Token导航 LogoToken导航TokenDH.com
研究检索external-servicegithub未标认证来源可访问许可证需确认审计通过

pencil-design铅笔设计

Agent Skill

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。

总安装

212

周安装

9

GitHub Stars

公开资料未说明

下载量

74
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

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:pencil-design(铅笔设计)
来源仓库:https://github.com/joneshong-skills/pencil-design
仓库路径:skills/pencil-design
安装命令:
npx skills add https://github.com/joneshong-skills/pencil-design --skill pencil-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/joneshong-skills/pencil-design --skill pencil-design

简介

pencil-design 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,适合生成 UI 方案和改进组件层级。

  • 适用于需要页面结构整理、视觉一致性检查或设计系统维护的场景。
  • 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。
  • 该技能归类为研究检索工具,主要用于界面设计和用户体验优化。

SKILL.md

Pencil Design

Expert design assistant for.pen files via Pencil MCP tools. Handles the full workflow: ideation, style guide, component reuse, section-by-section building, visual verification, and design-to-code export.

Prerequisites

  • Pencil Desktop App or VS Code/Cursor extension must be running BEFORE Claude Code starts
  • MCP tools are deferred — load via ToolSearch(query: "+pencil") before first use
  • .pen files are JSON-based, Git-friendly, no auto-save (remind user: Cmd+S)

Agent Delegation

This skill runs in main context — MCP tool calls require ToolSearch which only main context and general-purpose agents have. For large multi-screen designs, delegate individual screens to general-purpose sub-agents with explicit instructions.

Design Workflow

Standard 10-Step Flow

1. ToolSearch("+pencil")           ← Load MCP tools (once per session)
2. get_editor_state                ← Current file, selection, context
3. batch_get(reusable: true)       ← Discover existing components
4. get_variables                   ← Read design tokens
5. get_guidelines(topic)           ← Load rules for task type
6. get_style_guide_tags            ← Browse style options (optional)
   get_style_guide(tags)           ← Get visual direction (optional)
7. find_empty_space_on_canvas      ← Find placement space
8. batch_design                    ← Build section by section
9. get_screenshot                  ← Visual verify EACH section
10. snapshot_layout(problemsOnly)  ← Detect overlaps/clipping

Repeat steps 8-10 for each section. Never build an entire page in one batch_design call.

Topic Selection for get_guidelines

TaskTopic
Landing page / marketinglanding-page
Web application / SaaSweb-app
Mobile app screenmobile-app
Presentation slidesslides
Data table / dashboardtable
Using a design systemdesign-system
Generating code from designcode
Tailwind CSS implementationtailwind

Load multiple topics when relevant (e.g., web-app + design-system).

batch_design Operations

Seven operations, max 25 per call. Operations execute sequentially; any failure rolls back all.

OpSyntaxPurpose
Insertid=I(parent, {type,...})New node. Never set id manually.
Copyid=C(path, parent, {descendants, positionDirection, positionPadding})Clone node. Use descendants to modify children.
UpdateU(path, {props})Modify properties. Cannot change id, type, or ref.
Replaceid=R(path, {nodeData})Replace node (component instance slot swaps).
MoveM(nodeId, parent, index?)Reposition in tree.
DeleteD(nodeId)Remove node.
Image`G(nodeId, "ai"\"stock", prompt)`Fill frame/rectangle with generated or stock image.

Binding Rules

  • Every I/C/R must have a binding name: myCard=I(...) not just I(...)
  • document is the predefined root binding
  • Bindings are valid only within a single batch_design call
  • Use + for path navigation: U(card+"/titleText", {content: "New"})

Copy Gotcha

Copy gives all descendants new IDs. Do NOT U() on copied children — use the descendants parameter in the Copy call itself to modify children during copy.

Critical Rules

  1. Reuse components — Always batch_get(reusable: true) first, insert via ref
  2. Use variables — Reference token names, not hex values
  3. Prevent text overflow — Set wrap, constrain width, use fill_container
  4. Verify every sectionget_screenshot + snapshot_layout(problemsOnly: true)
  5. No image node type — Images are fills on frame/rectangle via G()
  6. Semantic namingUserAvatarImage not Frame 42
  7. 25 ops max per batch_design — split larger builds across calls
  8. Section-by-section — Build iteratively: plan → build → verify → fix → next

Design-to-Code Export

  1. get_guidelines("code") + get_guidelines("tailwind")
  2. get_variables → map tokens to Tailwind @theme directives
  3. batch_get → read full design tree
  4. Map Pencil components → shadcn/ui (or target library)
  5. Generate semantic classes: bg-primary not bg-[#3b82f6]
  6. Use CVA for variants, cn() for class merging, Lucide for icons

Style Systems

4 built-in design systems: Shadcn UI, Halo, Lunaris, Nitro. Select one early — reduces token cost by referencing known patterns.

Quick Reference

Keyboard Shortcuts

ActionShortcut
AI prompt panelCmd+K
Deep select elementCmd+Click
Select parentShift+Enter
Make reusable componentCmd+Alt+K
GroupCmd+G
Create frameCmd+Alt+G
DuplicateCmd+D
SaveCmd+S
Zoom to fitCmd+0

Figma Import

Ctrl+C in Figma → Ctrl+V in Pencil. Images don't transfer (re-add manually). Complex designs: paste in sections to avoid clipboard limits.

CLI (Desktop App Only)

pencil --agent-config config.json

Config is a JSON array — each entry opens a separate Pencil window running its own agent:

[
  {"file": "./screen1.pen", "prompt": "Design a dashboard", "model": "claude-4.5-sonnet"},
  {"file": "./screen2.pen", "prompt": "Design a login page", "model": "claude-4.5-haiku"},
  {"file": "./screen3.pen", "prompt": "Design settings", "model": "claude-4.5-haiku"}
]

Multi-agent parallel design:

  • Each entry = independent window + independent Claude API call
  • No documented concurrency limit (tested up to 6)
  • .pen files must be pre-created (CLI cannot create new files)
  • Use haiku for simple screens, sonnet/opus for complex ones
  • Bottleneck: RAM per window + API rate limit, not Pencil itself
  • Headless CLI (no GUI) coming soon — better for large-scale parallel runs

Pre-create empty files: for i in {1..6}; do echo '{}' > "screen${i}.pen"; done

Token Optimization

  • Select a design system upfront (AI references known patterns)
  • Hide/delete irrelevant canvas frames before starting
  • Use low readDepth / searchDepth values in batch_get
  • Build section by section, not entire page at once

Integration

  • frontend-design — Load for unique visual identity (avoid generic AI aesthetic)
  • brand-guidelines — Apply brand tokens to.pen designs
  • canvas-design — Different tool: PNG/PDF art, not.pen files

Additional Resources

Reference Files

  • references/batch-design-examples.md — Common batch_design patterns and recipes
  • references/troubleshooting.md — Known issues, gotchas, and workarounds

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

33.93%
按下载量换算25

Claude

30.81%
按下载量换算23

Cursor

19.88%
按下载量换算15

Gemini CLI

9.69%
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

external-service

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

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。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills