SKILL·055ACF

committed-aesthetic

avelikiy
更新于 6 days ago
2 次查看
89
13
89
在 GitHub 上查看
aidesign

关于

The `committed-aesthetic` skill defines a single, precise design style for consistent agent execution, unlike a catalogue of options. It's used to enforce a specific house style from scratch, prevent generic outputs, and ensure repeatable results. Apply it when starting a new brand surface without existing references or when you need an agent to reliably implement a defined aesthetic.

快速安装

Claude Code

推荐
主要方式
npx skills add avelikiy/great_cto -a claude-code
插件命令备选方式
/plugin add https://github.com/avelikiy/great_cto
Git 克隆备选方式
git clone https://github.com/avelikiy/great_cto.git ~/.claude/skills/committed-aesthetic

在 Claude Code 中复制并粘贴此命令以安装该技能

技能文档

A committed aesthetic beats a catalogue

The problem this exists for

ui-ux-pro-max ships 84 styles, 161 palettes and 57 font pairings, and design-advisor is wired to all of it. The catalogue is not the bottleneck.

The bottleneck is that a catalogue lets a model pick, and an unconstrained pick is the modal one. Asked for a landing page with no brand, any model reaches for the same safe centre: a soft gradient, a large sans heading, three rounded cards with a left accent border, generous neutral grey. Every input in the training distribution votes for it. Nothing in a catalogue entry saying "Swiss minimalism — clean, grid-based, generous whitespace" is specific enough to override that, because that sentence describes the safe centre too.

Compare a rule from a committed aesthetic:

every dimension is calc(fraction * 100vw) against a baseline width, so one page at 1024px reads as a perfect scaled-down of the same page at 1920px

You can check that. A design either obeys it or does not. That is the whole difference: a catalogue entry is a label, a committed aesthetic is a constraint.

The form

A committed-aesthetic skill has five parts. The first is the one that does the work.

1. Transferable DNA — numbered, specific, falsifiable

Between 8 and 14 rules. Each must be checkable by someone who was not in the room. Test every rule against this: could two designers disagree about whether a page obeys it? If yes, it is a vibe, rewrite it.

Vibe (useless)Constraint (usable)
"thin, elegant type""weights 100 or 300 only, never 400+"
"lots of whitespace""~60% of any above-the-fold area is empty"
"restrained motion""no shadows, no gradients, no bounce; one shadow exists in the system, on toasts"
"technical feel""product codes use en-dash U+2013, never hyphen"
"dark and dense""five surface steps, each ~5% apart; depth is a surface step, never a shadow"

2. What this does NOT do — the IP boundary, as a table

An aesthetic is inspired-by, not cloned. Name the specific assets that belong to whoever you are borrowing from — wordmark, icon set, information architecture, product names, copy — and for each give the reason and what to do instead. This is not a legal disclaimer at the bottom; it is a working table an agent reads while building, and it is the difference between an homage and a trademark problem. Where the source is a real company, default to inspired-but-distinct and ask before taking the clone path.

3. A component library that runs

One HTML file the agent can open, with every component rendered and its own CSS inline, copy-pasteable. A token table in prose is read and paraphrased; a rendered gallery is copied exactly. This is why the form works at all.

4. Templates for the signature patterns

The two or three moves nobody would reconstruct from the DNA list alone — the navigation shape, the section header, the hero. With swap these for your own brand marked in comments.

5. Reference captures, marked "study, do not pixel-copy"

Screenshots are for understanding the patterns. Building FROM a screenshot rather than from the rules produces a bad tracing of one page instead of a design in the family.

The banned list, and why we needed one

ui-ux-pro-max's anti-pattern tables are about accessibility, touch targets and performance. They are good and they are not about taste. Nothing in this repository told an agent what cliché looks like, so the only defence against slop was one sentence in design-advisor: "no generic clean modern minimal filler".

Banned by default on any surface without an existing system that calls for them:

  • A gradient as the page ground. Especially purple-to-blue, especially at 135°.
  • The card triplet: three equal cards, rounded corners, a coloured left border, an icon in a tinted circle.
  • Inter, Roboto, Arial, Open Sans, Fraunces, Poppins as the display face. Not because they are bad — because they are the default vote.
  • Emoji as iconography. Draw the icon. (Emoji are fine if the brand uses them.)
  • The hero that describes a category rather than taking a position. "The modern platform for X" is the sentence a model writes when it has not been told what to believe.
  • Glassmorphism, neumorphism, and a mesh-gradient blob unless the brief names one of them.
  • Symmetric three-column feature grids as the default second section.

A banned list is not a style. It removes the centre so a choice has to be made.

Using one

  1. If a system already governs, this skill does not apply — match what exists.
  2. If nothing governs and you can ask: offer 2–4 genuinely different directions, each an axis you can name, before building. Not five shades of one aesthetic — that is not a choice.
  3. If nothing governs and you cannot ask: commit to ONE, state the assumption in a line at handover, and put 1–2 low-fi alternates beside the deliverable, never instead of it.
  4. Once a direction is chosen, load its committed-aesthetic skill and execute the DNA rules literally. Cite the rule number for each choice, the way design-advisor already cites a ui-ux-pro-max rule.

Instances in this repository

Related: skills/ui-ux-pro-max/SKILL.md is the catalogue this stands beside, not a replacement for it — the catalogue is where you look when you need to KNOW something; this is what you load once you have DECIDED something.

GitHub 仓库

avelikiy/great_cto
路径: skills/committed-aesthetic
0
agentic-codingai-agentsclaude-codeclaude-code-pluginclaude-code-skillsclaude-code-subagents
FAQ

常见问题

什么是 committed-aesthetic Skill?

committed-aesthetic 是一个 Claude Skill,作者为 avelikiy。Skill 将 Claude 按需加载的说明和资源打包,让 Claude 无需额外提示即可执行与 committed-aesthetic 相关的任务。

如何安装 committed-aesthetic?

使用本页的安装命令:将 committed-aesthetic 作为插件添加到 Claude Code,或将其仓库克隆到 skills 目录,然后重启 Claude 以加载该 Skill。

committed-aesthetic 属于哪个分类?

committed-aesthetic 属于元分类。

committed-aesthetic 可以免费使用吗?

可以。committed-aesthetic 已收录在 AIMCP,可免费安装。

相关推荐技能

content-collections

Content Collections 是一个 TypeScript 优先的构建工具,可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计,提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。

查看技能
polymarket

这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持,涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流,可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。

查看技能
creating-opencode-plugins

该Skill帮助开发者创建OpenCode插件,用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式,适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。

查看技能
sglang

SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。

查看技能