Token导航 LogoToken导航

Agent Skill · 前端设计

prototype(prototype 搜索)

prototype 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要围绕仓库状态、代码变更或协作事项进行整理时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

prototype 是前端设计类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为操作浏览器;许可证信息还不明确,第三方扫描有未通过项,按安装量在前端设计类的 10,216 个 Skill 里排第 6,423。

总安装 512周安装 22GitHub Stars 55下载 180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:prototype(prototype 搜索)
来源仓库:https://github.com/adobe/skills
仓库路径:skills/prototype
安装命令:
npx skills add https://github.com/adobe/skills --skill prototype
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/adobe/skills --skill prototype

这个 Skill 是做什么的

前端设计类 Skill 关注页面结构、视觉规范和组件写法,输出通常是可运行的界面代码。

prototype 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要围绕仓库状态、代码变更或协作事项进行整理时使用。适用于前端设计场景,可结合来源仓库、安装命令和原始 README 继续核验具体用法。

通过 npx skills add 命令从指定 GitHub 仓库安装,支持在主流宿主环境中使用。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。prototype 属于前端设计类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/prototype

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/adobe/skills --skill prototype;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

    它需要操作浏览器,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

prototype 怎么安装?

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

prototype 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。

prototype 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

prototype 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。

prototype 用的人多吗?

公开数据是累计安装 512 次,近一周安装 22 次,下载 180 次,GitHub 55 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。

prototype 安全吗?

第三方扫描有未通过项,风险较高。除非你确认过源码,否则不建议直接安装到常用环境。

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。

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。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 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。
  • 需要处理前端设计类任务、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。

前端设计类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
frontend-design

前端设计 · Codex、Claude

8,823,360126,072只读MIT通过
vercel-composition-patterns

Vercel 组合模式 · Codex、Claude

3,858,81625,893需要联网MIT通过
vercel-react-native-skills

Vercel React native skills 前端 · Codex、Claude

2,468,66425,878只读MIT通过
adapt

adapt 搜索 · Codex、Claude

2,027,04023,125操作浏览器未明确通过
bolder

bolder 分析 · Codex、Claude

2,004,28823,130只读未明确通过
prototype

prototype 搜索 · 本页

51255操作浏览器未明确未通过

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

Prototype

Design in the browser. Produce branded, high-fidelity HTML for each page — real fonts, real colors, real copy — and iterate on it directly with the user until they approve. The output is self-contained static HTML.

This is the stage where visual design decisions happen: type scale, spacing, proportions, button sizing, visual weight, section rhythm.

When to use this skill

  • The user is ready to design visuals — type scale, spacing, proportions, layout, visual weight.
  • The user asks to change, refine, review, critique, or iterate on visual styling or any file under stardust/prototypes/**/*.html.
  • The user types /stardust:prototype, or asks to "design the page" after briefings are ready.

Do NOT use this skill

  • For page copy — copy is owned by briefings. Propose a writeback to briefings/{page}.md if the user wants to save a new headline.
  • For brand voice or identity decisions (colors, typography system, tone). Hand off to brand.
  • For grey structural wireframes. Hand off to wireframes.
  • To build EDS blocks or production HTML. Prototype stops at static HTML — EDS implementation is a separate downstream effort.

Pre-flight

Run the procedure in ../_shared/preflight.md first.

Contract

Needs (reads if present):

  • stardust/brand-profile.json
  • stardust/briefings/{page}.md (including optional # Copy and # Imagery)
  • stardust/wireframes/{page}.html
  • .impeccable.md

Produces:

  • stardust/prototypes/{page}.html (self-contained, branded, desktop fidelity)

If missing:

  • No brand-profile.json → synthesize a neutral brand shape (system-ui fonts, mono palette, straight voice). Stamp provenance in the design's <head>.
  • No briefing → prompt the user for a one-line page intent; synthesize a minimal briefing (in memory only, unless the user says "save it"); stamp provenance.
  • Briefing has no # Copy → generate on-brand copy following brand-profile.json voice rules and .impeccable.md. Stamp provenance per section.
  • No wireframe → shape structure from the briefing directly.
  • No .impeccable.md → use brand-profile defaults only.

Copy Ownership

The briefing is the source of truth for copy.

  • If a section has # Copy in the briefing, use those strings verbatim. Never rewrite.
  • If a section has no # Copy, generate on-brand copy and record provenance (per-section, in the <head> provenance block — e.g. hero.headline: synthesized).
  • Never auto-write generated copy back to the briefing. If the user asks ("also save this to the briefing") perform a single, targeted writeback and report what changed.

Phase 0: Setup

Two setup questions before rendering begins. Ask both once per session (not per page). When the skill is re-invoked on an existing project, read answers from the last prototype's provenance block if present; re-ask only if any answer is missing.

0a · Variant count

Ask the designer to confirm, with 2 as the default:

"I'll produce 2 variants of each prototype so you can compare — confirm, or give a different number (1–4)?"

