Claude Scaffold Skill
veekunth217/claude-scaffold-skill
Gere projetos completos com /scaffold, descubra skills e sincronize contexto entre máquinas
Ver no GitHubColeção de 30+ skills para Claude Code que automatiza scaffolding de projetos, geração de boilerplate production-ready e sincronização de contexto entre dispositivos. Inclui `/scaffold` para criar projetos descritos em linguagem natural, `/bootstrap` para descobrir skills relevantes e `/sync` para manter memória de projeto sincronizada.
Instalação
npx convez-ai veekunth217/claude-scaffold-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Inicializar novos projetos full-stack sem templates manuais
- Descobrir e instalar skills comunitárias relevantes para sua stack
- Sincronizar memória de projeto e configuração entre desenvolvimento local e remoto
Veredito

Skill robusta que resolve fragmentação real no fluxo de Claude Code — elimina repetição de boilerplate e contextualização entre projetos. Comunidade ativa com 35+ skills registradas e validação automática, mas requer familiaridade com Claude Code e arquitetura para máximo aproveitamento.
README
A tradução para o português deste README ainda está na fila.
claude-scaffold-skill
Turn Claude Code into a full dev toolkit — scaffolding, skills, and workflow automation in one install.
One install. Scaffold any stack. Discover the right skills. Sync context across devices.
┌─────────────────────────────────────────────┐
│ 📹 Demo GIF coming — recording in progress │
│ /scaffold → "FastAPI backend with Postgres" │
│ → plan shown → GO → files generated │
└─────────────────────────────────────────────┘
Why this exists
Claude Code is powerful. But out of the box, every new project starts the same way — blank directory, blank CLAUDE.md, you explaining your stack from scratch again.
This skill collection fixes that:
/scaffold— describe your project in plain English, get production-ready files, a CLAUDE.md, and.vscode/config. No templates to browse, no boilerplate to copy./bootstrap— after every scaffold, Claude surfaces the right community skills for your stack, tiered by how much you need them./sync— Claude stores project memory in a path-hashed folder that breaks between machines. Sync exports it to git, imports it anywhere.
Everything else (/hooks, /budget, /handoff, /dark-mode, /launch, /saas...) extends your workflow once the foundation is set.
The registry also surfaces Anthropic's official document skills (PDF, Excel, Word, PowerPoint) and the Vercel cross-agent skills CLI as awareness cards — so you always know what's available beyond this collection.
Install
git clone https://github.com/veekunth217/claude-scaffold-skill.git \
~/.claude/skills/claude-scaffold-skill
Open Claude Code in any project and type /scaffold. That's it.
No restart needed. Claude Code picks up skills from
~/.claude/skills/automatically.
Keeping skills updated
New skills land in this repo every week. To pull the latest:
# Quick — just one command
cd ~/.claude/skills/claude-scaffold-skill && git pull
Or use the built-in updater (it scans every git-installed skill in ~/.claude/skills/, shows you what changed, asks before pulling):
/update-skills
Other install paths:
- Skills installed via
/plugin install(e.g. Anthropic document skills) → update with/plugin update - Skills installed via
npx skills add(Vercel CLI) → update withnpx skills update
Quick Start — 60 seconds to first scaffold
mkdir my-app && cd my-app
Open Claude Code → type:
/scaffold
Claude checks your environment silently, then asks one question:
What are you building?
Type anything. Plain English. Claude routes it, asks 3-5 focused questions, shows you the full plan, and waits for:
GO
Before touching a single file.
How the routing works
The main SKILL.md reads your description and routes to the right specialist — no flags, no subcommands.
/scaffold "a FastAPI backend with PostgreSQL"
→ /python → pyenv check → venv → DB → Docker → CI → GO → files
/scaffold "Node REST API with Prisma"
→ /nodejs → nvm check → TypeScript → ORM → auth → GO → files
/scaffold "Terraform AWS EKS cluster"
→ /terraform → AWS EKS preset or custom picker → GO → .tf files
/scaffold "WordPress plugin for WooCommerce"
→ /wordpress → plugin boilerplate → DDEV → WP-CLI → GO → files
/scaffold "not sure what stack to use"
→ /suggest → options with tradeoffs → you pick → routes to specialist
Every route follows the same contract: show the plan, wait for GO, then generate.
