Token导航 LogoToken导航

Agent Skill · 研究检索

extract-saas-design(提取 SaaS 设计)

用于辅助界面设计、视觉规范、排版、配色、布局和交互体验优化。它适合让 Agent 根据产品场景整理页面结构、生成 UI 方案、检查视觉一致性或改进组件层级。使用时需要结合现有品牌、设计系统和用户任务,不应只堆装饰元素;涉及真实页面改动时,应通过截图或浏览器预览检查文本溢出、对齐和响应式表现。

extract-saas-design 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为只读;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 52,401。

总安装 559周安装 24GitHub Stars 5下载 196核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:extract-saas-design(提取 SaaS 设计)
来源仓库:https://github.com/yigitkonur/skills-by-yigitkonur
仓库路径:skills/extract-saas-design
安装命令:
npx skills add https://github.com/yigitkonur/skills-by-yigitkonur --skill extract-saas-design
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/yigitkonur/skills-by-yigitkonur --skill extract-saas-design

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于将现有 SaaS 仪表盘或管理界面的视觉系统提取为结构化文档。适合归档当前 UI 规范、组件规格和布局模式,便于跨项目还原或审计。聚焦于“存在即记录”,不修改原始代码,仅输出设计事实和行为说明。使用时需结合品牌与设计系统,避免装饰性堆砌;涉及页面改动应通过截图验证表现。

extract-saas-design 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/extract-saas-design

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/yigitkonur/skills-by-yigitkonur --skill extrac…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

extract-saas-design 怎么安装?

在终端执行 npx skills add https://github.com/yigitkonur/skills-by-yigitkonur --skill extract-saas-des…。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

extract-saas-design 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

extract-saas-design 需要什么权限?

权限级别是只读。该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

extract-saas-design 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

extract-saas-design 用的人多吗?

公开数据是累计安装 559 次,近一周安装 24 次,下载 196 次,GitHub 5 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

extract-saas-design 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 只想先试试效果、不希望它改动本地文件的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
extract-saas-design

提取 SaaS 设计 · 本页

5595只读未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Extract SaaS Design

Extract the implemented visual system of an existing SaaS dashboard, admin panel, or internal tool into structured documentation. This is forensic work: document what exists, where it comes from, and how it behaves.

Use this skill when

  • The user wants the current visual system documented: tokens, component specs, layout patterns, dark mode, or dashboard-specific UI behavior.
  • The user wants to recreate the current look in another codebase through documentation, not by copying source.
  • The user wants scoped extraction: foundations only, named components, a navigation or data-display slice, or an evidence-based consistency audit.

Do not use this skill when

  • The real task is to build, port, or generate a working UI.
  • The real task is redesign, restyling, “make it better,” or speculative UX advice.
  • The target is mainly a marketing or brochure site instead of a SaaS/dashboard/admin product.
  • There is no implemented UI evidence to inspect.

Operating boundaries

  1. Document, don't rebuild. Output documentation and evidence, not production UI code.
  2. Honor the requested scope. Do not default to a full-system extraction unless the user asked for it.
  3. Work from evidence only. Read actual source files and supporting artifacts; never infer unimplemented behavior.
  4. Resolve aliases. Convert utilities, tokens, and variables to final values and explain their semantic role.
  5. Absence is evidence. If a state or mode does not exist, write not implemented.
  6. Separate canon from noise. Document the dominant system, then flag inconsistencies and exceptions explicitly.

Scope routing

If the user asks for...Example requestsDeliver
Full extraction / "visual DNA" / "document the design system""Extract the full design system", "Document the visual DNA"system.md, foundation docs, relevant component docs, and meta-docs (INDEX.md, _summary.md)
Foundations / tokens / theme only"Extract the color tokens", "Document the spacing system"system.md plus foundation docs only; skip component docs unless requested
Specific components or patterns"Document the sidebar", "Extract button specs"Minimal foundation context plus full docs for the named components or patterns only
Consistency / quality / accessibility audit"Audit the color consistency", "Check accessibility"Evidence-based findings grounded in extracted docs; do not drift into redesign proposals

Minimal workflow

  1. Frame the job

- Restate what is in scope, what is out of scope, and which artifacts count as ground truth. - Prefer the smallest extraction that satisfies the request.

