Token导航 LogoToken导航

Agent Skill · 研究检索

slides(slides 演示文稿)

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。

slides 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,权限为需要联网;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 68,015。

总安装 238周安装 10GitHub Stars 6下载 83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:slides(slides 演示文稿)
来源仓库:https://github.com/neethanwu/visuals
仓库路径:skills/slides
安装命令:
npx skills add https://github.com/neethanwu/visuals --skill slides
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/neethanwu/visuals --skill slides

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。

辅助界面设计、视觉规范、排版、配色、布局和交互体验优化。适用于整理页面结构、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。

slides 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/slides

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/neethanwu/visuals --skill slides;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

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

slides 怎么安装?

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

slides 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。

slides 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

slides 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。

slides 用的人多吗?

公开数据是累计安装 238 次,近一周安装 10 次,下载 83 次,GitHub 6 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。

slides 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。

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。

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
slides

slides 演示文稿 · 本页

2386需要联网未明确通过

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

Slides — Single-File HTML Presentation Builder

MANDATORY PREPARATION

Before building anything, confirm you have sufficient context: content (or topic), audience/tone, and whether brand assets are involved. If any of these are unclear, STOP and use AskUserQuestion to clarify before proceeding.

Read reference/template.md before building your first deck in a session.

Architecture

Zero dependencies. Each presentation is a single self-contained .html file using CSS scroll-snap — no framework.

  • Slides are 100dvh sections with scroll-snap-type: y mandatory
  • Typography scales via clamp(), layouts are natively responsive
  • Navigation, animations, fragments — agent-generated per deck, not a fixed engine
  • Libraries (Chart.js, GSAP, D3, etc.) added via CDN only when the content needs them

Context Assessment

Assess what context already exists. Do NOT follow a rigid step-by-step wizard. Figure out what you have, what you need, and fill the gaps.

Check for existing context

  1. Conversation history — has the user been working on content, research, or an outline?
  2. Working directory — are there markdown files, outlines, notes, or existing HTML slides?
  3. User arguments — did they pass a topic, file path, or content directly?

Determine what's missing

If you have...Then...
Full content + style preferenceBuild directly
Full content, no style preferenceInfer style from content, confirm briefly, build
Topic onlyHelp structure content, then infer style, build
Existing HTML slides to improveRead them, identify gaps, enhance
Nothing (bare /slides invocation)Ask what the presentation is about

Communication Philosophy

Be an opinionated collaborator with taste, not a menu system. Make the best choices you can, and communicate with the user in plain language when there's a meaningful fork in the road.

Always ask when needed:

  • Content gaps — you need the user's actual content, topic, or key points. Don't fabricate.
  • Brand context — if the presentation is for a company or product, ask whether they have brand colors, fonts, or a logo to incorporate. Don't ask for a personal/casual deck.
  • Speaker notes — for talk-style decks, ask if they want speaker notes embedded.
  • Meaningful style forks — when 2-3 styles would work equally well, present them in plain language the user can evaluate: "I see two directions for this:\n\n1. Warm editorial — cream tones, serif headlines, magazine feel\n2. Dark and bold — black background, neon accent, high-impact\n\nWhich feels right, or should I pick?"

Agent decides, then communicates what it chose:

  • Animation level — pick what fits the style and content. Mention it in the summary ("I used subtle fade-in animations to match the editorial feel").
  • GSAP vs CSS animations — use whichever produces the best result. The user doesn't need to know the implementation.
  • Background effects — add particle fields, shaders, or gradients when the style calls for it. Skip for restrained styles. Don't ask permission.
  • Font trio — read reference/font-guide.md, pick the best match for the style, load via Google Fonts. Mention the fonts in the summary.
  • Charts, tables, data viz — when content includes data, metrics, or comparisons, add appropriate visualizations. Don't ask "should I add a chart?" — just do it if the data warrants it.
  • Icons, diagrams, code blocks — include when content benefits. Don't ask.
  • Fragment reveals — add click-to-reveal for lecture/talk content where sequential builds help. Don't add for pitch decks or simple presentations.

