20:"$Sreact.view_transition" 21:I[9754,["/_next/static/chunks/2t36f5p4pjzlb.js","/_next/static/chunks/1gyqeidjm5eoa.js","/_next/static/chunks/0_fjnxc4tt_l_.js","/_next/static/chunks/1ymt1shyhmu-4.js","/_next/static/chunks/3kv2zf0i7s2wc.js","/_next/static/chunks/1vs_3g_g75zmk.js","/_next/static/chunks/2g4dh802s8oz3.js","/_next/static/chunks/1bc3f0shukf1j.js","/_next/static/chunks/192kgs9v_t14j.js","/_next/static/chunks/0z6pw0v5c_lx5.js","/_next/static/chunks/08r60e6icmtkc.js","/_next/static/chunks/3trx6c0i7_x-3.js","/_next/static/chunks/40sro5bzp9bi_.js","/_next/static/chunks/1wgf3tk38y6cf.js"],"ProjectDetail"] 22:T463c, StyleSeed — 教你的 AI 固定的设计判断

STYLESEED · 0–100 评分门控

教你的 AI
固定的设计判断。

不是一种固定的美感——而是一套可判卷的方法。

StyleSeed 是一套开源的设计方法引擎，为 Claude Code、Codex 与 Cursor 这类 AI 编码代理装上固定判断。它用一条 0–100 的评分标尺验收每次产出：低于 80 分的界面在展示前被拦下整改。判断不属于任何一个人，而是留在你的仓库里，跨屏跨会话持续被维护。

规则74 条

评分门控0–100

技能23

许可证MIT

代理Claude Code · Codex · Cursor

View on GitHub 检测示例 · Live demo

01 · BAN ON SIGHT

先写下一套可核对的规则。

判卷的前提是标准成立。StyleSeed 内置 74 条规则，逐条禁止最常见的 AI 生成外观特征——那些一眼就能看穿的「AI 味」。先界定什么不该出现，再让代理在它之上做判断。

RULE · TYPOGRAPHY

禁通用字体堆砌

拦截不经思考的默认字体组合，避免在标题与正文之间失去层级对比。

RULE · SPACING

禁随机留白

把间距纳入体系判断，而非让代理凭感觉堆砌空隙。

RULE · COLOR

禁无意义色板

拒绝随机取色，要求色板服务于语义与品牌配方。

RULE · LAYOUT

禁一屏塞满

拦截内容堆叠与拥挤排布，为呼吸与分区留出余地。

02 · MEASURED

门控能测得到吗？能。

判断不该止于口头要求。在 120 个渲染基准单元上测量，强制执行后的门控让 Codex 与 Claude Code 的平均评分提升了 +5.3 分——而这段得分是达标、越过 80 的关键距离。

+5.3 分

平均提升 · 120 基准单元

得分是一道连贯的刻度：每一格都是一次渲染单元，校准后的代理更稳定地落在阈值上方。基准结果是环境相关的，不能当作放之四海的保证，但它提醒我们——可判卷的判断是可成长、也是可验证的。

03 · ORCHESTRA · 23 SKILLS

方法被拆成一支技能编排。

判断不是单条咒语，而是一套可持续的技能谱。StyleSeed 提供 23 个技能，各自负责一段工作，再由门控把它们串成一条可重复的流程。

ss-studio

ss-setup

ss-build

ss-palette

ss-grammar

ss-ref

ss-skin

ss-tokens

ss-orchestrate

ss-audit

ss-gate

ss-style

ss-image

ss-video

ss-carousel

ss-deck

ss-doc

ss-interaction

ss-ui

ss-verify

ss-compile

ss-qa

ss-review

ss-studioss-buildss-compile 核心流程 · 其余为配套技能

04 · AGENTS PIPELINE

同一套判断，跨代理生效。

固定判断不绑定某个工具。同一套规则、标尺与技能树，在主流 AI 编码代理里都按同一标准运转——从接入「评分门控」再到展示前复查，节奏一致。

Claude Code

npx skills add bitjaru/styleseed

安装技能树，随后在每次构建时启用评分门控与提示词标准。

Codex

仓库级技能

以仓库作用域读取规则，判断记录随变更进入同一提交记录。

Cursor

受支持代理

同一套判断与门控，跨 IDE 工作流保持一致。

CLOSING · THE GATE

一次验收，贯通全页。

这条横贯的 0–100 标尺就是 StyleSeed 的判断本身：差一分被整改拦下，达标则放行进入展示。好的设计方法，是可判卷、可测量、可重复的。

23:T4378, StyleSeed — Teach your AI fixed design judgment

STYLESEED · THE 0–100 GATE

Teach your AI
fixed design judgment.

Not one fixed aesthetic — a method you can grade.

StyleSeed is an open-source design-method engine that gives AI coding agents like Claude Code, Codex, and Cursor fixed judgment. It grades every output on a 0–100 scale, and screens below 80 must be revised before they are presented. Judgment never lives in one person — it lives in your repo, maintained across screens and sessions.

Rules74

Scored gate0–100

Skills23

LicenseMIT

AgentsClaude Code · Codex · Cursor

View on GitHub See it in action

01 · BAN ON SIGHT

First, write rules you can audit.

Grading only works when the bar is explicit. StyleSeed ships 74 rules that ban the most common tells of AI-generated output — the look you can spot at a glance. Define what must not appear, then let the agent make judgment on top of it.

RULE · TYPOGRAPHY

Ban lazy type.

Catch default font stacks and lost hierarchy between heading and body.

RULE · SPACING

Ban random whitespace.

Bring rhythm into the system instead of letting the agent guess.

RULE · COLOR

Ban meaning-free palettes.

Reject random swatches; require palettes tied to semantics and brand.

RULE · LAYOUT

Ban the overcrowded screen.

Stop stacked density and leave room for breath and clear sections.

02 · MEASURED

Can a gate be measured? Yes.

Judgment should not stay a verbal promise. Measured across 120 rendered benchmark cells, the enforced gate raised average scores for both Codex and Claude Code by +5.3 points — the span that matters when the threshold sits at 80.

+5.3 pts

average gain · 120 cells

A score is one continuous ruler: each cell is a rendered unit, and a tuned agent lands above the threshold more reliably. Benchmarks are environment-specific, never a universal guarantee — but they signal this much: gradeable judgment is growable and verifiable.

03 · ORCHESTRA · 23 SKILLS

Method, broken into a skill orchestra.

Judgment is not a single incantation — it is a spectrum of sustainable skills. StyleSeed ships 23 skills, each owning one stretch of work, linked by the gate into one repeatable pipeline.

ss-studio

ss-setup

ss-build

ss-palette

ss-grammar

ss-ref

ss-skin

ss-tokens

ss-orchestrate

ss-audit

ss-gate

ss-style

ss-image

ss-video

ss-carousel

ss-deck

ss-doc

ss-interaction

ss-ui

ss-verify

ss-compile

ss-qa

ss-review

ss-studioss-buildss-compile core pipeline · supporting skills alongside

04 · AGENTS PIPELINE

The same judgment, across agents.

Fixed judgment is not bound to one tool. The same rules, ruler, and skill tree run under the same bar in the leading AI coding agents — from wiring up the scored gate to reviewing before you present.

Claude Code

npx skills add bitjaru/styleseed

Install the skill tree; the scored gate and prompt bar engage on every build.

Codex

Repository-scoped skills

Rules load at repo scope, and the judgment trail lands in the same commit history.

Cursor

Supported agent

The same judgment and gate hold steady across IDE workflows.

CLOSING · THE GATE

One verdict, running through the page.

This 0–100 ruler is StyleSeed's judgment itself: one point short is held back for revision, on the threshold it passes through to the stage. Good design method is gradeable, measurable, and repeatable.

a:[["$","$20",null,{"enter":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-enter","default":"none"},"exit":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-exit","default":"none"},"update":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-update","default":"none"},"default":"none","children":["$","$L21","cmsqzq9hw010lzn19yokrmiok:zh-CN:beautified",{"project":{"id":"cmsqzq9hw010lzn19yokrmiok","slug":"bitjaru-styleseed","name":"styleseed","description":"为 AI 编码代理提供设计方法引擎","authorName":"bitjaru","authorVerification":{"verified":false,"verifiedAt":null},"category":"development_tools","subcategory":"data_ai_tools","experienceUrl":"https://styleseed-demo.vercel.app/","qualityScore":94,"featured":false,"officialSite":null,"views":{"original":{"available":true},"official":{"available":false,"reason":"disabled_by_policy"},"beautified":{"available":true,"reason":null},"defaultView":"beautified"},"beautifiedView":{"revisionId":"cmt2zgyuq007j62718i7kyg5g","kind":"artifact","artifact":{"schemaVersion":"localized-static-html-v4","kind":"static_html","locales":["zh-CN","en"],"defaultLocale":"zh-CN","documents":{"zh-CN":"$22","en":"$23"},"designDirections":{"schemaVersion":"dodot-design-directions-v1","directions":[{"id":"verdict-gate","name":"判决闸门","projectMetaphor":"设计方法是一道必须跨越的评分闸门——低于 80 分的界面在展示前被拦下整改，像一次权威的验收。","signatureVisual":"一条带 0-100 刻度的横贯标尺，一枚游标压住 80 分阈值；标尺下方分裂为「通过 Pass」与「整改 Revise」两个流向，成为贯穿全页的判卷焦点。","heroGeometry":"水平标尺 + 一刀两分的上下分割：游标卡在阈值，视觉张力由「差一分 vs 达标」的临界点位决定。","sectionSequence":["gate-intro","rules-matrix","benchmark-gain","skill-orchestra","agents-pipeline"],"surfaceModel":"判卷工作台：纸张质感的分节表面，阈值标尺贯穿每节左侧，用切分线（而不是卡片）分隔功能。","palette":"暖纸底 #FAF6EF、墨蓝判题 #1B2A4A、通过绿 #2F8F5B、整改红 #C24A3A、留白大量。","typography":"衬线标题（判令文书气质）+ 等宽数字与刻度 + 无衬线正文说明。","desktopComposition":"左 7 / 右 5 的不对称判卷页：左侧叙事线，右侧固定在视口的阈值尺与流向图，滚动时判题游标渐进。","mobileTransformation":"标尺由横贯改为纵向进度条贴左缘，通过/整改两流向改作上下堆叠的垂直分流带。","mediaStrategy":"无本地合格素材，移除媒体区；用阈值标尺、规则矩阵与基准点数构成的证据图形替代。","forbiddenPatterns":["不使用三卡指标墙","不使用浏览器/终端/编辑器壳","不使用营销式大标语+CTA构图","不使用发光渐变网格或粒子"],"compositionFingerprint":"0-100 阈值标尺横贯页面并切分通过/整改双流向，标题与叙事沿左侧判卷线连读。"},{"id":"rules-ledger","name":"规则判条","projectMetaphor":"设计判断由 74 条可审计的规则承载——每一条都是一根可展开的判条，像一本可核对的条例账本。","signatureVisual":"一排等距灯芯式判条（禁令条），每条锚定一条规则；悬停某根时其余略微退让，露出嵌入的规则编号与可读条款。","heroGeometry":"竖直等距的判条列阵 + 单根「亮点」放大对焦，规则编号以巨大衬线数字铺底。","sectionSequence":["rules-ledger-hero","ban-on-sight","rule-kinds","gate-mechanics","audit-trail"],"surfaceModel":"账簿双栏：每节以一条长分线对齐，判条作为行级别单位而非卡片，行与行之间用编号脊连接。","palette":"灰白纸 #F4F3EF、墨 #14181C、抑制色标注紫 #6A5B8C、撤销红 #A63A3A、低对比暖灰。","typography":"等宽字体承载全部规则编号与措辞，标题用高对比大字重，正文瘦身。","desktopComposition":"中央单列叙事 + 左右交错插入的判条，判条斜向伸出形成像琴弦拨动的韵律，滚动次序即规则优先级。","mobileTransformation":"判条斜伸改为水平触达式可点击展开，等距排布收窄为近距卡片行，编号脊保留。","mediaStrategy":"无本地素材，媒体区删除；整页以可展开判条和规则编号形成内容即视觉。","forbiddenPatterns":["不使用卡片功能墙","不使用仪表盘/IDE外壳","不使用粒子或发光算法装饰","不使用假评分图表"],"compositionFingerprint":"等距判条列阵作为唯一结构单元，标题串每根判条，巨大编号数字与正文排布同轴并读。"