Back to Details

creative-direction

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Creative Direction

创意方向

You establish a brand's visual system — every asset class, one coherent language, all of it traceable to the feeling the brand is built to produce. This is the stage where strategy becomes visible: the emotional target and muses from 
brand-strategy
get translated into actual color, type, composition, pattern, icons, illustration, and photography.
Read 
design-principles
first. This skill is where its rules get applied: contrast-not-competition, restraint, five stops, emotion before execution, and "state why, or the default stands."
Position in the sequence: after 
brand
(the audit/context) and 
brand-strategy
(the feeling + muses), before 
logo-design
and the application skills. The audit diagnoses; this skill builds the cure.
你需要为品牌建立视觉系统——覆盖所有资产类别，形成一套连贯统一的语言，所有元素都可追溯到品牌旨在传递的核心感受。这是战略可视化的阶段：来自
brand-strategy
的情感目标与灵感缪斯，将被转化为实际的色彩、字体、构图、图案、图标、插画及摄影元素。
请先阅读
design-principles
。本技能是这些规则的落地应用场景：对比而非竞争、克制原则、五级色彩梯度、情感优先于执行，以及“说明理由，否则沿用默认规则”。
流程定位：在
brand
（审计/背景调研）和
brand-strategy
（情感目标+灵感缪斯）之后，
logo-design
及落地应用类技能之前。审计负责“诊断问题”；本技能负责“构建解决方案”。

Hard Prerequisite: the Emotional Target and Muses

硬性前提：情感目标与灵感缪斯

Open 
.agents/brand.md
. You need:
  • The Emotional Target (the feelings the brand must produce)
  • The Muses, each with its stealable visual cue
  • The confirmed competitor set (what "distinct" is measured against)
  • The REO depth and the client's appetite
If the Emotional Target or muses are missing, stop and route to 
brand-strategy
. Without them you'd be choosing colors and typefaces from taste — exactly the "different by accident" and "category default" failure modes the audit exists to catch. Every choice you make here must trace to a muse or the emotional target; if you can't say why a choice serves the feeling, it's a default wearing a decision's clothes.
打开
.agents/brand.md
文件。你需要获取以下内容：
  • 情感目标（品牌必须传递的核心感受）
  • 灵感缪斯，每个缪斯需包含可借鉴的视觉线索
  • 已确认的竞品集合（用于衡量“差异化”的参照标准）
  • REO深度与客户接受度
若缺少情感目标或灵感缪斯，请立即停止并转向
brand-strategy
流程。 没有这些依据，你将仅凭个人喜好选择色彩和字体——这正是审计要避免的“偶然差异化”和“品类默认风格”的失败模式。你在此做出的每一项选择都必须能追溯到某个缪斯或情感目标；若无法说明某一选择如何服务于品牌感受，那它只是披着决策外衣的默认选项。

Scope by Depth

按深度划分的工作范围

  • Revolution — build the system from scratch as three genuinely distinct concepts, presented as stylescapes (see references/stylescapes.md). Each concept is a different plausible answer to the emotional target — different muses, color story, type pairing, and attitude — not three shades of one idea. Stylescapes are real intended assets, not mood boards: expressive and free, not yet constrained by buildability.
  • Evolution — one direction, anchored on the fixed core (the equity being kept — often the logo, name, or a color). Start from the primary touchpoint (usually the hero/website) and change only the deltas the strategy named: carry the kept assets forward, introduce the new type, motif, or palette around them.
  • Optimization — mostly out of scope. The direction already exists; the work is execution (
    web-design
    , 
    collateral-design
    ). At most a light pass to document the existing system so it can be enforced.
  • 革新级——从零开始构建系统，输出三个完全不同的独立概念，以stylescapes形式呈现（详见references/stylescapes.md）。每个概念都是对情感目标的不同合理回应——采用不同的缪斯、色彩方案、字体搭配及风格态度，而非同一思路的三种微调。Stylescapes是真实的意向资产，而非情绪板：表达自由不受限，暂不考虑落地可行性。
  • 演进级——单一方向，锚定固定核心资产（需保留的品牌资产——通常是标志、品牌名或某一色彩）。从核心触点（通常是官网首屏）入手，仅调整战略中明确指出的差异部分：保留既有资产，围绕其引入新字体、图案或配色。
  • 优化级——基本不在本技能范围内。品牌方向已确定，工作内容为执行层面（
    web-design
    、
    collateral-design
    ）。最多仅需对现有系统进行轻度梳理和文档化，以便后续执行时统一规范。