When you say "fresh project," /scaffold asks where it should land — current directory, a new subfolder you name, or an absolute path — before generating anything. Never overwrites a non-empty directory unless you explicitly chose "Existing" mode.
Curious what
/scaffoldactually generates? See examples/ for committed reference outputs (FastAPI + Postgres, Next.js + Tailwind) — same shape/scaffoldwrites into your project.
Document support — Anthropic's official skills, surfaced for you
If your project ever touches PDFs, Excel, Word, or PowerPoint, /bootstrap surfaces Anthropic's official document skills as an awareness card:
# Install the official Anthropic marketplace
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
This gives Claude native read/write for .pdf, .xlsx, .docx, and .pptx — automatically suggested for SaaS, WordPress, Python data, reports, docs, and ETL projects.
The registry also surfaces the Vercel cross-agent Skills CLI (npx skills add / npx skills find) so you know about the broader skills ecosystem beyond this collection.
Skills
Wizards — ask questions, generate real code
| Skill | Command | What it does |
|---|---|---|
| Scaffold | /scaffold | Main entry — plain English → right specialist |
| Python | /python | FastAPI / Django / Flask / Celery / Jupyter — pyenv, venv, DB, Docker, CI |
| Node.js | /nodejs | Express / Fastify / NestJS / Hono — nvm, TypeScript, Prisma/Drizzle, auth |
| Terraform | /terraform | AWS EKS preset · DO Kubernetes preset · custom component picker |
| Web App | /webapp | Angular/React/Vue + Node — DB, auth, Docker, deploy |
| GraphQL | /graphql | Apollo / Yoga / Pothos + Prisma / ScyllaDB + DataLoader + codegen |
| Database | /database | Schema-first wizard — PG / ScyllaDB / Redis, migrations, query patterns |
| Deploy | /deploy | DigitalOcean or AWS EC2 — Nginx, SSL, systemd/PM2, deploy script |
| WordPress | /wordpress | Site / plugin / theme — DDEV, WP-CLI, hooks, REST endpoints |
| Suggest | /suggest | No stack preference → options with tradeoffs + recommendation |
| Clone | /clone | Any GitHub repo as skeleton — strips history, detects stack, adds CLAUDE.md |
| Design | /design | Extract design tokens from any website via designmd.me → CSS vars, Tailwind, theme.ts |
| Dark Mode | /dark-mode | Full dark mode + Black or White animated toggle (MJ tribute) — View Transitions API |
| SaaS | /saas | Product wizard — auth, billing, onboarding → GSD phase plan + UI/UX Pro Max brief |
| Launch | /launch | Submission packs for CodeCanyon · WP.org · Product Hunt · SaaS pre-launch audit |
| Hooks | /hooks | Wire .claude/settings.json — auto-lint, block rm -rf, run tests on edit |
| Budget | /budget | Route Claude Code + Roo Code traffic to free models via local proxy |
| Handoff | /handoff | Split tasks between Claude and Roo Code — generates self-contained Roo prompts |
| Context Sync | /sync | Export/import Claude memory across machines, and push it as adapter files to Cursor, Codex, Roo, Cline, Aider, Windsurf (cross-tool memory bridge — git-native alternative to Vilix) |
| Bootstrap | /skill-bootstrap | Tiered skill installer — runs automatically after every scaffold |
| New Skill | /new-skill | Author a new Claude Code skill — generates SKILL.md + draft registry entry, ready for PR |
| Update Skills | /update-skills | Pull latest commits for every git-installed skill in ~/.claude/skills/ — preview changes, ask before applying |
| Review Skills | /review-skills | Triage the discovered-skills queue 25 at a time — keep good ones into the registry, reject the rest (persisted, resumable) |
Reference Skills — config guides and snippet collections
These are directly-invoked skills (type /aws, /kubernetes, etc.) — they're not part of the /scaffold router. Use them when you want focused help on a specific topic.
Honest status: Most reference skills below are stubs — they have the structure (sections, RULE, frontmatter) but the actual snippet content is still being filled in. They activate and surface the topic to Claude, but won't yield deep snippets until contributors write them. Each stub is clearly marked at the top of its
SKILL.md. PRs welcome — see CONTRIBUTING.md. The/dbskill is the only fully-written reference today.
| Skill | Command | Covers | Status |
|---|---|---|---|
| AWS | /aws | EKS, ECR, VPC, RDS, ElastiCache, S3, Route53, ACM, IAM | 🚧 Stub |
| Kubernetes | /kubernetes | Helm, ArgoCD, Ingress-nginx, HPA, Blue/Green, debug playbook | 🚧 Stub |
| CI/CD | /cicd | GitHub Actions, self-hosted runners, Docker push, OIDC to AWS | 🚧 Stub |
| Server | /server | Nginx, PHP-FPM, Certbot, UFW, fail2ban, Redis, MySQL/PG, PM2 | 🚧 Stub |
| DigitalOcean | /digitalocean | Droplets, Managed DBs, Spaces, Load Balancers, DNS | 🚧 Stub |
| Docker | /docker | Dockerfile best practices, multi-stage, compose, networking | 🚧 Stub |
| Security | /security | OWASP Top 10, WP hardening, server hardening, SSL/TLS, WAF | 🚧 Stub |
| PHP | /php | OOP, Laravel, WordPress plugin dev, REST, wpdb, nonces | 🚧 Stub |
| WooCommerce | /woocommerce | Products, pricing, payment gateways, hooks, checkout | 🚧 Stub |
| WordPress Server | /wordpress-server | Nginx+WP, PHP 8.3-FPM, Redis cache, WP Rocket | 🚧 Stub |
| Database | /db | MySQL, PostgreSQL, MongoDB, Redis, ScyllaDB, Meilisearch | ✅ Ready |
Speed & cost modes — what you can actually toggle
Three usage patterns. Only one is a real toggle; the other two are behavioral patterns you ask for in plain English.
| Mode | How to activate | How to check | What you get |
|---|---|---|---|
| Fast | /fast (real toggle) | type /fast again — it shows ON/OFF | Opus 4.6 with faster output. Costs more tokens. Toggle on for hard problems, off when done. |
| Default | (no toggle — it's the baseline) | Status bar shows model = Sonnet 4.6 | Sonnet 4.6 with sub-agents on demand. Balanced. |
| Careful | Ask: "do this serially, no sub-agents" | (no command — pattern only) | One task at a time. Slowest, cheapest. |
Other useful asks:
- Force parallel: "spawn 4 sub-agents to do X" → faster, costs more
- Check current model: look at the Claude Code status bar — model name shows there
- Restart from fresh context:
/clear
The 3-Tool AI Dev Stack
The most cost-effective way to build with AI in 2026. Claude thinks, Roo types, you pay only for thinking.
| Role | Tool | Best for | Cost |
|---|---|---|---|
| 🧠 Architect | Claude Code | Decisions, debugging, code review, architecture | Paid (Anthropic) |
| ✍️ Executor | Roo Code / Cline | Bulk generation, boilerplate, mechanical refactors | Lower — or free via proxy |
| 🔀 Router | free-claude-code | Routes Roo traffic → OpenRouter free / Ollama / NIM / DeepSeek | Free (local) |
Set up in 5 minutes:
/budget → installs the proxy, wires VS Code + shell
/handoff → splits any task list between Claude and Roo
The /handoff skill detects when the proxy is running and auto-routes Roo tasks through it. Heavy lifting runs on free models. Critical work stays on Anthropic.