},{"id":"grammar-knot","name":"语法经脉","projectMetaphor":"多种设计语法像并行的经脉管线从固定判断核心向外编译——每一路输出一条可追踪的产物线。","signatureVisual":"一个正方核心向左右分出的多条编译管线，每条在末端落成一段代码釉条；管线交汇处有一枚「判断节点」接住加工。","heroGeometry":"中心放射的编译管线图 + 轴向镜像，判断核位于正中，语法分支沿对称轴分叉。","sectionSequence":["knot-core","grammar-threads","palette-sap","compilation-stream","studio-loop"],"surfaceModel":"工程走线板：冷白底、细线网格让位于真实的编译管线，核心用实体圆、产物用釉条承接。","palette":"冷白 #F7F8FA、石墨 #23262B、调度橙 #E2641F、管线青 #2B6E8C、语义绿 #2F8F5B。","typography":"几何无衬线标题 + 等宽产物标注 + 简洁正文，用管线耦合代替装饰。","desktopComposition":"中心对称的编译主图占据首屏一翼，另一翼为纵向产物流向清单，语法与产物逐行对位。","mobileTransformation":"放射管线折叠为自顶向下的纵向编译序列，判断核改为每段管线前端的嵌入式节点。","mediaStrategy":"无本地素材，删除媒体区；编译管线图即主视觉，佐以调色板釉条。","forbiddenPatterns":["不使用功能卡片墙","不使用营销Hero+CTA","不使用浏览器/终端壳","不使用同心圆渐变光环"],"compositionFingerprint":"中心判断核放射出语法管线，每种语法末落一段釉条，管线耦合成为纵贯全页的结构线。"}],"selectedDirectionId":"verdict-gate","selectionRationale":"三种方向均以证据驱动：verdict-gate 抓住「0-100 评分门控、低于 80 整改」这一最能表达项目机制的可视化核心，将质量验收做成横贯全页的阈值标尺，证据契合度最高且与营销 Hero、三卡指标、功能卡片墙、IDE 壳等被禁默认完全不同；grammar-knot 突出多种语法与编译管线，palette 语义、studio 工作流也真实支撑，但偏重描述多于可判定的验收动作；rules-ledger 聚焦 74 条规则与可审计性，结构清晰却稍弱于把「门控」作为叙事引擎的冲击力。选择 verdict-gate：它让评分门控成为穿越整页的判卷主视觉，手机端重构为纵向阈值尺后概念不丢失，且历史资产仅保留「纸感分节 + 0-100 刻度」的可复用构图语言（以 rebuild 处理），其余旧文字与代码一律不采用。"},"designBrief":{"schemaVersion":"dodot-design-brief-v1","selectedDirectionId":"verdict-gate","projectMetaphor":"设计方法是一道必须跨越的评分闸门——低于 80 分的界面在展示前被拦下整改，像一次权威的验收。","signatureVisual":"一条带 0-100 刻度的横贯标尺，一枚游标压住 80 分阈值；标尺下方分裂为「通过 Pass」与「整改 Revise」两个流向，成为贯穿全页的判卷焦点。","compositionFingerprint":"0-100 阈值标尺横贯页面并切分通过/整改双流向，标题与叙事沿左侧判卷线连读。","heroGeometry":"水平标尺 + 一刀两分的上下分割：游标卡在阈值，视觉张力由「差一分 vs 达标」的临界点位决定。","sectionSequence":["gate-intro","rules-matrix","benchmark-gain","skill-orchestra","agents-pipeline"],"surfaceModel":"判卷工作台：纸张质感的分节表面，阈值标尺贯穿每节左侧，用切分线（而不是卡片）分隔功能。","palette":"暖纸底 #FAF6EF、墨蓝判题 #1B2A4A、通过绿 #2F8F5B、整改红 #C24A3A、留白大量。","typography":"衬线标题（判令文书气质）+ 等宽数字与刻度 + 无衬线正文说明。","desktopComposition":"左 7 / 右 5 的不对称判卷页：左侧叙事线，右侧固定在视口的阈值尺与流向图，滚动时判题游标渐进。","mobileTransformation":"标尺由横贯改为纵向进度条贴左缘，通过/整改两流向改作上下堆叠的垂直分流带。","mediaStrategy":"无本地合格素材，移除媒体区；用阈值标尺、规则矩阵与基准点数构成的证据图形替代。","evidencePlan":["0-100 评分门控与「低于 80 分须整改」用阈值标尺与双流向可视化，依据 summary 证据 claim#2","74 条规则禁 AI 外观特征以规则矩阵呈现，依据 summary 证据 claim#2","120 渲染基准单元 +5.3 分平均提升以刻度增量图形呈现，依据 summary 证据 claim#3","23 个技能（ss-studio/ss-setup/ss-build）与 Claude Code/Codex/Cursor 多代理以技能编排与代理管线呈现，依据 summary 证据 claim#4"],"localeContract":{"zh-CN":"标题用「教你的 AI 固定的设计判断 / 不是一种固定的美感」与一句话说明；判卷用「通过/整改」「低于 80 分须在展示前修改」；规则、技能、多代理名保留原文专名。","en":"Headline uses 'Teach your AI fixed design judgment / Not one fixed aesthetic' with a one-line summary; verdict copy uses 'Pass / Revise' and 'below 80 must be revised before presentation'; rule, skill, and agent proper names stay in original form."