  1. Build the evidence base

- Detect the styling stack first -- this determines everything else: - Offline snapshot / plain HTML+CSS mode: if the input is just HTML/CSS files with no package.json, no src/, and no Tailwind/shadcn markers, treat the CSS files as the token source of truth. Skip framework-specific checks and write not implemented / N/A where those systems do not exist. - Tailwind v4 uses @import "tailwindcss" in CSS and @theme blocks instead of tailwind.config.js. Check for both. - shadcn/ui: look for components/ui/ directory and @/lib/utils imports with cn(). - CSS entry point: find globals.css or app.css -- this is where tokens live. - Normalize the target root before running any routed command. Reference snippets that mention src/, app/, or package.json are common repo examples, not fixed paths. In offline snapshot mode, search the snapshot root directly. See references/extraction/target-modes.md. - Identify token sources: global CSS, theme files, Tailwind config (if v3), component source, Storybook. - Resolve representative token chains end-to-end before writing docs: Tailwind class -> CSS variable -> final value bg-primary -> var(--primary) -> oklch(0.21 0.006 285.88) Plain CSS snapshot chain.metric-card -> var(--card-bg) -> #111827 - Detect the color space: modern shadcn uses oklch(), older uses hsl(). - Check for variant systems: CVA (class-variance-authority) defines variant/size matrices. - Capture recurring values, mode switching, state styling, and composition patterns.

  1. Extract foundations first when needed

- Spacing, colors, typography, shadows/depth, radius, animations, and shared state conventions. - For dashboard products, explicitly check sidebar overrides, chart palettes, density zones, and tabular/monospace usage.

  1. Extract components and patterns in scope

- For each component, document anatomy, variants, sizes, states, state transitions, dark-mode differences, and composition. - In offline snapshot mode, treat repeated DOM fragments plus their matching selectors as the component evidence. Document only what the snapshot proves; if the capture does not expose a variant or state, call out the gap explicitly instead of inventing it. - Treat sidebars, tables, command palettes, and form layouts as mega-components that must be decomposed.

  1. Verify before handoff

- Use references/quality-checklist.md as your verification checklist. - Every important value should be resolved to a final computed value, not just a class name or alias. - Every relevant state should be explicit, including not implemented. - Pay special attention to disabled, focus-visible, loading, error/invalid, empty, and dark-mode behavior. - Document opacity semantics and animation composition when they carry design meaning. - If a recreator would still need to guess, the extraction is incomplete.

  1. Package the output

- Create all extraction artifacts in the .design-soul/ directory at the codebase root (not inside the skills repo). - "Codebase root" may be the original repo root or a writable working copy of the target UI (for example a copied fixture under a work directory). The rule is about target ownership, not about the original location on disk. - Use the .design-soul/ structure defined in references/documentation/output-format.md. - Follow the exact templates and file layout from the references instead of inventing new formats.

Guardrails: do this, not that

Do thisNot that
Resolve bg-primary, rounded-md, p-3, and similar aliases to actual values and meaningsStop at class names or token aliases
Write Hover: not implemented / Loading: not implemented when absentInvent expected states or “best practice” behavior
Scope to the user's requestDefault to documenting the entire app
Document composition and contextDescribe components only in isolation
Decompose mega-components into sub-parts and modesFlatten sidebar, table, or form layout into one vague blob
Flag inconsistencies with file evidenceSmooth them over into a cleaner system than the product really has
Keep audits evidence-basedTurn the task into redesign or subjective critique

Recovery rules

  • Evidence is incomplete: look for adjacent ground truth (global CSS, theme files, config, Storybook, screenshots). If still unresolved, document the gap explicitly.
  • The codebase is too large: split into foundations plus the requested categories or components; do not widen scope silently.
  • Values conflict across files: document the dominant pattern, then list exceptions with source references.
  • Dark mode or a state does not exist: say so plainly; absence is part of the extraction.
  • The task starts drifting toward implementation: stop generating build instructions and return to documentation deliverables.

Output contract

When you are asked to produce extraction files, keep the structure predictable:

  • system.md for foundations in scope
  • numbered foundation docs when doing system-level extraction
  • one numbered doc per extracted component
  • numbered pattern docs in components/[app-specific]/ when the scope includes dashboard-level compositions or product-unique flows
  • INDEX.md and _summary.md for multi-document or full extractions

Use the exact structure, naming, and content expectations in references/documentation/output-format.md, references/system-template.md, and references/component-template.md.

Reference routing

Start with the smallest relevant set. Only expand if the task genuinely needs more depth.

NeedWhat it containsRead
Target root and snapshot path handlingHow to reinterpret repo-style example commands for repo-backed UIs vs copied HTML/CSS snapshotsreferences/extraction/target-modes.md
Foundation extraction methodAgent prompts for scanning tokens + grep commands + output templatesreferences/extraction/target-modes.md, references/foundations-agent.md, references/extraction/color-extraction.md, references/extraction/typography-extraction.md, references/extraction/spacing-extraction.md, references/system-template.md
Component extraction methodAgent prompt for per-component visual specs + template with all required sectionsreferences/extraction/target-modes.md, references/components-agent.md, references/component-template.md, references/extraction/icons-and-assets.md
Dashboard/admin-specific patternsSidebar, metrics, tables, charts, cmdk, mega-component decompositionreferences/extraction/target-modes.md, references/dashboard-patterns.md, references/layout/grid-and-responsive.md
Token translation and namingW3C DTCG, CSS custom properties, oklch format, shadcn naming patternreferences/extraction/target-modes.md, references/tokens/token-formats.md, references/tokens/naming-conventions.md
Packaging the docs.design-soul/ directory structure, INDEX.md and _summary.md templatesreferences/documentation/output-format.md, references/system-template.md, references/component-template.md
Verification and auditExtraction completeness checklist, token/component consistency matrix, WCAG contrast checksreferences/extraction/target-modes.md, references/quality-checklist.md, references/audit/consistency-checklist.md, references/audit/accessibility-review.md

Steering Notes for Agents

These notes prevent the most common extraction failures. Read the linked references for full detail.

Tailwind v4 Detection

Grep for @import "tailwindcss" or @theme in CSS files — if found, the project uses Tailwind v4 (CSS-native config, no tailwind.config.js). If tailwind.config.ts exists with theme.extend, it uses Tailwind v3. Always check before extracting tokens. See references/foundations-agent.md for detection commands.

oklch Color Space

Modern shadcn/ui (2024+) uses oklch(L C H) instead of hsl(). L=lightness (0-1), C=chroma (0-0.4), H=hue (0-360). Grays have C < 0.01. Always note lightness to distinguish dark/light colors. See references/extraction/color-extraction.md and references/tokens/token-formats.md for reading guides.

CVA and cn() Patterns

The cva() call IS the component's visual spec — extract every variant/size key-value pair. The cn() utility merges classes via tailwind-merge where the last class wins. See references/component-template.md for the CVA extraction checklist and cn() merging guide.

Token Chain Resolution

Never stop at the Tailwind class. Resolve the full chain: bg-primary → var(--primary) → oklch(0.205...). Some tokens chain through multiple variables. Resolve ALL the way to the literal value. See references/extraction/color-extraction.md for variable chain resolution.

Final reminder

This skill should feel like an orchestrator, not a template dump. Keep the main flow tight, use references for detail, and stay anchored to the implemented UI rather than rebuilding or redesigning it.

平台分布

Codex

32.32%
按下载量换算63

Claude

30.84%
按下载量换算60

Cursor

18.74%
按下载量换算37

Gemini CLI

8.58%
按下载量换算17

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills