A Claude Code configuration for designers and vibe-coders. Includes 7 specialized agents, 6 rule files, 3 custom commands, a curated library of design skills, and a structured workflow pipeline that turns Claude Code into your design & development team.
For: Designers who build their own interfaces, vibe-coders, and anyone who wants to go from idea to working product without deep coding knowledge.
Stack: Any. React, Vue, Svelte, vanilla HTML — this config adapts to your project.
- Download the kit — Code → Download ZIP — and unzip it.
- Rename the folder to your project's name (e.g.
photo-portfolio). - Open that folder in Claude Code (
claude ., or open it in the desktop app).
That's it. On first launch the kit installs its design skills automatically, in
the background — you don't have to run anything. Claude reads CLAUDE.md, loads
the rules, and has all agents ready. Just describe what you want to build.
Prefer git?
git clone https://github.com/Dishain/shain-designer-kit.git my-projectthen open the folder. (Manual install is never required, but./install.shis there as a fallback.)
# Grab the kit
git clone https://github.com/Dishain/shain-designer-kit.git /tmp/cdk
# Copy config files into your project
cp -r /tmp/cdk/.claude /path/to/your-project/
cp /tmp/cdk/CLAUDE.md /path/to/your-project/
cp /tmp/cdk/.mcp.json /path/to/your-project/
cp -r /tmp/cdk/docs /path/to/your-project/
cp /tmp/cdk/install.sh /tmp/cdk/update.sh /path/to/your-project/Open your project in Claude Code — the design skills install automatically on the first session.
- Claude reads
CLAUDE.mdand loads all rules automatically - You describe what you want: "Make me a landing page for my photography portfolio"
- Claude picks the right agents for the job (BA → UX → UI → Developer → Tester)
- Progress is tracked in
docs/todo.md - When you correct Claude, it saves the lesson to
docs/lessons.md(project memory) - Next session, Claude reads
lessons.mdfirst — it won't repeat the same mistakes
The kit includes a self-improvement system:
docs/todo.md— Claude writes the task plan here and tracks progress with checkboxesdocs/lessons.md— After every correction, Claude records what went wrong and the rule to follow. This file persists between sessions, so Claude learns your preferences over time.
| Agent | Purpose | Model |
|---|---|---|
ba |
Requirements analysis, feature planning, MVP scoping | opus |
ux-researcher |
User flows, information architecture, wireframes | opus |
ui-builder |
Visual design, component building, Figma-to-code, styling | opus |
design-reviewer |
Anti-AI-slop & craft review of built pages (via impeccable) |
opus |
accessibility-auditor |
WCAG 2.1 AA compliance, contrast, keyboard nav, ARIA | sonnet |
developer |
Code implementation, API integration, state management | sonnet |
copywriter |
UX copy, microcopy, error messages, button labels | sonnet |
tester |
Visual QA, responsive testing, functional testing | sonnet |
Design-system work is handled by the
shain-dlsskill (see Design Skills below), not a dedicated agent.
| Rule | Purpose |
|---|---|
workflow.md |
Agent pipeline: BA → UX → UI → Design-Reviewer → A11y → Developer → Tester |
design-skills-policy.md |
Default style (impeccable), choosing another, and the anti-slop pass |
design-principles.md |
Visual hierarchy, typography, color, spacing, responsive, motion |
code-style.md |
Multi-stack code conventions (React, Vue, Svelte, Python, HTML) |
verification.md |
Self-verification loop before reporting work done |
skills.md |
How design skills are installed, updated, and security-checked |
updates.md |
How the kit keeps itself up to date with upstream |
git-operations.md |
Commit/push safety, PR format |
| Command | Purpose |
|---|---|
/figma-to-code |
Convert Figma designs to working code |
/review-design |
Comprehensive design review (UX + UI + A11y + Copy) |
/use-design-md |
Drop a brand DESIGN.md (Linear, Vercel, …) into the project for matching UI |
Installed automatically on first launch (in the background) from pinned upstream sources — never silently vendored. No command needed; ./install.sh exists only as a manual fallback. See .claude/skills/SOURCES.md for the full manifest, source links, pinned commits, and licenses, and .claude/rules/skills.md for the update + security policy.