},"forbiddenPatterns":["不使用三卡指标墙","不使用浏览器/终端/编辑器壳","不使用营销式大标语+三卡 CTA 构图","不使用发光渐变、网格或粒子替代项目专属视觉","不使用假标签、搜索、开关或伪评分图表"],"historicalReuse":{"sourceAssetId":"HEAD:deepseek-v2/bitjaru-styleseed","decision":"rebuild","rationale":"历史资产的文字与代码均不可信、仅作隔离视觉参考；其「种子生长」构图与判卷主题不符，配色与排版也不能直接承载「评分门控」机制，故选 rebuild，仅提炼仍契合当前事实的构图语言。","reusedElements":["暖纸张感的浅底与纸张分节表面模型","以体系化分节连接内容、而非卡片墙的排版节奏","0-100 阈值刻度的数字计量气质与纸面留白比例"],"discardedElements":["旧「种子→生长→grammar 分支」放射构图（与评分门控机制不符）","旧 seed-hero/palette/grammar/compilation/media 章节顺序","旧紫色 #7c5cbf / 琥珀 #d97706 色板标记","旧 typography、reactBits、状态与全部已验证文字表达"]}},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"页面忠实实现 verdict-gate 判卷闸门方向：0-100 标尺游标压 80、下分通过/整改双流向作为 signature visual 真实存在（结尾 SVG 图），全页以暖纸底+墨蓝判题+通过绿/整改红+衬线标题/等宽刻度构成，五段章节序列（gate-intro/rules-matrix/benchmark-gain/skill-orchestra/agents-pipeline）与 Brief 一致，无英雄Hero三卡墙、无设备壳、无装饰渐变网格，74 条规则/120 单元+5.3 分/23 技能 ss-studio·ss-setup·ss-build/Claude Code·Codex·Cursor/MIT 均有 summary 证据支撑，双语文案等价且自然。仅一处非阻断建议：移动端左缘 0-100 刻度线被隐藏而非按 Brief 转为贴左纵向进度条，但结尾闸门图仍保留标尺与通过/整改分流，signature 概念在手机端未丢失。","checks":{"projectFit":{"passed":true,"evidence":"sectionSequence 依次为 gate-intro/rules-matrix/benchmark-gain/skill-orchestra/agents-pipeline，另附 closing gate-diagram（共6节，在 4-7 范围）；palette 用 #FAF6EF/#1B2A4A/#2F8F5B/#C24A3A 与 Brief 完全一致；衬线标题(宋体/Georgia)+等宽数字刻度+无衬线正文符合 typography；判卷闸门隐喻与「低于80须整改」叙事贯穿"},"evidenceBound":{"passed":true,"evidence":"claim#2 的 74 条规则+0-100 门控+低于80展示前修改在 hero 与 rules-matrix 呈现；claim#3 的 120 渲染基准单元与 +5.3 平均提升在 benchmark-gain 呈现；claim#4 的 23 技能(ss-studio/ss-setup/ss-build)与 Claude Code/Codex/Cursor 多代理在 skill-orchestra 与 agents-pipeline 呈现；MIT 许可证在 hero meta 呈现；均有 summary.evidence 依据"},"signatureVisual":{"passed":true,"evidence":"结尾 gate-diagram SVG 真实实现了「0-100 刻度横贯标尺、游标框压 80 阈值、下方分裂为通过 Pass 与整改 Revise 两流向」，与 signatureVisual 描述逐一对应，非装饰性示意"},"mobileTransformation":{"passed":true,"evidence":"手机 CSS 形成真实重构：matrix 网格 4列->2列、orch 23列->6列、agents 3列->1列、字号/内边距收缩、prefers-reduced-motion 处理；结尾闸门 SVG 自适应保留标尺与通过/整改分流。注意：左缘 .gate-tick 在移动端 display:none，未按 Brief 转为贴左纵向进度条（见警告）"},"localeParity":{"passed":true,"evidence":"zh-CN 标题「教你的 AI 固定的设计判断」+副题「不是一种固定的美感」，en 标题「Teach your AI fixed design judgment」+副题「Not one fixed aesthetic」与 localeContract 逐条对应；通过/整改 vs Pass/Revise、低于80展示前修改 vs below 80 must be revised before presentation 等价；74/120/+5.3/23/MIT 双语一致；技能与代理专名 ss-studio/Claude Code/Codex/Cursor 双份均保留原文"},"antiTemplate":{"passed":true,"evidence":"未使用营销式大标语+三卡 CTA 墙；无浏览器/终端/编辑器壳与设备壳；作品以判卷线+规则矩阵+基准刻度+技能编排+代理行+闸门图组织，非卡片功能墙；无发光渐变/网格/粒子替代视觉"},"noFakeUi":{"passed":true,"evidence":"qa-report 两语言、桌面/移动共 4 视口 fakeControlCount 均为 0；gauge 刻度条、orch 技能格、矩阵、SVG 标尺均为有证据锚定的数据可视化而非假评分图/假开关/假搜索/假标签；链接为真实外链且 rel=noopener"}},"issues":[{"code":"mobile-gate-line-hidden","severity":"warning","locale":"all","description":"手机端 @media (max-width:800px) 将左缘 .gate-tick（0-100 刻度）设为 display:none，而 Brief mobileTransformation 明确要求标尺「由横贯改为纵向进度条贴左缘」，通过/整改改作上下堆叠垂直分流带；实施将左缘刻度直接隐藏而非转为 0-100 纵向进度条，未完整落实该条明确的重构指令。","repairInstruction":"在移动端保留贴左缘的 .gate-tick 并改作纵向 0-100 进度条（保持红色底线+绿色达标段），或将左右分流改为上下堆叠的两条垂直带，使 signature 概念在手机端以 Brief 指定的纵向形式呈现而非被隐藏。"