A fourth, optional tier — async agents
For fire-and-forget work (dependency bumps, mechanical PRs, "fix this flaky test"), there's a whole category of asynchronous cloud agents: Jules (Google, Gemini-powered), the OpenAI Codex agent, Devin, GitHub Copilot agent mode. You hand them a task + a repo, they spin up a VM, make the change, run tests, and open a PR you review later.
| Tier | Tool | Interaction model |
|---|---|---|
| 🧠 Architect | Claude Code | live, turn-by-turn |
| ✍️ Executor | Roo Code / Cline | live, bulk in-editor |
| 🔀 Router | free-claude-code | (config — routes Roo to free models) |
| 📨 Delegate | Jules / Codex / Devin | async — fire a task, get a PR |
Not Claude Code skills (so not in the registry) — but if you're already splitting work with /handoff, the "delegate the boring PR to an async agent" tier slots in naturally above it.
Cross-tool memory bridge — git-native alternative to Vilix/Mem0
Use Claude Code and Cursor and Codex CLI and Roo Code and Aider on the same project? Each one has its own instruction/memory file — CLAUDE.md, .cursorrules, AGENTS.md, etc. — and you re-explain your project to every one of them.
/sync tools push writes them all from one canonical source (.claude-context/MEMORY.md, committed to git):
make sync-tools-detect # see which tool configs exist in this project
make sync-tools # generate adapter files for every detected tool
make sync-tools DRY=1 # show what would change without writing
make sync-tools ONLY=cursor,claude
| Tool | Adapter file | Written when |
|---|---|---|
| Claude Code | CLAUDE.md | always |
| Cursor | .cursor/rules/memory.mdc | .cursor/ or .cursorrules exists |
| Roo Code | .roo/rules/memory.md | .roo/ exists |
| Cline | .clinerules/memory.md | .clinerules exists |
| Codex CLI | AGENTS.md | .codex/ or AGENTS.md exists |
| Aider | CONVENTIONS.md | .aider/ or CONVENTIONS.md exists |
| Windsurf | .windsurfrules | .windsurfrules exists |
vs Vilix / Mem0 (proprietary SaaS at ~$20/mo):
- ✅ Git-native — memory lives in your repo,
git blame-able, code-review-able - ✅ Zero vendor lock-in — plain markdown, works offline forever
- ✅ One source of truth, regenerable — new dev runs
make sync-toolsaftergit cloneand every AI tool they use immediately knows the project - ⏳ Not live yet — re-run after editing the source. Phase 2 (spec'd, not built): skills/sync/PHASE2-MCP.md — a small MCP server so Claude / Cursor / Codex / Roo / Cline / Windsurf see edits live, no re-run. ~half-day build, contributions welcome.
Context Sync — take your Claude memory anywhere
Claude stores project memory in ~/.claude/projects/<path-hash>/. The hash is derived from your absolute path, so it breaks between machines:
/home/alice/myapp → -home-alice-myapp ← Machine A
/Users/john/myapp → -Users-john-myapp ← Machine B (Claude loses everything)
The /sync skill solves it. Export to .claude-context/ (committed to git), push, import on any device:
# After a productive session on Machine A
python scripts/sync-export.py && git push
# On Machine B after pull
python scripts/sync-import.py
The import script calculates the correct path hash for the current machine automatically.
/sync export → export + commit
/sync import → pull + restore to correct hash
/sync status → diff between exported and live
/sync clean → strip *.jsonl before export (reduces size)
What syncs: MEMORY.md, memory/*.md, feedback files, project notes
What stays private: *.jsonl (conversation history), *.json (may contain keys)
After every scaffold — the skill bootstrapper
Once your project is generated, /bootstrap opens automatically and shows you what to install next:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
CLAUDE CODE SKILL INSTALLER
Project detected: python, fastapi, docker
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔴 ESSENTIALS — pre-selected
[1] ✓ Get Shit Done (GSD) ⭐ 59,791
[2] ✓ Awesome Claude Code ⭐ 42,429
[3] ✓ Claude Code Expert ⭐ 0
[4] ✓ Code Review Graph ⭐ 15,203
🟡 STACK MATCH — for your python + docker project
[5] Claude Memory ⭐ 71,750
🟢 COMMUNITY PICKS
[6] Everything Claude Code ⭐ 172,979
[7] Superpowers ⭐ 177,817
...