Wait for the designer's answer. Accept:

  • "ok", "confirm", "yes", Enter, "2" → 2 variants (default)
  • 1 → single variant
  • 3 or 4 → that many variants
  • Any other input → re-ask once with the valid options

Default to 2 when:

  • The designer skips or accepts the default
  • The skill is invoked as part of an automated pipeline with no explicit variant count
  • 1 variant — produce stardust/prototypes/{page}.html.
  • N ≥ 2 variants — the skill produces {page}-a.html, {page}-b.html, … {page}-{letter}.html, each exploring a distinct design direction. Pick directions along axes that are load-bearing for *this* brand (e.g. if the brand voice is unsettled, vary type-voice; if the palette is rich, vary color-energy). Canonical axes to pick from:

- type-voice — editorial serif-forward ↔ software sans-forward - density — airy/spacious ↔ catalog/compressed - color-energy — restrained/ink-led ↔ saturated/surface-led - imagery-role — decorative/background ↔ content/lead

  • Record the chosen direction per file as a one-line comment inside the prototype's provenance block (e.g. variant_direction: "editorial, Fraunces-forward, quiet").

0b · Imagery mode

Ask: "For images in the prototype, do you want: (a) branded placeholder rectangles — fast, offline, free, or (b) real generated images? If (b), you'll need to pick a model and point me at a credential."

Default to (a) placeholders when the user skips, the skill runs in full-pipeline auto-approve mode, or no credential is available.

If the user picks (b) generated, ask two follow-ups:

  1. "Which image model?" — the assistant presents options based on what's installed. Currently known providers: Answer Provider Model Credential env var gemini Google Gemini 3 Pro Image Preview GOOGLE_API_KEY flux fal.ai FLUX Schnell FAL_KEY imagen Google Vertex Imagen 3 GOOGLE_APPLICATION_CREDENTIALS dalle OpenAI DALL-E 3 OPENAI_API_KEY
  2. "Where's the credential?" — accept any of:

- env → the credential is already exported as the env var from the table above - A file path ending in .env → the skill reads the env var from that file (do NOT echo the key to the designer; just read it in memory) - A raw token → the skill treats the string as the key value and uses it only for this session (do not write it to disk)

Record the imagery mode in each prototype's provenance block:

<!-- stardust:provenance
  imagery_mode: placeholder
  ...
-->

or, when generation was used:

<!-- stardust:provenance
  imagery_mode: generated
  imagery_provider: gemini
  imagery_credential_source: /Users/paolo/excat/az-sitebuilder/.env
  imagery_generated_at: 2026-04-24T15:12:00Z
  images: [
    { section: "hero", path: "images/landing-hero.png", prompt_hash: "abc123" },
    ...
  ]
-->

Never log the actual key value anywhere — not in provenance, not in terminal output, not in error messages.

Imagery-mode consequences for Phase 2

  • placeholder → Phase 2 renders branded placeholder rectangles per design-guide.md *Imagery* section, option (b).
  • generated → Phase 2 invokes the ai-image-generator skill (from eds-site-builder, sumi, or testing plugin — whichever is registered) with the chosen provider + credential, passing the briefing's # Imagery direction + brand photography rules per section. Generated images land at stardust/prototypes/images/{page}-{section}.png. Prototype HTML <img src> points at those relative paths.

Fallback behaviour when imagery_mode = generated but the generation fails:

  1. Retry once with a slightly simplified prompt
  2. On second failure, fall back to a branded placeholder for that specific image slot (not the whole page)
  3. Stamp imagery_fallback: true and the failed-slot list in the provenance block so the designer knows which slots to retry later
  4. Continue rendering the prototype — do not abort the whole page over one failed image

Phase 1: Plan

For each briefing:

  1. Load structural input:

- If stardust/wireframes/{page}.html exists, use it as the section order and layout reference. - If not, plan the sections from the briefing: - If the impeccable plugin is installed (/impeccable shape is registered), delegate to /impeccable shape. - Otherwise, use the "For wireframe section planning" pattern in ../_shared/fallback-brainstorm.md. Per ../_shared/soft-deps.md, impeccable fallback runs silently.

  1. For each section, check the briefing's # Copy:

- Present → use verbatim. Do not rewrite under any feedback loop unless the user explicitly says to change the words in the briefing. - Absent → generate on-brand copy; add the slot to the design's provenance block (e.g. hero.headline: synthesized). - Never write generated copy back to the briefing automatically. Offer: "Want me to also save these lines to the briefing?" after the first render — act only on explicit confirmation.

  1. Re-read the briefing's # Imagery section — follow source hints; otherwise generate branded placeholders.

Phase 2: Render (Branded Mode)

Render each design as a self-contained HTML file at desktop fidelity (1440px design target):

  • Brand fonts via @import or <link> to web fonts.
  • Brand colors for surfaces, text, CTAs, and accents per brand-profile.json color roles. Derive the page ground from the brand palette — do NOT default to cream (or any rebrand: vellum, kami, bone, ivory, eggshell). See ../_shared/divergence-toolkit.md § 2.5 Ground-color seed.
  • Real component styles — button patterns, border-radius, motifs from the brand profile.
  • Real copy — briefing # Copy verbatim, or on-brand generated copy.
  • Images — briefing # Imagery sources, or branded placeholders.
  • CSS custom properties in :root that expose type scale, spacing, max-width, section padding, button proportions. These values are the authoritative desktop tokens — any downstream translation (to EDS CSS, another framework, a design handoff) reads them from here.
  • Preserve data-section, data-intent, data-layout attributes from the wireframe so downstream stages can read structure.
  • Provenance block — if any input was synthesized (brand, briefing, wireframe, or any # Copy slot), include a <!-- stardust:provenance... --> comment as the first child of <head> per ../_shared/skill-contract.md. List each synthesized input and, for copy, each synthesized slot.

Write to stardust/prototypes/{page}.html (single-variant mode) or stardust/prototypes/{page}-{letter}.html for each variant (multi-variant mode). In multi-variant mode, each file is rendered independently — variants share the briefing's copy and the brand tokens, but differ on the design-direction axes chosen in Phase 0.

Follow the rendering rules in design-guide.md.

Phase 3: Serve

Prototypes are self-contained HTML files. Open each file in the designer's default browser immediately after writing per ../_shared/skill-contract.md *Opening HTML artifacts*. Do not require the designer to open the files manually.

  • Single variant: open "stardust/prototypes/<page>.html" on macOS (xdg-open on Linux). Tell the user: "Your prototype is open."
  • Multiple variants: open each variant file at the same time so the designer can flip between browser tabs. Then present the comparison table: Variant Direction Key visual move Risk a editorial, Fraunces-forward oversized serif headline, narrow measure may read as too quiet for a tech audience b software-catalog, Inter-dense table-led hero, hard geometric rhythm may read as cold without warm imagery............ Ask the user: "Which variant should we take forward?" Wait for an explicit answer (a, b,...) before moving on.

If the designer needs a real HTTP origin (fetch, service workers, absolute asset URLs), additionally run python3 -m http.server 8000 --directory stardust/prototypes in the background and tell them the http://localhost:8000/<page>.html URL — but keep the browser-open via open first, which handles 95% of cases via the file:// protocol.

In pipeline-automation mode (no designer present), skip the open.

Phase 4: Iterate in the Browser

This is a design loop, not a one-shot render. Expect multiple rounds.

In multi-variant mode, iterate only on the variant the user named (e.g. landing-b.html). Unchosen variants stay on disk as a design-history reference; do not delete or rename them without the user's explicit ask. Every subsequent iteration request refers implicitly to the chosen variant until the user picks a different one.

Common feedback and how to handle it:

  • "Headlines are too big/small" → Adjust --heading-* custom properties, re-render, refresh.
  • "Section feels cramped" → Adjust --section-padding or per-section spacing, re-render.
  • "Button needs more weight" → Adjust button padding/font-size/weight in the design CSS.
  • "Try a different accent color" → Swap the CSS variable value; do not edit brand-profile.json unless the user wants to make it the new brand default.
  • "This section should feel quieter" → Adjust typographic weight or surface color on that specific section.

Every iteration updates the file under review — stardust/prototypes/{page}.html in single-variant mode, or stardust/prototypes/{page}-{letter}.html for the variant the user named in multi-variant mode. Re-open the file in the browser after every iteration so the designer sees the change immediately (browsers with live-reload will pick up the change on re-open; otherwise the user may need to refresh). Keep iterating until they approve.

Before asking the user to look at a new iteration, run a critique:

If the impeccable plugin is installed (/impeccable critique is registered), use it. Otherwise, apply the rubric in ../_shared/fallback-critique.md. Per ../_shared/soft-deps.md, impeccable fallback runs silently.

Phase 5: Approval Gate

Hard gate in interactive mode — do not proceed until the user approves each prototype.

Pipeline automation: When invoked as part of a full pipeline run, auto-approve after two critique passes and continue.

When the user approves:

  1. Confirm all prototypes are saved under stardust/prototypes/*.html.
  2. Tell the user: "Prototypes approved. Run /stardust to see your next step."

Why Prototypes Are Static HTML

Prototypes are platform-agnostic — they could in principle feed any downstream build later (EDS, a different SSG, a different CMS). They are pure HTML + CSS judgment calls.

Keeping prototypes static and platform-free means:

  • Visual design can be iterated freely without implementation complications.
  • Any downstream translation (to EDS CSS, to a framework component library, etc.) is mechanical, not judgmental.
  • The prototype layer is portable across target platforms.

Artifacts Written

FileDescription
stardust/prototypes/{page}.htmlBranded, high-fidelity static HTML per page — self-contained, platform-independent. Single-variant mode.
stardust/prototypes/{page}-{letter}.htmlOne file per variant (-a, -b, -c, -d) when the user asked for 2–4 variants. Each carries a variant_direction line in its provenance block. Unchosen variants stay on disk as design history.

平台分布

Codex

34.08%
按下载量换算61

Claude

29.06%
按下载量换算52

Cursor

19.34%
按下载量换算35

Gemini CLI

9.73%
按下载量换算18

安全审计

Gen Agent Trust Hub

未通过

Socket

通过

Snyk

通过

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。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