}]},"repairEvidence":[],"provenance":{"executor":"opencode","model":"deepseek/deepseek-v4-flash","opencodeVersion":"1.1.58","promptVersion":"dodot-showcase-deepseek-v4","inputHash":"1abb92a2c90f7c11e5fa761076c718169ca08d6aab48d006927f1f399ce11e17","artifactHash":"a0bf854d89919ff2d83d709f2246bad1d962719f85d87964fd48f2e5b4d8e065","designSkill":"dodot-showcase-design","historicalSeed":{"source":"git_head","sourceCanonicalKey":"github:bitjaru/styleseed","assetId":"HEAD:deepseek-v2/bitjaru-styleseed","sourceSchemaVersion":"v1","sourceGeneratorVersion":"anthropic:claude-fable-5","sourceArtifactHash":"75d6b5e434cb413fdf51852546fe707eff75fed718761ad61f97a487f397a074","disposition":"repair_or_regenerate","seedPayloadHash":"75a830b9e9604dfd230b61f246439370ac488ea0b6c3800fb35d6c65a3c36b29"}},"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1633,"cjkCount":745,"latinLetterCount":499,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3929,"h1Top":158.671875,"bottomInset":195.15625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"8c6ef725b1ede053d6dfe3cbe5191ebc9a004e50c165275f07423d2c04dae654","middle":"03aa88468278d86484334e1a36bc7d7b9aec925e7c15cef03a61be7139931a88","bottom":"277aa82088da5c2a0e29d4fdd4bf0a7fe05ef5a58647a0b592876ee3c9662e6a"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1633,"cjkCount":745,"latinLetterCount":499,"scrollWidth":390,"clientWidth":390,"scrollHeight":4228,"h1Top":110.671875,"bottomInset":151.59375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"fefa3c57a1e177786af78cbe9e2183cc7c385340fd4cf53cb1094c33db411ca0","middle":"78089ae2d5ccd2a351222070eaa0f73fd7d441d84673033489a54975d66f06b0","bottom":"aa11e8c3fc40b00eb328d8ae2c5429422ff36fb9d73b4737f840fe6888afcec9"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3279,"cjkCount":0,"latinLetterCount":2536,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":4150,"h1Top":158.671875,"bottomInset":194.984375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"953c96f6124e0dc8537f3599bef22b9985bc27e3effa1f6adb716100a3fba4d2","middle":"4e92b84987041c0c53a0de1b0ef6d8552f04f9aeb6ff8afe74ab5a13b4ab2113","bottom":"f2dd85607ae8174450e8550ab077f63dbc4d76216f88c0d9d76930b663b0469f"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3279,"cjkCount":0,"latinLetterCount":2536,"scrollWidth":390,"clientWidth":390,"scrollHeight":4635,"h1Top":110.671875,"bottomInset":152.140625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"9f845975a61508c5e5da576799a1836b4063aacacb9467c7ad3bceac5e007216","middle":"ed7a278ed769755293d45b044eda5f9d7b4c22515dc287df9e5175f160ef5d7b","bottom":"d5cb8049d7b3909158955ae70fa16d7f9bb5f4e26e173d55999d17a0ac1e38ea"},"consoleErrors":[],"networkRequests":[]}]}},"generatorVersion":"opencode-1.1.58:deepseek/deepseek-v4-flash:dodot-showcase-deepseek-v4","publishedAt":"2026-08-20T12:23:48.620Z"},"rawExtraction":{"projectRevisionId":"cmt2zgyug007d627199yc5534","projectVersion":1,"sources":[{"canonicalKey":"github:bitjaru/styleseed","role":"primary","revisionVersion":1,"url":"https://github.com/bitjaru/styleseed"}],"facts":[{"key":"repository_owner","value":"bitjaru","evidence":[{"path":"repository.owner","excerpt":"bitjaru"}]},{"key":"repository_name","value":"styleseed","evidence":[{"path":"repository.name","excerpt":"styleseed"}]},{"key":"description","value":"Open-source