2.4K ·GitHub
What is this?
A tool-agnostic recipe library of 21 verified hand-drawn art styles that turns your content description into a ready-to-copy image generation prompt, built as a Claude Code skill.
A tool-agnostic recipe library of 21 verified hand-drawn art styles that turns your content description into a ready-to-copy image generation prompt, built as a Claude Code skill.Is this a skill package or an image generator itself?It is a text-only prompt/skill package (PROTOCOL.md + STYLES.md plus a SKILL.md entry point) with no image backend; it only produces the prompt text or a structured JSON call, leaving actual image generation to a model like gpt-image or Midjourney.
It is a text-only prompt/skill package (PROTOCOL.md + STYLES.md plus a SKILL.md entry point) with no image backend; it only produces the prompt text or a structured JSON call, leaving actual image generation to a model like gpt-image or Midjourney.How do I install this skill?Download the lightweight hand-drawn-skill.zip release, unzip it, and drop the hand-drawn/ folder into ~/.claude/skills/ (or ~/.codex/skills/ for Codex); other tools point their rules file at AGENTS.md/PROTOCOL.md inside the folder.
Download the lightweight hand-drawn-skill.zip release, unzip it, and drop the hand-drawn/ folder into ~/.claude/skills/ (or ~/.codex/skills/ for Codex); other tools point their rules file at AGENTS.md/PROTOCOL.md inside the folder.Do I need anything installed to use it?Just Python 3.10 or later to run the renderer/build scripts, and the project has zero third-party Python dependencies.
Just Python 3.10 or later to run the renderer/build scripts, and the project has zero third-party Python dependencies.Any catches before relying on a specific style?Styles 3.1, 19 and 20 require official reference anchor images and a multi-stage, scored validation workflow before output can be labeled 'final'; if the calling tool can't support anchors, high-fidelity references, or snapshot locking, production generation fails closed.
Styles 3.1, 19 and 20 require official reference anchor images and a multi-stage, scored validation workflow before output can be labeled 'final'; if the calling tool can't support anchors, high-fidelity references, or snapshot locking, production generation fails closed.Which AI agents can use this skill?Any agent that can read custom instructions, explicitly including Claude Code, Codex, Cursor, Gemini CLI, Jules, Cline, Windsurf, and Continue.
Any agent that can read custom instructions, explicitly including Claude Code, Codex, Cursor, Gemini CLI, Jules, Cline, Windsurf, and Continue.How many drawing styles are included?21 verified recipes in total (20 numbered styles plus a 3.1 variant), grouped into 8 categories like crayon doodle, Ghibli, ink wash, retro pixel, and Nordic paper craft.
21 verified recipes in total (20 numbered styles plus a 3.1 variant), grouped into 8 categories like crayon doodle, Ghibli, ink wash, retro pixel, and Nordic paper craft.How do I pick a specific style?Call it by its number, Chinese name, or a registered English alias such as 'ghibli' or 'pixel-art'; placeholders like subject, title, or color scheme are automatically inferred from your description.
Call it by its number, Chinese name, or a registered English alias such as 'ghibli' or 'pixel-art'; placeholders like subject, title, or color scheme are automatically inferred from your description.Can I add my own style recipe?Yes, contributors can add a new entry to STYLES.md and register it in PROTOCOL.md's menu and alias table, following the project's placeholder and sample-image validation guidelines.
Yes, contributors can add a new entry to STYLES.md and register it in PROTOCOL.md's menu and alias table, following the project's placeholder and sample-image validation guidelines.Is there a CLI script I can run directly?Yes, scripts/renderprompt.py extracts a recipe exactly as written (avoiding agent paraphrasing); running it with --list shows all recipes' parameters and reference images.
Yes, scripts/renderprompt.py extracts a recipe exactly as written (avoiding agent paraphrasing); running it with --list shows all recipes' parameters and reference images.What are similar projects?photo-abstract-editorial, ai-aug-edu, deep-research-skill, SwiftUI-Agent-Skill, brag, kid-papercraft, prompt-master, seedance2-skill
photo-abstract-editorial, ai-aug-edu, deep-research-skill, SwiftUI-Agent-Skill, brag, kid-papercraft, prompt-master, seedance2-skillWhich topics is it in?cline, art, prompt, gemini-cli, claude-code-skill, claude-skills, image-generation, prompt-engineering, skill, cursor, generative-ai, china
cline, art, prompt, gemini-cli, claude-code-skill, claude-skills, image-generation, prompt-engineering, skill, cursor, generative-ai, china