Never ask about:

  • Hex colors, font names, animation easing, CSS values, or technical implementation
  • Which of the 37 styles to use — curate for them
  • Things you can infer from content (data-heavy → data-appropriate style)
  • Things with a clear single best answer
  • "Should I add GSAP?" "Should I include a progress bar?" — make the call

When using AskUserQuestion, keep options to 2-3 max. Describe in terms users care about (mood, feel, impression), not specs. Always include: "or should I just pick?"

Content Structuring

Slide types

  • Opening: Title + subtitle/tagline + optional author/date
  • Section divider: Bold section title + optional section number
  • Content: Heading + body (text, bullets, two-column, image+text)
  • Data: Heading + chart/table/metrics
  • Quote: Large quote + attribution
  • Comparison: Side-by-side columns
  • Code: Syntax-highlighted code block + annotation
  • Image: Full-bleed or framed image with caption
  • Closing: Final message + CTA/contact/links

Slide count

  • Lightning talk (5 min): 5-8 slides
  • Standard talk (15-20 min): 12-20 slides
  • Keynote (30-45 min): 25-40 slides
  • Pitch deck: 10-15 slides

One idea per slide. Dense slides lose audiences.

Style Selection

37 curated design styles live in reference/styles/. Read the chosen style file for its Color Palette, Typography, Design Principles, and Slide Application Notes. Apply with taste — they're guides, not rigid templates.

Interpreting style files — terminology glossary

Style files use inconsistent terminology. Normalize when reading:

You may see in a style fileTreat asCSS custom property
Background, Canvas, Outer FrameBackground color--bg-color
Surface, Card Surface, Elevated, PanelCard/container background--surface-color
Text Primary, Text Emphasis, Text Body (dark)Primary text color--text-primary
Text Secondary, Text Muted, Body TextSecondary text color--text-secondary
Accent, Accent Color, CTA, EmphasisPrimary accent--accent-color
Border, Rule, Divider, Grid LineBorder color--border-color
Positive, Success, GreenPositive semantic color--color-positive
Negative, Danger, RedNegative semantic color--color-negative
Display Font, Headline FontDisplay/heading typeface--display-font
Body Font, Text Font, UI FontBody typeface--body-font
Data Font, Mono Font, Code FontMonospace typeface--mono-font

Always map style file tokens to these CSS custom properties in your output.

How to choose

Infer from content. Do NOT present all 37 as a menu.

Content typeRecommended styles
Startup pitch / product launchp-bold-signal, p-minimal-corporate, p-dark-classy, p-swiss-clean
Financial / data-heavyp-editorial-financial, p-editorial-data, p-swiss-elegant
Technical / engineeringp-bauhaus-digital, p-swiss-clean-expressive, 08-neo-brutalism, blueprint
Creative / portfolio05-neo-expressionism, 16-contemporary-collage, p-monochrome-expressive, pop-art
Corporate / formalp-minimal-corporate, p-swiss-clean, p-editorial-calm
Academic / researchp-editorial-calm, p-swiss-elegant, 01-minimalist
Luxury / premium brandp-elegant-luxury, p-brutalist-luxury, p-dark-classy, apple-liquid-glass
Futuristic / tech-forward07-neo-futurism, p-bold-signal, p-bauhaus-digital, glitch-art
Editorial / storytellingp-editorial-warm, p-editorial-calm, p-industrial-humanist
Bold / high-impactp-bold-signal, p-nordic-brutalist-v2, p-brutalist-luxury-v2, 08-neo-brutalism
Warm / approachablep-editorial-warm, p-industrial-humanist, 01-minimalist
Elegant / refinedp-swiss-elegant, 12-art-deco, 06-neo-classicism
Retro / nostalgic / funsynthwave, pop-art, risograph
Print / analog aestheticrisograph, 16-contemporary-collage, blueprint
Disruptive / edgyglitch-art, 04-deconstructivist, 08-neo-brutalism
Modern / glass / premium techapple-liquid-glass, p-dark-classy, 07-neo-futurism

Strong single match → just use it. Multiple equally good → AskUserQuestion with 2-3 visual descriptions.

Font Selection

49 curated font trios (heading + body + mono) in reference/font-guide.md. Always read the font guide and cross-reference with the style — do not blindly hardcode fonts from the style file. Pick the trio that best fits the style's mood.