| Source | What it adds |
|---|---|
| pbakaus/impeccable | impeccable — holistic interface craft, audit, critique & polish |
| anthropics/skills | frontend-design |
| nextlevelbuilder/ui-ux-pro-max-skill | ui-ux-pro-max — UI styles, palettes, typography, UX guidelines |
| Owl-Listener/designer-skills | ~95 design skills (research, IA, UI, interaction, critique, design systems) |
| emilkowalski/skills | emil-design-eng, review-animations — micro-animations & interaction polish |
| Mindrally/skills | lottie — performant Lottie animations on the web |
| Dishain/Shain-Design-System-Skill | shain-dls — token-based design systems |
Reference resources (not skills): awesome-design-md (brand DESIGN.md files, via /use-design-md) and awesome-design-systems (a curated list of real-world design systems for inspiration).
How styles are chosen: by default the kit builds UI in the impeccable style. Say "use ui-ux-pro-max", "follow our design system" (shain-dls), or /use-design-md <brand> to switch the look. Whatever the style, after each page is built the design-reviewer agent runs an impeccable pass to clean up AI slop — without overriding your chosen style. See .claude/rules/design-skills-policy.md.
Standard Feature: BA → UX-Researcher → UI-Builder → Design-Reviewer → A11y Auditor → Developer → Tester
Design Only: BA → UX-Researcher → UI-Builder → Design-Reviewer → A11y Auditor
Bug Fix: Developer → Tester
The .mcp.json includes these integrations (configure API keys as needed):
| Server | Purpose |
|---|---|
| Figma | Fetch designs, extract tokens, screenshots |
| Shadcn UI | Component library reference |
| Context7 | Up-to-date framework documentation |
| GitHub | PR, issues, code access |
Edit .mcp.json and set:
FIGMA_API_KEY— Figma → Settings → Personal Access TokensCONTEXT7_API_KEY— context7.comGITHUB_PERSONAL_ACCESS_TOKEN— GitHub → Settings → Developer settings → Tokens
Create .claude/agents/my-agent.md:
---
name: my-agent
description: "What this agent does. Trigger words — EN: keyword1, keyword2. Trigger words — UA: слово1, слово2."
model: sonnet
color: blue
---
# Agent Title
Instructions for the agent...Create .claude/rules/my-rule.md with markdown content. Rules are auto-loaded for every conversation.
All agents include trigger words in English (primary) and Ukrainian (fallback). To add another language, extend the description field in the YAML frontmatter.
Projects are usually started from a downloaded archive, so the kit files can fall behind. The kit checks for its own updates and can pull them safely:
- On session start, a hook (
.claude/scripts/bootstrap.sh) installs skills on first run and quietly checks the upstream repo for kit updates (at most once a day). If a newer version exists, Claude tells you and offers to update. - Run
./update.shto apply. It backs up the current kit files to.kit-backup/first, then replaces only kit-owned files (agents, rules, commands, scripts,CLAUDE.md,README.md,GUIDE.md,install.sh). - It never touches your work, your project memory (
docs/lessons.md,docs/todo.md), your API keys (.mcp.json), or.claude/settings.json. ./update.sh --checkjust reports whether an update is available.
See .claude/rules/updates.md for the full policy. The current version is in
VERSION.
- MooseDesign1/DDD — Design Driven Development — a full design-first Claude Code framework (separate project, not bundled here).
MIT for this kit. Bundled design skills keep their own upstream licenses — install.sh pulls them from source rather than redistributing copies. See .claude/skills/SOURCES.md for each skill's license (note: emilkowalski/skills ships without an explicit license, so it is fetched at install time and never copied into this repo).