1实现原理 · 为什么它能做到
功能全部由 71 行单文件指令文本实现:零脚本、零工具、零外部文件。机制是「把反模板审美编码为可执行的上下文规则」,靠宿主模型执行 + 自律批判产出,不依赖任何自动化。
git ls-files frontend-design/ = LICENSE.txt + SKILL.md(71 行,无 scripts/references/templates);全文 grep 无 URL/命令/代码围栏
用「负样本校准法」避免模板化:先把当前 AI 生成视觉语汇归纳成 5 条带具体色值/模式的特征清单,让模型能自查输出是否落入模板簇——不是空喊『要有创意』,而是给出可比对的规避对象。
For calibration, AI-generated design right now clusters around some traits: 1. a warm cream background (near #F4F1EA) with a high-contrast serif display and a terracotta or warm-clay accent (often near #D97757 — Anthropic's own Claude-interaction accent...); ... 5. template chrome that appears whatever the subject: a tracked-out ALL-CAPS eyebrow label above every heading; meta strings joined with middle dots ('A · B · C'); ... tinted near-black (#0B0B0B, #111) standing in for black; a monospace face for small data labels; a '→' appended to link and button text.
强制「题材定位」作为差异化来源:brief 缺题材时先自行提出 subject/audience/primary job 并与客户确认;题材的行业、物料、行话被规定为视觉选择的出发点。
If the brief does not identify what the product or subject matter is, identify it yourself before designing, and confirm with the client. You can come up with one concrete subject, the design's audience, and the design's primary job, as a proposal. ... The subject's industry, subject matter, materials, and vernacular are where distinctive visual choices come from
两遍流程 + uniqueness 自测门槛:先产出紧凑 token 计划(色/字/布局/原则),写代码前必须对照 brief 评审,凡读起来像『任何同类页都能套的默认』即修改并说明改了什么;确认相对独特后才动手写码。
Work in two passes. First, brainstorm a short design plan based on the client's design brief: create a compact token system with color, type, layout, and principles. ... Then review that plan against the brief before building: if any part of it reads like the generic default you would produce for any similar page ... revise that part, say what you changed and why. Only after you've confirmed the relative uniqueness of your design plan should you start to write the code, following the revised plan.
内嵌排印与版式领域知识作为硬规则:字阶参照 The Elements of Typographic Style、行长 <80 字符、serif 行高略增;并给出『生成页最常见排版穿帮』的禁止清单。
set a clear type scale following the default guidance of The Elements of Typographic Style ... Default to line lengths of less than 80 characters. Serif typefaces can have slightly longer line lengths; give serif body text slightly more line-height than a sans-serif. ... Avoid these default typographic treatments; they are the commonest tells of a generated page: - Accenting just a single word or phrase in a headline ... - Using all caps for labels.
把「界面文案」提升为一等设计内容:CTA 用主动语态且精确("Save changes," not "Submit")、动作命名全程一致(Publish → toast 显示 Published)、错误/空状态写为引导语而非情绪。
Use active voice as default. A CTA says exactly what happens when it is used: "Save changes," not "Submit." An action keeps the same name through the whole flow, so the button that says "Publish" produces a toast that says "Published." ... Treat failure and emptiness as moments for direction, not mood. Explain what went wrong and how to fix it ... Errors don't apologize, and they are never vague about what happened. An empty screen is an invitation to act.
结构诚实与动效克制规则:序号/分隔线等结构装置必须编码真实信息(如内容确为序列才用 01/02/03);非用户触发的动效只保留『单个编排队列时刻』,逐块 fade-slide-up + 全卡 hover 被点名批评为 AI 味。
Many generic designs use numbered markers (01 / 02 / 03), but that's only appropriate if the content actually is a sequence — like a stepped process or a timeline. Before adding numbered markers, check the content really is a sequence. ... A single orchestrated moment — one page-load sequence or one reveal — lands better than scattered effects; fade-and-slide-up entrances on each section and hover transitions on every card are the generic default and read as AI-generated.
质量底线与自批判闭环:单个记忆点 + 其余克制("Spend your boldness in one place")、响应式/键盘焦点/reduced motion/无障碍硬底线、条件性截图自评("a picture is worth 1000 tokens")、Chanel『出门前摘掉一件配饰』类比。
Spend your boldness in one place. Let one element be the memorable thing, keep everything around it quiet and disciplined, and cut any decoration that does not serve the brief. Build to a quality floor without announcing it: responsive down to mobile, visible keyboard focus, reduced motion respected, visually accessible ... Critique your own work as you build, taking screenshots to review if your environment supports it — a picture is worth 1000 tokens.
2核心能力
4风险提醒 风险提醒:绿色 · 放心使用
- 无强制机制,效果全押宿主模型自律 — uniqueness 自测(L53)、截图自批判(L59)均无工具检查点;弱模型可能跳过评审直接产出模板风。效果随宿主模型波动 [INFERENCE:源码无校验面,按执行面依赖标注]
- 规避清单时效性有限 — 5 类特征锚定当前 AI 视觉语汇(本 commit 2026-09-03 刚为此大改 #1713);语汇演化后清单会失真,需持续跟进版本
- 『brief 永远赢』的绝对化指令 — L45 要求客户 brief 点名要模板风也照做;若用户 brief 审美低质,skill 文本本身不授权专业否决,结果可能既模板又不独特(这是其设计取舍,但对期望『帮我变好看』的用户是落差点)
- 文案/错误态建议可能催生编造 — 『Errors ... are never vague about what happened』(L69) 需要模型确知产品真实行为;无配套检索/验证工具时,模型可能虚构具体错误原因与修复步骤
- 不产出结构化设计资产 — token 系统停留在文字描述(4–6 个命名 hex 等),不生成 design token 文件/组件库代码,跨会话与跨人复用依赖文本传递 [INFERENCE:源码无任何导出/文件生成逻辑]
5第二遍独立确认
- [ok] 目录文件清单(有无隐藏脚本/数据) — git ls-files frontend-design/ 仅 LICENSE.txt + SKILL.md;glob 无其他文件,无隐藏资源
- [ok] SKILL.md 中 URL/CLI/API/包引用 — 正则扫 http|curl|fetch|exec|api_|npm|pip|gh |git |env[ 等零命中;全文本唯一 URL 是 LICENSE.txt 中官方许可地址 http://www.apache.org/licenses/(非功能依赖)→ external_deps 空数组成立
- [ok] 代码围栏 / 注入标记(```、ignore previous、system-reminder 等) — 零命中;SKILL.md 为纯 markdown 叙述
- [ok] 5 条 AI 特征清单及色值逐字存在 — L39-43 逐字含 #F4F1EA、#D97757、rgba(0,0,0,.1)、#0B0B0B、#111、'A · B · C'、'→',非编造
- [ok] meta.license / stars / last_push 与源一致 — gh api repos/anthropics/skills:stars 174801、pushed_at 2026-09-03T16:37:14Z、license null(仓库根无 LICENSE,故 meta 以 skill 目录内 Apache-2.0 LICENSE.txt 为准并注明差异)
- [ok] description 与正文能力是否夸大 — description 措辞为 'Guidance/Helps with',正文确为指南文本,未宣称自动化/保证结果——无吹牛
- [ok] 『两遍流程/截图自批判』是否真被强制 — L47-53 流程顺序明确(评审→修改说明→才写代码),L59 截图附 'if your environment supports it' 条件;但执行严格性依赖宿主模型自律,源码内无强制检查点——此点已在 how_it_works 标注为执行面依赖
- [ok] 行号/引用与文件实际内容对应 — 二次 grep 带行号比对全部引用段落,无张冠李戴;截断引用未扩写
6结论
67d68803068d5ae9…41bbe19d1a