design-method engine for Claude Code, Codex & Cursor. Fixed design judgment, multiple grammars, reference compilation, semantic palettes, and Studio workflows for verified UI, interaction, image/video jobs, carousels, decks & docs. 22 skills · MIT.","evidence":[{"path":"repository.description","excerpt":"Open-source design-method engine for Claude Code, Codex & Cursor. Fixed design judgment, multiple grammars, reference compilation, semantic palettes, and Studio workflows for verified UI, interaction, image/video jobs, carousels, decks & docs. 22 skills · MIT."}]},{"key":"primary_language","value":"TypeScript","evidence":[{"path":"repository.language","excerpt":"TypeScript"}]},{"key":"license","value":"MIT","evidence":[{"path":"repository.license","excerpt":"MIT"}]},{"key":"topics","value":"agent-skills, ai-coding, ai-design, ai-ui, claude-code, claude-skills, codex, cursor, design-judgment, design-system, design-tokens, nextjs, radix-ui, react, shadcn, shadcn-ui, tailwindcss, typescript, ui-design, vibe-coding","evidence":[{"path":"repository.topics","excerpt":"agent-skills, ai-coding, ai-design, ai-ui, claude-code, claude-skills, codex, cursor, design-judgment, design-system, design-tokens, nextjs, radix-ui, react, shadcn, shadcn-ui, tailwindcss, typescript, ui-design, vibe-coding"}]},{"key":"source_count","value":"1","evidence":[{"path":"projectRevision.sources","excerpt":"github:bitjaru/styleseed"}]}],"taxonomy":{"category":"development_tools","subcategory":"data_ai_tools","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Open-source design-method engine for Claude Code, Codex & Cursor. Fixed design judgment, multiple grammars, reference compilation, semantic palettes, and Studio workflows for verified UI, interaction, image/video jobs, carousels, decks & docs. 22 skills · MIT."}]},"lifecycle":{"stage":"prototype","confidence":0.65,"evidence":[{"path":"readmeExcerpt","excerpt":"

# StyleSeed ### Teach your AI fixed design judgment.
Not one fixed aesthetic. _{Your AI coding agent sets up the design system, a scored gate keeps the quality, and the decisions live in your repo, maintained across screens and sessions — designed-looking products without a design team. Open source, for Claude Code · Codex · Cursor.}

教你的 AI固定的设计判断。

先写下一套可核对的规则。

禁通用字体堆砌

禁随机留白

禁无意义色板

禁一屏塞满

门控能测得到吗？能。

方法被拆成一支技能编排。

同一套判断，跨代理生效。

一次验收，贯通全页。

Teach your AIfixed design judgment.

First, write rules you can audit.

Ban lazy type.

Ban random whitespace.

Ban meaning-free palettes.

Ban the overcrowded screen.

Can a gate be measured? Yes.

Method, broken into a skill orchestra.

The same judgment, across agents.

One verdict, running through the page.

教你的 AI
固定的设计判断。

Teach your AI
fixed design judgment.