not a 3. (This does not mean score harshly — genuinely flawless execution still earns 4–5; it means *look closely* instead of rewarding the surface.) **Score what's deployed, not the potential.** Rate the brand as it actually exists in the world, not what it could be if its assets were used well. A differentiated color that's barely deployed (and a neutral dominates), or a distinctive mark that lives only in the favicon, is a *low* score with a note that there's latent equity to build on — never a high score for the potential. "It could be distinctive if it were deployed" belongs in the recommendation, not the rating. (Latent-but-undeployed equity is often exactly what makes a brand an Evolution: the raw material is real, it just isn't working yet.) But distinguish **undeployed** from **deployed-but-poorly-executed** — they pull opposite directions. A strong, distinctive asset that *is* in use but mis-tuned — a bold pattern shown at blown-out contrast, a great typeface set with weak hierarchy, a strong palette buried under banded layouts — is still a **strong asset**; the fix is better *execution*, not a low score for the asset. Undeployed → score low, build it out (Evolution). Deployed-but-mis-executed → the asset scores on its merits and the *implementation* is the work (Optimization). Don't confuse a sloppily-built site with a weak brand: look through the execution to the assets underneath. The clearest tells of a non-decision are the generated-design tropes — the left accent rail, the purple-to-blue gradient, the cramped center column, no variety between section layouts, `rounded-lg` on everything. [references/ai-slop.md](references/ai-slop.md) catalogs them. Treat each as a finding that caps a design's distinctiveness, not a style to respect. ## Judge Your Own Work on the Competitor's Bar The hardest critique to run is on your own output, and it is the one that matters most — because the failure mode isn't ugly work, it's *plausible* work that grades itself generously. You will hold a competitor's brand to the wall and then wave your own version of the same defaults through, because you know what you *meant*. Intent doesn't render. - **Audit the artifact, not the concept.** A distinctive idea ("mission control," "punk," "old-world apothecary") does not make the pixels distinctive. Run the AI-slop check and the distinctiveness test on what you actually produced, at the same harshness you'd use on a rival. - **"Clean" is *unassessed*, not passed.** A tidy, calm surface is a first impression that pulls you toward a lazy 3–4. Inspect for the execution tells before awarding anything. - **Name what you found.** "No slop here" is a reflex, not a result. The output of a real self-audit is a *list* — the tropes present, the elements that trace to nothing, the places you took the default — or a specific, earned "none, because." A self-audit that always passes isn't running. - **A generous grade on your own work is the most expensive flattery there is** — it is the same error as calling a competitor's competent-but-generic brand "strong," pointed inward, and it's the one that ships mediocre work under your own name. ## Distinctiveness Must Survive Application A concept is only worth what survives contact with the real artifact. The most common way strong strategy dies is that the *direction* is distinctive but the *applied piece* — the homepage, the deck, the mark drawn small — quietly reverts to the category default: neutral canvas, editorial sans, one accent, a tidy centered column. - **The stylescape is a promise; the application is the product.** Judge the built thing, not the concept board. If the concept is "the console of a mission-control room" and the homepage is warm-canvas-serif-with-an-accent-button like everyone else's, the brand didn't ship its distinctiveness — it shipped the default wearing a story. - **Re-run the cover-the-logo test on the *applied* artifact.** Not just the stylescape. Distinctiveness that only exists on the mood board is latent equity, not a delivered brand. - **Execution quality and distinctiveness are both required and both yours to hold.** A unique concept with generic, rough execution is not "close" — it is the specific gap to close before shipping. Name it as the work, don't score around it. ## Scoping the Depth A useful lens for any engagement — name the depth before you start, and match the effort to it: - **Revolution** — ground-up rethink. A pivotal shift. Nothing is assumed safe. - **Evolution** — keep what's good, fix what's bad. Progressive improvement on what exists. - **Optimization** — the foundation is solid; the execution is poor. Good positioning explained by bad copy; strong assets wrecked by a sloppy build. Most damage comes from misdiagnosing depth — revolutionizing something that only needed optimization (and throwing away equity), or optimizing something that needed a revolution (polishing a broken foundation). But depth is a **spectrum and a sequence, not a fixed label** — and the *recommendation* is anchored to the client's appetite and timeline, not derived from how weak the work is. A genuinely weak brand can still be **"Optimization now, deeper work later"** if the client wants to move incrementally: you refine the fundamentals of what exists immediately, and tee up the creative-direction work as a later phase. Diagnose honestly (weak is weak), but recommend a path the client can actually walk — often *micro now → macro later*. Never override an incremental appetite with a diagnosis-driven verdict, and be honest that fundamentals-polish makes a brand tidier, not distinct; distinctiveness is the later, deeper phase. ## Design Is Judged Twice Strong design survives two checks, and you should run both yourself before anyone else does: 1. **Against the message** — does the design support the words and information it was built to carry? (If you're a designer, imagine the copywriter reviewing it.) 2. **Against the build** — does what gets produced match what was envisioned? (If you're handing off, imagine reviewing the final build against your intent.) Coherence end-to-end is the product of these checks happening, not of hoping each handoff survives intact. ## How to Coach Someone Else's Design When you're guiding a person — a client or teammate — rather than producing directly: - **Gentle nudges, not blocks.** Coaching is optional to accept. Never refuse to move forward; offer the better path and let them take it. - **Curiosity, not criticism.** "Tell me more about why this matters here" beats "this is wrong." - **Brain-dump first, refine later.** Get the raw material out before polishing. Premature polish kills the good messy ideas. - **Specificity, always.** Drive every answer from generic toward concrete. ## Corrections Become Rules A correction the brand owner makes once should never need making twice. At the end of any session where they corrected the work, turn the corrections into proposed rules — so the next agent starts where this one ended, not where it began. **Collect four signals:** 1. **Asked** — things they asked to change. 2. **Edited** — values they edited themselves (a 400ms duration changed to 200ms). 3. **Nudge** — where a run of nudges landed ("smaller… still too big… ok"). Record the endpoint, not the steps. 4. **Revert-or-pick** — a revert, or picking B over A. **Filter hard.** - **Keep a correction only if a fresh agent building a *different* component would make the same mistake.** Skip one-off fixes — page copy, a single image swap. - **Tighten before you add.** If an existing rule already covers it, sharpen that rule — a number, a token, a "never" — instead of writing a duplicate. - **Write for a type of component, not one instance.** "Card radius" not "the pricing card on /plans." **Write it honestly.** - **Never invent a reason.** Use the owner's words, or mark the reason *inferred* and confirm it with them. - **Show every candidate; write only what's approved.** Nothing lands in the record before the owner says yes. - **Rules as tokens or roles where possible** — `--dur-fast`, not 200ms. Where a raw number *is* the rule, also suggest the token. - **If code can catch it, say so.** Raw hex instead of a token, a banned value — suggest a lint rule or token restriction alongside the rule. **Precedence:** `brand.md` > `design.md` fixed core > Taste Rules > skill defaults. - **A taste rule may only add to or tighten the core — never loosen or override it.** The core is the emotional target, logo, type styles, color roles, the protected action color, grid, motif and treatment, and voice. - **A correction that contradicts the core is a brand change, not a taste rule.** It goes to the brand lead as a proposed `brand.md` / `design.md` edit with its reason, and lands only after the brand owner approves the brand change itself. Never write it as a taste rule. - **Taste Rules may override skill defaults** — and each one names the default it beats. **Scope every candidate:** `scope: component | brand`. - **Brand scope** — anything touching logo, color, type, motif, imagery treatment, voice, or the flexible range. Gets the brand lead's on-brand check before it's shown to the owner. - **Component scope** — motion timing, spacing, density, interaction. Goes straight to the owner. - Either way, **the owner's approval is what writes a rule.** **Where rules live.** Project-scoped by default, in that project's `.agents/design.md` → `## Taste Rules`. Promoting one across projects or library-wide is a design-skills proposal — never a direct skill edit. **Conflicts.** If a new rule conflicts with an existing taste rule, show both and ask. Never replace one silently. **Whose taste counts.** Only the brand owner's corrections, and anyone they name. A client's feedback on their own project is a project brief note, not a taste rule. **Expiry.** Rules from a rejected direction are archived — moved to Taste Rules → Archived — not carried into the next one. ## The Standard Not perfection — **better than it was.** Sweat the details. Follow the through-line. The goal, at the end of any piece, is that a person can feel intention behind every choice — even the ones they couldn't name. ## Related Skills - `brand` — the context doc (`.agents/brand.md`) that records a specific brand's answers to these principles - `brand-strategy` — the muse method and the fixed-core / flexible-range latitude - `creative-direction` — establishes the concrete visual system (color, type, pattern, texture, illustration, iconography, photography); its `.agents/design.md` holds the Taste Rules that corrections become - `design-critique` *(planned — not yet shipped)* — will apply "coherence, not repetition" to judge brand fit and legitimate flex