DesiSketch-Illustrations-AI-Hand-Drawn-Diagram-Generator-Agent-Skill
Translate judgments, workflows, system states, and metaphors in articles into clean, hand-drawn, absurd, and minimalist body illustrations Skill
Install / Use
npx skills add GunaTeja777/DesiSketch-Illustrations-AI-Hand-Drawn-Diagram-Generator-Agent-SkillInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
AutomationSupported Platforms
Tags
Skill content
View source on GitHubDesiSketch Illustrations (AI Hand-Drawn Diagram Generator & Agent Skill)
Translate text, workflows, and system metaphors into clean, wobbly, hand-drawn diagrams using a consistent visual grammar. Compatible with Claude Code, Codex, Cursor, Windsurf, and Claude Projects.
16:9 Horizontal | Desi IP | Pure White Hand-Drawn | Latin-Script Annotations (English, Hinglish, Tenglish, Kanglish) | AI Illustration Agent
What is this Repository?
DesiSketch Illustrations is a cross-platform AI Agent Skill designed to guide AI coding assistants and chat agents (such as Claude Code, Codex, Cursor, Windsurf, or Claude Projects) in designing and generating inline illustrations for articles, blog posts, Notion documents, and technical workflows written in English or transliterated Indian languages (such as Hinglish, Tenglish, Kanglish, etc.).
Unlike ad-hoc DALL-E prompts or generic diagram tools, this skill enforces a consistent, reproducible visual grammar—combining a wobbly pen-sketch aesthetic with a deadpan operator character ("Desi") to make abstract concepts instantly readable and memorable. It is not a generic clipart generator or a polished slide template. Its core mission is to analyze the cognitive anchors of your text and translate one key judgment, workflow, comparison, state, or metaphor into a memorable 16:9 hand-drawn diagram.
The default character is "Desi": a solid black creature with white dot eyes, thin legs, and a deadpan expression. Desi is not a decorative mascot, sticker, or corner-filler—Desi is an active, deadpan operator participating in the mechanics of the system.
In short: We guide the AI to draw out the key conceptual action of the text, rather than just "adding a pretty picture."
Compatibility
This skill has been tested and verified across multiple AI assistant/agent interfaces:
- Codex / Antigravity IDE: Fully supported, including automated visual feedback and built-in image tools.
- Claude Code (CLI): Fully supported, using custom local skill directories and
/desisketch-illustrationsinvocation. - Cursor / Windsurf: Verified using
.cursorrulesconfig. - Claude.ai Projects: Verified by pasting
SKILL.mdinto Custom Instructions and uploading reference files.
Who is this for?
Perfect for:
- Content creators writing technical blogs, Notion documentation, or newsletters.
- Creators explaining AI workflows, product design, or systems.
- People wanting to translate abstract concepts into clean, physical metaphors.
- Those who want a lighter, weirder, and more recognizable visual language than standard slide deck diagrams.
- AI team workforces (agents) that need a standardized, reproducible visual layout.
Not suitable for:
- Polished commercial marketing assets or corporate key visuals.
- Complex architectural block diagrams, detailed flowcharts, or full UML schematics.
- Cute, cartoonish, or babyish illustrations (emojis/stickers).
- Cramming long paragraphs, massive bullet lists, or entire course pages into a single image.
- Editable vector source files (SVG/Illustrator) that require strict node-by-node editing.
What does it output?
Default Outputs:
- 16:9 horizontal article illustrations.
- A strategic shot list for your article (4–8 illustrations).
- The theme, core concept, structure type, Desi's action, and suggested Latin-script annotations for each drawing.
- PNG image assets saved to your workspace's
assets/<article-slug>-illustrations/directory.
Does NOT output:
- Editable slides (.pptx, .key) or document PDFs.
- Editable vector layers (SVG/HTML/Canvas).
- Dense text-based informational posters.
Visual Style
This skill uses the "absurd wobbly line product sketch" style:
- Solid White Background: No paper textures, warm tints, gradients, drop shadows, or background noise.
- Black Hand-Drawn Outlines: Thin, slightly wobbly pen lines representing a raw, manual sketch.
- Negative Space: The subject occupies 40%–60% of the canvas; at least 35% must remain empty.
- Sparse Color Accents: Black for outlines/labels, Orange for paths/flows, Red for critical points/warnings, Blue for supplementary notes/mental states.
- Latin Script Only: Minimal annotations (5-8 labels, 2-8 words each) written in the Latin alphabet (English writing) to ensure high rendering accuracy in AI generators. Mixed-language transliterated words (e.g., Hinglish: "Kaam Shuru", Tenglish: "Pani Shuru", Kanglish: "Kelsa Shuru", English: "Process Output") are highly encouraged.
- Desi IP: Must be the primary driver of the physical action (operating levers, carrying boxes, filtering files, etc.).
Example Concepts (Calibration Samples)
Below are the available style-calibration drawings located in examples/images/:
Multilingual Sorting (4 Languages)