Load fonts via Google Fonts <link> tags. To convert a trio name to a Google Fonts URL, take the font family names and build: https://fonts.googleapis.com/css2?family=Font+Name:wght@400;500;600;700&display=swap

Style moodStrong trio matches
Editorial / serif authorityeditorial, gazette, novel, literary
Swiss / geometric precisionswiss, creative, modern-clean, minimal
Brutalist / raw powerbrutalist, poster, headline, dossier
Corporate / neutral claritycorporate, document, technical, product
Luxury / premiumclassic, folio, scholar, memoir
Warm / humaniststoryteller, journal, commons, health
Data / dashboarddashboard, fintech, devtool, technical
Bold / condensed impactheadline, impact, poster, brutalist
Retro / expressiveposter, headline, brutalist, creative

Reference Quick Lookup

All references live in reference/. Template and font guide are loaded in MANDATORY PREPARATION. Load the rest based on content needs:

ReferenceWhen to load
reference/animations.mdEntrance animations and transitions for the style
reference/effects.mdBackground effects — particles, shaders, scan lines, halftone, gradient mesh
reference/data-viz.mdData, metrics, charts, comparisons, tables
reference/media.mdImages, video, logos, screenshots
reference/elements.mdIcons, code blocks, diagrams, decorative SVGs, styled lists
reference/css-techniques.mdFrosted glass, text blending, outlined type, animated gradients, clip-path reveals

Key principle: Don't just react to what the user explicitly asks for. Consider what would make the deck better. If the outline has data, add charts. If the style calls for entrance animations, add them. The goal is the best possible deck, not the minimum viable one.

Building the HTML

Output

Single self-contained .html file. Styles in <style>, scripts in <script>, libraries via CDN only when needed.

Content sanitization: When interpolating user-provided text into slides, escape <script>, </style>, and raw HTML tags in body content to prevent accidental breakage of the document structure.

Structure

Follow the skeleton in reference/template.md:

  1. <head>: Google Fonts <link> → optional library CSS → <style> with structural CSS + agent-generated styles
  2. <body>: <div class="slide"> sections → optional progress bar / nav dots → CDN scripts → inline <script> with agent-generated JS

The structural CSS (~10 lines) is always present. Everything else — typography, layouts, colors, animations, navigation JS, fragment handling — is generated fresh for each deck based on style + content.

What to generate per deck

FeatureWhen to includeHow
Keyboard navigationAlwaysAgent generates JS for arrow keys, space bar
Touch/swipe supportMobile-friendly decksAgent adds touch event handlers
Entrance animationsMost decks (skip for ultra-minimal)IntersectionObserver + CSS transitions. See animations.md for style-specific patterns
Fragment revealsTalks/lectures with sequential builds.fragment + .visible class pattern. See template.md
Progress barLonger decks (10+ slides)Fixed bar at top, updates on scroll
Navigation dotsWhen style suits visible navPositioned buttons, highlight active
Background effectsWhen style calls for itCanvas/WebGL behind slides. See effects.md
GSAP animationsStaggered reveals, counters, complex sequencesLoad GSAP CDN, use with IntersectionObserver
ChartsData slidesChart.js or D3, render on slide visibility via IntersectionObserver
Code highlightingCode slidesHighlight.js CDN, call hljs.highlightAll() after DOM ready
DiagramsProcess/flow slidesMermaid CDN, init with startOnLoad: false, call mermaid.run() after DOM ready
IconsWhen content benefitsLucide CDN, call lucide.createIcons() after DOM ready
Speaker notesWhen user requestsHTML comments inside .slide divs

Gotchas

These are real issues to watch for:

  • Chart.js canvases need explicit dimensions. A <canvas> inside a slide collapses to 0x0 unless wrapped in a container with explicit dimensions or given width/height attributes.
  • Charts should render on visibility, not page load. Use IntersectionObserver to detect when a chart's slide enters the viewport, then create the chart. Otherwise charts render at 0x0 inside off-screen slides.
  • Three.js importmap must precede module scripts. Place <script type="importmap"> before any <script type="module">.
  • Background canvas z-index. Canvas for effects: position: fixed; z-index: 0. Slides need position: relative; z-index: 1.
  • Mermaid timing. Initialize with startOnLoad: false, call mermaid.run() after DOM is ready. Otherwise diagrams on non-first slides never render.
  • Lucide icons need explicit init. Call lucide.createIcons() after DOM is ready.
  • Fragment + navigation interaction. When fragments exist on a slide, keyboard navigation should reveal fragments first, then advance to the next slide when all fragments are visible. The agent must wire these together.
  • Google Fonts load order. Font <link> tags must appear before any <style> that references them to avoid flash of unstyled text.
  • Content overflow. Each slide has overflow: hidden. If content is too dense, it clips. Respect the "one idea per slide" rule. Use responsive clamp() sizing so content scales down on smaller screens.

On-Brand Customization

When users provide brand assets (logos, brand colors, brand fonts, style guides):

  1. Brand colors override the style's accent palette. Map the user's primary brand color to --accent-color. Map secondary brand colors to --color-positive or secondary accent roles. Keep the style's background and text colors unless the brand guide specifies otherwise.
  2. Brand fonts override the style's typography. If the user provides a brand font, use it for display. Fall back to the style's body font unless the brand specifies a body face too. Load brand fonts via Google Fonts if available, or ask the user for a hosted URL.
  3. Logos — see reference/media.md for placement patterns. Ask the user whether the logo should appear on every slide (persistent bottom-corner) or only on title/closing slides.
  4. Don't fight the brand. If the user's brand is warm and rounded but they picked a brutalist style, gently suggest a better-fitting style via AskUserQuestion rather than forcing a mismatch.

Design Quality & Polish

While building, ensure:

  • Typography hierarchy is clear (display > heading > body > caption)
  • Color palette applied consistently across all elements — charts, icons, tables, dividers
  • Spacing feels intentional — vary it (tight for grouped elements, generous for hero content)
  • Slide layouts vary within the style — don't make every slide look the same
  • Opening and closing slides are distinct and memorable
  • The overall deck feels cohesive — one design language throughout

Polish pass

The /polish, /animate, and /delight skills can elevate the design further — offer them to the user after the initial build.

For advanced GSAP usage beyond what's in reference/animations.md, the official GSAP skill provides deeper guidance on timelines, ScrollTrigger, and complex sequences.

Validation

After writing the HTML file, verify before telling the user it's done:

  1. Read the file back — scan for:

- Structural CSS present (scroll-snap, 100dvh, overflow hidden, reduced-motion) - All <div class="slide"> tags properly closed - CSS custom properties defined in :root and used consistently - No unstyled elements (bare <table>, <ul>, <canvas> without dimensions) - Font families in CSS match what's loaded in Google Fonts <link> - Navigation JS present (keyboard at minimum) - IntersectionObserver wired up if entrance animations exist

  1. Check for common breaks:

- Chart canvases have explicit width/height or a sized container - Mermaid uses startOnLoad: false - Three.js importmap appears before module script - Fragment handler integrates with navigation (reveals before advancing) - Lucide createIcons() called after DOM ready

  1. If issues found — fix them and re-read. Do not proceed until validation passes.
  2. Design polish pass (optional — offer to the user):

- "Want me to run a polish pass to refine the design?" - If yes, invoke /polish, /animate, /delight in sequence - If the user declines, skip.

  1. Tell the user:

- The file path - How to open it (open filename.html or browser) - What style, fonts, and key design choices you made - Offer refinements: "Want me to adjust colors, swap the style, add speaker notes, or rework any specific slide?"

What NOT to do

  • Don't ask users to pick hex colors, font names, or animation curves
  • Don't present all 37 styles as a menu
  • Don't add Three.js or shaders to a simple corporate deck
  • Don't ignore the style's do/don't rules
  • Don't generate placeholder content — if content is missing, ask for it
  • Don't over-animate — restraint is professional
  • Don't use default/unstyled charts — always theme to match
  • Don't embed large images as base64 — use URLs or file paths
  • Don't tell the user the file is ready before running validation
  • Don't override brand colors with the style's default accent — brand takes priority
  • Don't hardcode fonts from style files without checking the font guide
  • Don't skip data visualization when the content has data — charts and styled tables make decks better
  • Don't ask the user technical implementation questions they can't answer

平台分布

Codex

38.02%
按下载量换算32

Claude

29.59%
按下载量换算25

Cursor

16.39%
按下载量换算14

Gemini CLI

8.9%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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