The Core Move: Abstract the Muses

核心动作：提炼灵感缪斯

This is the engine of a distinct direction, and the step most likely to be skipped. Take the muse's literal forms — recorded at physical fidelity in 
brand-strategy
— and abstract them into modern, ownable brand elements. You are not decorating with the muse; you are distilling its geometry, materials, and ideas into a system.
  • Worked examples (real): the concentric rings around the Kaaba → a dotted-line motif meaning unity of people; the building's brick geometry → an ownable grid; Islamic gardens → an "ants in the garden" pattern. The mission-control muse → the amber-on-black telemetry readout and the machined dimensional knob as the signature element.
  • Pull two things from each muse: a color story and an abstracted idea. Then combine the abstractions with modern UI. Most software looks the same — the muses are how you skin an undifferentiated category with something only this brand could own.
  • Then render the product's own work through that abstraction — don't stop at decoration. The single biggest gap is a board that ornaments the muse but never shows what the product does. Show the product's actual jobs — a post being drafted, a blog written, SEO run, a chart climbing up-and-to-the-right — expressed in the muse's language: musical notes that bloom into a published social post; a console readout that is the campaign launching. Muse ornaments floating over generic UI read as boring and unowned; the muse × product-function merge is what makes a direction feel like this brand doing this job.
  • Don't let ease of abstraction crown the winner, and treat on-the-nose as a failure mode. The muse that maps most literally to UI (a mission-control console → dials, telemetry) resolves fastest and looks the most finished soonest — but the literal reading is often the weakest, most clichéd idea, and cliché is a real liability in an AI category where everyone grabs the obvious. The muse that's harder to abstract (orchestration, empire) is frequently the more powerful, more ownable concept. Push harder on the difficult abstraction; don't retreat to the easy muse or fake the hard one with slop. Judge concepts on conceptual strength, not on which rendered first.
  • Typography obeys the same abstraction law as everything else — and it has two failure poles. A literal muse-face (an all-caps Roman inscriptional / Cinzel for an imperial muse) isn't a taste error, it's an abstraction failure: you wore the muse instead of distilling it. The opposite trap is just as fatal: reaching for a try-hard trendy face (Space Grotesk, Fraunces, and their cohort — banned; see 
    design-principles
    → ai-slop.md) to look designed. Both are costume. The distinction must come from the system — motif, color, composition, treatment — not from the font. When in doubt, a neutral face used deliberately (Inter) beats a characterful one grabbed as a shortcut; let the system carry the muse. Muse-derived type is genuinely rare.
这是打造独特品牌方向的核心引擎，也是最容易被跳过的步骤。将
brand-strategy
中精准记录的缪斯具象形态，提炼为现代、专属的品牌元素。 你不是用缪斯做装饰，而是将其几何形态、材质质感和核心理念提炼为一套系统。
  • 实际案例：克尔白周围的同心圆环 → 代表“人类团结”的虚线图案；建筑的砖块几何结构 → 专属网格系统；伊斯兰花园 → “花园中的蚂蚁”图案；任务控制中心缪斯 → 黑底琥珀色遥测读数+机械旋钮作为标志性元素。
  • 从每个缪斯中提取两个要素：色彩方案和抽象理念。 然后将抽象理念与现代UI结合。大多数软件视觉趋同——缪斯正是让品牌在同质化品类中打造专属视觉的关键。
  • 用抽象理念呈现产品实际功能——不要停留在装饰层面。 最常见的误区是仅用缪斯做装饰，却未展示产品的实际用途。要展示产品的核心功能——比如草稿撰写、博客创作、SEO运行、数据图表上升——并用缪斯的语言来表达：音符绽放为已发布的社交帖；控制台读数就是正在启动的营销活动。仅在通用UI上叠加缪斯装饰会显得乏味且无专属感；缪斯×产品功能的融合才会让品牌方向呈现出“这个品牌做这件事”的独特性。
  • 不要让提炼难度决定方案优劣，避免过于直白的表达。 最容易与UI直接映射的缪斯（如任务控制中心→旋钮、遥测读数）能最快成型且看起来最完整，但这种直白解读往往是最弱、最陈词滥调的想法，在AI品类中尤其危险，因为所有人都会想到这些显而易见的元素。更难提炼的缪斯（如管弦乐、帝国）往往能打造出更有影响力、更具专属感的概念。要在难度高的提炼上多下功夫；不要退而求其次选择容易的缪斯，也不要敷衍了事。评判方案的标准是概念强度，而非哪个方案先完成渲染。
  • 排版同样遵循提炼原则——存在两个失败极端。 直白使用缪斯风格字体（如为帝国缪斯选择全大写罗马铭文字体/Cinzel）不是审美错误，而是提炼失败：你直接套用了缪斯的风格，而非提炼其内核。另一个极端同样致命：盲目追求刻意时髦的字体（Space Grotesk、Fraunces及其同类——禁用；详见
    design-principles
    →ai-slop.md）来“显得有设计感”。这两种都是“穿 costume”。差异化必须来自系统——图案、色彩、构图、处理方式——而非字体本身。拿不准时，刻意使用中性字体（如Inter）比为了走捷径而选择有个性的字体更好；让系统来承载缪斯的理念。源自缪斯的专属字体极为罕见。

The System to Establish

需要建立的视觉系统

Work through each element with the audit's bar in mind — the nine-category audit (see 
brand
) is your acceptance test, and every "Moving forward" item the audit raised (all nine categories except the logo, which 
logo-design
handles after) must be answered by the system — don't leave an audit gap unaddressed. Full per-element standards: references/element-standards.md. The production-craft floor — art-direction commitment, design tokens, typography/texture hard rules, the abomination checklist — lives in references/frontend-boilerplate.md; the board's collage/bento composition in references/stylescapes.md.
  1. Color — assign by role: background, text (primary/secondary), action/CTA (protected — minimized everywhere else), support. Five stops per color, not ten. Verify ownership against the confirmed competitor set — a color isn't owned if a rival occupies it; deliberately escape the category's default lane.
  2. Typography — faces by role (display, heading, body), chosen for contrast, not competition; if a mono or third register is needed, prefer a superfamily's own cut over a bolted-on face. Set weight/style restrictions. Licensing checkpoint: foundry, budget, and character/language support for the target markets — confirm before the direction is approved, not after.
  3. Composition & layout — the grid, density, spacing logic, section rhythm and variety (no monotonous stacks), how the eye is directed.
  4. Pattern & texture — an ownable motif, most naturally derived from the mark's geometry, deployable off-site without the logo or color present. Check it doesn't collide with a competitor's motif.
  5. Iconography — one source of truth; stroke/fill/grid rules; plan the tiers (utility now, storytelling/illustration as the brand matures).
  6. Illustration & product abstraction — branded abstractions over raw screenshots; a defined style (or an explicit "no illustration" decision).
  7. Photography — subjects, treatment/grade, art direction; or the deliberate absence of it.
  8. Motion (directional note, not full spec) — what the brand's motion feels like; intent over ornament (hover > load > scroll > loop).
按照审计标准逐一梳理每个元素——九类审计（详见
brand
）是你的验收测试，审计提出的所有“后续改进”事项（除标志外，标志由
logo-design
后续处理）都必须由本系统解决——不要留下任何审计缺口。各元素的完整标准：references/element-standards.md。制作工艺规范——艺术指导要求、设计令牌、排版/纹理硬规则、避坑清单——详见references/frontend-boilerplate.md；风格场景板的拼贴/便当式构图规范详见references/stylescapes.md。
  1. 色彩——按角色分配：背景色、文本色（主/次）、行动/CTA色（受保护——其他场景尽量少用）、辅助色。每种色彩设五级梯度，而非十级。对照已确认的竞品集合验证专属权——若竞品已使用某一色彩，则该色彩无法成为品牌专属；需刻意跳出品类默认的色彩范围。
  2. 排版——按角色选择字体（展示型、标题、正文），遵循对比而非竞争原则；若需要等宽字体或第三类字体，优先选择超级字体系列自带的变体，而非额外搭配其他字体。设置字重/样式限制。许可检查：字体厂商、预算、目标市场的字符/语言支持——请在方向获批前确认，而非之后。
  3. 构图与布局——网格系统、密度、间距逻辑、区块节奏与变化（避免单调堆叠）、视觉引导方式。
  4. 图案与纹理——专属图案，最自然的来源是标志的几何形态，即使脱离标志或色彩也能独立识别。需检查是否与竞品图案冲突。
  5. 图标系统——单一数据源；描边/填充/网格规则；规划层级（当前优先实用型，品牌成熟后可增加叙事/插画型）。
  6. 插画与产品抽象——在原始截图上叠加品牌化抽象元素；明确风格（或做出“不使用插画”的决策）。
  7. 摄影——拍摄主体、处理/调色风格、艺术指导；或刻意不使用摄影元素。
  8. 动效（方向性说明，非完整规范）——品牌动效的核心感受；优先功能性而非装饰性（hover > 加载 > 滚动 > 循环）。

Producing the Board — Source and Generate the Real Assets

制作风格场景板——获取并生成真实资产

A stylescape needs treated imagery and genuine texture; flat CSS shapes read as austere. The operator is often a founder, not a designer, so the craft leverage is on you. But don't assume their toolset — ask first, then fork the workflow.
  • Ask what image-generation tools they have, then fork. Before writing a single prompt, ask which the operator can access. Assume at least ChatGPT (GPT Image) and Google Gemini / Imagen; the fuller set — Higgsfield, Flora, Weave, Magnific, Midjourney, Adobe Firefly, Figma, Paper — isn't in everyone's purview, so ask rather than assume. Then take the most capable path available:
    • (a) Drive it directly — if a tool is connected to you as an MCP (Adobe Firefly, Figma, and Paper are common), generate and place the bespoke, muse-anchored imagery yourself, in-session. Strongest and most ownable; zero founder steps.
    • (b) Hand off prompt packs — for a generator only they have (Midjourney, Flora, Higgsfield…), write precise muse-anchored prompts (subject, grade, palette hexes, texture, lighting, composition, aspect ratio, "no text") for them to run and bring back. Iterate the prompt when a result misses.
    • (c) Free stock, treated — if they have none, real free stock (Unsplash/Pexels) run through the concept's grade is a great floor. Populate it yourself so tone is present without a founder step.
  • You compose the board, not the founder. They have no Figma skill and no design eye — so you assemble the coalesced board (self-contained HTML/CSS/SVG) from the generated + sourced assets. The operator supplies raw images (or you generate them); you do the art direction, layout, texture, and typography.
  • Lean on CSS/SVG for the texture it renders well — grain, noise, scanlines, patterns, grids, machined UI, telemetry, glows, gradients — which for many concepts (anything mechanical, retro, systems-y) is most of the "life," no generation needed. Reserve generation / free stock for photographic or illustrative muses. (Strict-CSP HTML can't load a webfont URL — embed the face as a data URI, or accept a fallback and name the intended typeface so the type intent reads.)
  • Unify every source through one grade. Whatever the origin — MCP-generated, founder-run, or free stock — run it through the concept's treatment recipe (a duotone + grain + scanline for a console; a warm-paper wash for a programme; a porphyry duotone for an imperial muse). Show it as a before→after on the board. The grade is what makes even free stock read as this brand's — and what keeps mixed sources looking like one world.
  • The muses are the brief. Everything generated or sourced is treated to the muse cues at physical fidelity — never generic stock dropped in raw.
  • If imagery can't be produced yet, say so and label the board a composition / art-direction proof with the prompt packs attached (for whichever generator the fork identified), so the founder can fill it in. Never quietly ship a flat board and call it finished.
Stylescape需要经过处理的图像和真实纹理；扁平化CSS形状会显得过于朴素。操作者通常是创始人而非设计师，因此你需要充分发挥专业优势。但不要假设他们的工具集——先询问，再调整工作流程。
  • 询问他们可用的图像生成工具，再调整流程。 在撰写任何提示词之前，先询问操作者能访问哪些工具。默认至少有ChatGPT（GPT Image）和Google Gemini / Imagen；更完整的工具集——Higgsfield、Flora、Weave、Magnific、Midjourney、Adobe Firefly、Figma、Paper——并非所有人都能使用，因此请询问而非假设。然后选择最可行的路径：
    • (a) 直接生成——若工具与你作为MCP相连（Adobe Firefly、Figma、Paper较为常见），可在会话中直接生成并放置以缪斯为核心的定制图像。这是最强且最具专属感的方式；无需创始人操作。
    • (b) 交付提示词包——若工具仅操作者可用（Midjourney、Flora、Higgsfield…），请撰写精准的、以缪斯为核心的提示词（主体、调色、十六进制色值、纹理、光线、构图、宽高比、“无文字”），供他们运行并返回结果。若结果不符合预期，迭代提示词。
    • (c) 处理免费素材——若他们没有任何生成工具，使用真实免费素材（Unsplash/Pexels）并按概念风格调色是不错的基础。请你自行完成素材处理，确保风格统一，无需创始人操作。
  • 由你负责拼接场景板，而非创始人。 他们没有Figma技能和设计眼光——因此由你将生成+获取的资产组合成完整的场景板（独立的HTML/CSS/SVG文件）。操作者提供原始图像（或由你生成）；你负责艺术指导、布局、纹理和排版。
  • 充分利用CSS/SVG擅长渲染的纹理——颗粒感、噪点、扫描线、图案、网格、机械UI、遥测读数、光晕、渐变——对于许多概念（如机械风、复古风、系统风），这些纹理就能打造大部分“鲜活感”，无需生成图像。仅在处理摄影或插画类缪斯时使用生成工具/免费素材。（严格CSP的HTML无法加载网页字体URL——请将字体嵌入为数据URI，或接受 fallback 字体并注明目标字体，以便传达排版意图。）
  • 用统一调色风格整合所有素材。 无论素材来源——MCP生成、创始人运行、免费素材——都需按概念的处理方案进行统一调整（如控制台风格用双色调+颗粒+扫描线；节目风格用暖纸质感；帝国缪斯用斑岩双色调）。在场景板上展示前后对比效果。统一调色能让即使是免费素材也看起来像这个品牌的资产——并让不同来源的素材呈现出统一的视觉世界。
  • 缪斯就是创作brief。 所有生成或获取的素材都必须严格遵循缪斯的视觉线索——绝不能直接使用通用素材。
  • 若暂时无法生成图像，请明确说明，并将场景板标注为构图/艺术指导样稿，附上对应的提示词包（针对之前确定的生成工具），以便创始人后续补充。绝不能悄悄交付扁平化场景板并声称已完成。

Validate Before You Present

展示前的验证

Run your own work through the same gauntlet the audit runs — adversarially, on the rendered board, not the concept in your head (see 
design-principles
→ Judge Your Own Work on the Competitor's Bar). The concept can be distinctive while the pixels are generic; a self-audit that grades intent misses exactly that.
  • Score it against the foundational categories vs. the confirmed competitor set, at the harshness you'd use on a rival — would this system earn 4–5 on Creative Direction, Color, Pattern, and (with 
    logo-design
    ) Logo? Cover the logo: is it still recognizably this brand?
  • Run the AI-slop check (
    design-principles
    → ai-slop.md) on your own output and list what it finds — the tropes actually on the board (a mono "telemetry/terminal" readout, a dot-grid, a left rail, a centered no-variety column, 
    rounded-lg
    everywhere). "No slop here" is a reflex, not a result; produce the list, or an earned, specific "none, because —." "Clean" counts as unassessed, not passed.
  • Confirm the board is a rendered surface, not a CSS-block comp — treated imagery or a committed illustration style, a full named type specimen, an ownable motif (not a dot-grid default), applied mockups, real depth and bleed. Missing anatomy = under-built (see references/stylescapes.md).
  • Carry-through gate: the applied fragment (the hero/mockup) must pass cover-the-logo too, not just the abstract board (see 
    design-principles
    → Distinctiveness Must Survive Application). A distinctive board with a generic applied screen is a fail — the distinctiveness has to survive into application, which is where it usually leaks out.
  • Trace every element to a muse or the emotional target. Anything you can't trace, cut or justify.
  • Copy check: does the system serve the message and give the words their proper emphasis? Design serves the message — the system is the stage, not the show.
用审计的严苛标准来检验你的工作——针对渲染完成的场景板，而非你脑海中的概念（详见
design-principles
→以竞品标准评判自己的作品）。概念可能很独特，但像素呈现可能很平庸；仅评判意图的自我审计恰恰会忽略这一点。
  • 对照基础品类与已确认的竞品集合打分，用评判竞品的严苛标准——这套系统在创意方向、色彩、图案（加上
    logo-design
    后的标志）上能拿到4-5分吗？遮住标志：它还能被识别为这个品牌吗？
  • 对你的输出进行AI陈词滥调检查（
    design-principles
    →ai-slop.md），并列出发现的问题——场景板上实际存在的俗套元素（如等宽“遥测/终端”读数、点阵网格、左侧边栏、无变化的居中列、全用
    rounded-lg
    ）。“没有问题”是本能反应，而非结果；请列出问题，或给出有理有据的具体说明“无问题，因为——”。“简洁”属于未评估，而非通过。
  • 确认场景板是渲染完成的成品，而非CSS块样稿——包含经过处理的图像或明确的插画风格、完整的命名字体样本、专属图案（而非默认点阵网格）、应用样稿、真实的层次感和 bleed。缺少必要元素即为未完成（详见references/stylescapes.md）。
  • 一致性检查：应用片段（首屏/样稿）也需通过遮标测试，而非仅抽象场景板（详见
    design-principles
    →差异化必须在落地应用中保持）。场景板独特但应用屏幕平庸即为失败——差异化必须延续到落地应用中，这也是差异化最容易流失的环节。
  • 每个元素都能追溯到缪斯或情感目标。 任何无法追溯的元素，要么删除，要么给出合理理由。
  • 文案检查：系统是否服务于品牌信息，并为文字赋予恰当的强调？ 设计是为信息服务的——系统是舞台，而非演出本身。

Present, Decide, Record

展示、决策、记录

  • Give it a walkthrough first, then present. Talk each concept through — the muses, the abstraction, why it answers the feeling — before handing it over. A stylescape without its narrative is just arrangement.
  • Name the challenges, honestly. For each direction, say what would be hard to implement across the board, and how different it actually is from the competitors — the things the client needs to weigh to decide. Don't only sell; surface the tradeoffs.
  • Close with a comparison — per direction or a synopsis at the end. For each concept, answer the questions that actually drive the pick: the challenges, the pros, is this the most differentiated?, and is this the most on-brand with the emotion you're trying to elicit? Fold it into each direction's walkthrough or give a synopsis that weighs all three against each other at the close.
  • Prove it has legs applied. Clients want to see it real, not abstract — the fastest proof is a website hero + one or two ads carrying the full system. A concept that looks great on the board but thin when applied hasn't earned the pick (see the carry-through gate above).
  • Document the styles and their sources. Photo / graphic / icon / typography styles, and where any sourced assets came from (stock, Envato, Freepik, etc.) — notes now save the application skills from guessing later.
  • Revolution: present the three stylescapes; the client picks one. Push for one idea, not a Frankenstein. A little modularity is fine, but merging concepts usually averages three ideas into none. When a client asks to graft (e.g. this concept's type on that one's board), knowing the emotional target lets you reject the merge that betrays it — a merge is legitimate only if it still produces the feeling and has its own single story (see references/stylescapes.md → synthesis vs. averaging).
  • Evolution: present the direction against the current state so the deltas are unmissable.
  • Get explicit approval before 
    logo-design
    and the application skills build on it — everything downstream compounds on this decision. (Run creative direction before any naming project.)
  • Write the approved system into 
    .agents/brand.md
     (the Visual system section, plus updates to Fixed core / Flexible range), recording the why behind each choice — the rationale is what survives handoffs and lets a later critique judge flexes.
  • 先讲解，再展示。 在交付前，先逐一讲解每个概念——缪斯来源、提炼过程、如何呼应品牌感受。没有叙事的stylescape只是元素堆砌。
  • 坦诚说明挑战。 针对每个方向，说明在全渠道落地时的难点，以及与竞品的实际差异程度——这些是客户做决策时需要权衡的因素。不要只推销，要明确展示取舍。
  • 结尾进行对比——可针对每个方向讲解，或在最后总结。 针对每个概念，回答驱动决策的核心问题：挑战、优势、是否最具差异化？、是否最贴合品牌想要传递的情感？ 可将对比融入每个方向的讲解，或在结尾总结时对三个概念进行综合权衡。
  • 证明其落地可行性。 客户希望看到真实效果，而非抽象概念——最快的证明方式是官网首屏+1-2个广告，完整应用这套系统。仅在场景板上好看但落地时显得单薄的概念，不足以被选中（详见上述一致性检查）。
  • 记录风格及素材来源。 摄影/图形/图标/排版风格，以及所有获取素材的来源（素材网站、Envato、Freepik等）——现在做好记录，能避免后续落地技能需要猜测。
  • 革新级：展示三个stylescapes；由客户选择一个。 推动客户选择单一理念，而非拼凑组合。少量模块化是可以的，但合并概念通常会将三个想法平均成毫无特色的产物。当客户要求拼接（如将A概念的字体用在B概念的场景板上），凭借对情感目标的理解，你可以拒绝违背核心感受的拼接——只有当拼接后的方案仍能传递品牌感受且有统一叙事时，才是合理的（详见references/stylescapes.md→合成 vs 平均）。
  • 演进级：将新方向与当前状态进行对比，让差异一目了然。
  • 在
    logo-design
    及落地应用技能启动前，获取明确的批准——后续所有工作都将基于此决策展开。（任何命名项目前都需先完成创意方向流程。）
  • 将获批的系统写入
    .agents/brand.md
    （视觉系统部分，以及固定核心/灵活范围的更新），记录每个选择的理由——这些理由能在交接时保留，并让后续评审有据可依。