Square Desi (1:1)

RAG System

LLM Process

Docker Pipeline

Binary Tree (Hinglish Concept)

Merge Sort (Tenglish Concept)

Note: These are visual style calibrations (for stroke width, line jitter, and color density), not rigid layout templates. For each new article, the AI will invent a new metaphor.
Installation
Clone the repository:
git clone https://github.com/GunaTeja777/desisketch-illustrations.git
cd desisketch-illustrations
Prerequisites (Image Generation Dependency)
This skill designs the visual strategy, plans the shot list, and generates image prompt guidelines, but it requires your AI agent to have image-generation capabilities to produce final PNGs:
- Claude Code (CLI): Requires an MCP (Model Context Protocol) server or custom tool for image generation (e.g., connected to Midjourney, DALL-E, or Stable Diffusion APIs).
- Codex / Antigravity IDE: Works out of the box using built-in image tools.
- Cursor / Windsurf / Claude.ai Projects: Works when using a model or workspace configuration that has access to image generation features/extensions.
Choose the setup method depending on your environment:
A. For Cursor & Windsurf IDEs
A .cursorrules file is provided at the root of this repository.
- Copy the
.cursorrulesfile (or append its contents) into the root of your workspace:cp .cursorrules /path/to/your/workspace/ - When you use the Cursor chat or composer, Claude/Gemini will automatically recognize the
DesiSketchcommands and follow the rules.
B. For Claude Projects (Claude.ai Web UI)
If you are using Claude Projects in the web interface:
- Create or open a Project in Claude.ai.
- Click Add Content and upload the files inside the
desisketch-illustrations/references/folder as project files. - Copy the contents of desisketch-illustrations/SKILL.md and paste it into the Custom Instructions section of your Project.
C. For Claude Code (CLI)
Copy the skill to your Claude custom skills directory:
- Global Installation (available in any project/workspace):
mkdir -p ~/.claude/skills cp -R ./desisketch-illustrations ~/.claude/skills/ - Project-level Installation (only active inside this specific workspace):
mkdir -p .claude/skills cp -R ./desisketch-illustrations .claude/skills/
D. For Codex & OpenAI Agent Frameworks
Copy the skill to your agent configuration workspace's custom directory:
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
cp -R ./desisketch-illustrations "${CODEX_HOME:-$HOME/.codex}/skills/"
How to Use
Depending on your agent platform, trigger the skill using its identifier:
- Claude Code: Use the
/desisketch-illustrationscommand or let Claude trigger it automatically. - Codex & Other Agents: Reference the skill as
$desisketch-illustrations.
1. Illustration Planning (Shot List Strategy)
[Claude Code] /desisketch-illustrations
[Codex/Other] Use $desisketch-illustrations
Do not generate images yet.
Analyze the article below, identify where visual explanations are needed, and output a shot list of around 5 images.
For each image, specify:
- Placement paragraph
- Theme
- Core metaphor
- Structure type
- Desi's action
- Suggested elements
- Recommended transliterated annotation labels (in Hinglish / Tenglish / Kanglish / English)
<Paste article text here>
2. Full Image Generation
[Claude Code] /desisketch-illustrations
[Codex/Other] Use $desisketch-illustrations
Generate 4 illustrations for this article.
Requirements: 16:9 horizontal, pure white background, black hand-drawn outline, sparse red/orange/blue Latin-script annotations (e.g., Hinglish or Tenglish transliterated labels).
Do not create complex slides or cute mascot posters.
<Paste article text here>
3. Visualizing a Single Statement/Concept
[Claude Code] /desisketch-illustrations
[Codex/Other] Use $desisketch-illustrations
Generate a 16:9 illustration for this concept:
"Trust isn't built by shouting; it is laid down tile by tile with concrete proof."
Make it bizarre but clean. Desi must carry out the physical action. Use 3-5 short Hinglish transliterated labels.
License
MIT License. See LICENSE.
Related Skills
Agent-Reach
84.5kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
73.0k🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
nanobot
48.5kUltra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps
Scrapling
82.9k🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl! Don't be shy, join here: https://discord.gg/EMgGbDceNQ and follow here for daily tips and tricks: https://x.com/Scrapling_dev
Languages
Security Score
Audited on Jul 11, 2026