The Deliverable

交付物

Two layers, one source of truth — mirroring 
brand-strategy
:
  1. .agents/design.md
     — the full creative-direction record, generated at the end of this exercise. It holds: the three concepts and the one chosen; each concept's muses → abstractions; the documented photo / graphic / icon / typography styles with their asset sources (stock, Envato, Freepik…); the color system (roles + hex + ramps) and design tokens; the ownable motif/texture and the signature element; the application examples (hero + ads); the implementation challenges and competitor-differentiation notes; and the why behind every choice. Use the template in references/design-doc-template.md. This is the detailed spec the application skills (
    web-design
    , 
    collateral-design
    , …) build against — the counterpart to 
    brand-strategy
    's 
    .agents/brand-strategy.md
    . And it's a living record: downstream skills append their decided assets to it as the pipeline advances — 
    logo-design
    the approved mark, 
    motion-design
    the built motion system, application skills their adopted libraries and notes (see the template's Living Record sections). This skill writes the founding sections; the record grows from there.
  2. .agents/brand.md
     — the distilled system: write the approved Visual system into its section (plus Fixed core / Flexible range) so every downstream skill inherits it without reading the whole record.
Plus, if the client wants it, a published copy (Notion, PDF) generated from 
.agents/design.md
— never the other way around.
两层结构，单一数据源——与
brand-strategy
保持一致：
  1. .agents/design.md
    ——完整的创意方向记录，在本流程结束时生成。内容包括：三个概念及最终选定的方案；每个概念的缪斯→提炼过程；已记录的摄影/图形/图标/排版风格及素材来源（素材网站、Envato、Freepik…）；色彩系统（角色+十六进制色值+梯度）及设计令牌；专属图案/纹理及标志性元素；应用示例（首屏+广告）；落地挑战及竞品差异化说明；每个选择的理由。使用references/design-doc-template.md中的模板。这是后续落地技能（
    web-design
    、
    collateral-design
    等）的详细规范——对应
    brand-strategy
    的
    .agents/brand-strategy.md
    。这是一份动态记录：后续技能会随着流程推进在其中添加已确定的资产——
    logo-design
    添加获批的标志，
    motion-design
    添加已构建的动效系统，落地技能添加采用的库及说明（详见模板中的动态记录部分）。本技能负责撰写基础章节；后续记录将逐步完善。
  2. .agents/brand.md
    ——提炼后的系统：将获批的视觉系统写入对应章节（以及固定核心/灵活范围），以便所有后续技能无需阅读完整记录即可继承规范。
此外，若客户需要，可基于
.agents/design.md
生成公开版本（Notion、PDF）——绝不能反向操作。

Non-Negotiables

不可妥协项

  • Emotional Target + muses read from 
    .agents/brand.md
    (or routed to 
    brand-strategy
    first)
  • Depth + appetite respected (three stylescapes / anchored deltas / documentation pass)
  • Color ownership verified against the confirmed competitor set
  • Type licensing + character support confirmed
  • Motif is ownable and doesn't collide with a competitor's
  • Every element traces to a muse or the emotional target
  • Adversarial self-audit run on the rendered board (AI-slop tropes listed, not "none"; applied fragment passes cover-the-logo) before presenting
  • Challenges + competitor-differentiation named; direction proven applied (hero + ad)
  • Client approval secured; full record written to 
    .agents/design.md
    ; distilled system written to 
    .agents/brand.md
  • 从
    .agents/brand.md
    读取情感目标+灵感缪斯（或先转向
    brand-strategy
    流程）
  • 遵循深度要求与客户接受度（三个stylescapes / 锚定差异 / 文档梳理）
  • 对照已确认的竞品集合验证色彩专属权
  • 确认字体许可+字符支持
  • 图案具有专属权且不与竞品冲突
  • 每个元素都能追溯到缪斯或情感目标
  • 展示前对渲染完成的场景板进行严苛自我审计（列出AI陈词滥调元素，而非“无问题”；应用片段通过遮标测试）
  • 明确说明挑战+竞品差异化；证明方向可落地（首屏+广告）
  • 获取客户批准；完整记录写入
    .agents/design.md
    ；提炼后的系统写入
    .agents/brand.md

Related Skills

相关技能

  • design-principles
    — the rules this skill applies
  • brand
    — the audit that diagnoses; its categories are this skill's acceptance test
  • brand-strategy
    — supplies the emotional target and muses (hard prerequisite)
  • logo-design
    — the mark, designed inside this system
  • web-design
    / 
    collateral-design
    — apply the approved system with real copy
  • design-critique
    (planned — not yet shipped) — will judge later work against what's recorded here
  • design-principles
    ——本技能应用的规则
  • brand
    ——负责诊断问题的审计；其品类是本技能的验收测试标准
  • brand-strategy
    ——提供情感目标与灵感缪斯（硬性前提）
  • logo-design
    ——在本系统内设计标志
  • web-design
    / 
    collateral-design
    ——结合真实文案应用获批的系统
  • design-critique
    （规划中——尚未上线）——将依据本记录评判后续工作