💡 ALREADY IN CLAUDE CODE — no install needed
↳ Hooks · MCP Servers · /ultrareview · CLAUDE.md · IDE Extension
↳ The 3-Tool AI Dev Stack (Claude + Roo + free proxy)
↳ Anthropic Document Skills (PDF · Excel · Word · PowerPoint)
↳ Vercel Skills CLI (cross-agent skill discovery)
Currently selected: 1, 2, 3, 4
Type numbers to toggle, "all", "none", or "go" to install.
Community Registry
registry/skills.json tracks 35 community skills (incl. Anthropic's official document marketplace, the Vercel cross-agent CLI, MCP servers, and curated agent collections). It auto-updates every Sunday:
GitHub Action → scans 10 topic queries (claude-skill, claude-skills, etc.)
→ finds repos with SKILL.md in root
→ scores each on 6 quality signals (stars, recency, topics,
description length, README, archived status — max 100)
→ writes the full sorted queue to registry/discovered.json
→ diffs against last run → opens an Issue ONLY for the new
candidates (or stays silent if nothing new appeared)
→ maintainer reads the Issue, picks the worthwhile ones, PRs
→ PR blocked by CI if any repo returns 404
No skill enters registry/skills.json without human review. discovered.json is a sorted candidate queue you browse at your own pace; the weekly Issue only nudges you about what's newly appeared since the last scrape. Quality scoring criteria are documented in CONTRIBUTING.md.
Triaging the queue — page by page
The queue is grouped into sections (🆕 just launched · 🔥 popular · 💎 quiet gems · 📦 long tail) so you can review by flavour. Work through it 25 at a time:
make review-status # how many pending, broken down by section
make review # show the next page (default 25)
make review SECTION=just-launched # only the freshly-launched ones
make review PAGE=3 # jump to page 3
# Record decisions (they persist):
python scripts/review-queue.py --accept owner/repo # → appended to skills.json, verified
python scripts/review-queue.py --reject owner/repo # → added to rejected.json, never shown again
Or run it conversationally inside Claude Code with /review-skills — it shows you a page, you say "keep 1 3 7, reject the rest", it updates the files and offers to commit. Rejected repos go in registry/rejected.json; the scraper skips them on every future run, so the queue only ever shrinks toward "done."
A second registry (registry/claude-capabilities.json) tracks 16 built-in Claude Code features — surfaced as 💡 awareness items matched to your stack.
Add your skill
The easy way — open Claude Code and type /new-skill. The wizard walks you through name, description, tags, and trigger words; generates a properly-structured skills/your-name/SKILL.md, optionally wires a route in the main router, and appends a draft registry entry. Then you push it to your own GitHub repo and open a PR.
The manual way:
- Fork, edit
registry/skills.json:
{
"name": "Your Skill",
"repo": "your-username/your-repo",
"description": "One sentence, max 120 chars",
"tags": ["relevant", "tags"],
"install": "git clone https://github.com/your-username/your-repo.git ~/.claude/skills/your-skill",
"stars": 0,
"verified": false,
"added": "2026-05-05"
}
make validate-full— blocks if your repo 404s- Open a PR
Make your repo auto-discoverable
Add GitHub topics claude-skill or claude-code-skill and put a SKILL.md in root with YAML frontmatter:
---
name: your-skill-name
description: What it does
version: 1.0.0
author: your-github-username
tags: [tag1, tag2]
---
The weekly scraper finds it automatically.
Supported Stacks
| Category | Stacks |
|---|---|
| Frontend | React (Vite+TS), Vue 3, Angular, Next.js, Hugo |
| Backend | Express, Fastify, NestJS, Hono, FastAPI, Django, Flask, Laravel |
| CMS | WordPress — site, plugin, theme, WooCommerce |
| Full-stack | MERN, LAMP, LEMP, Angular+Node, React+Node |
| API | GraphQL (Apollo/Yoga/Pothos), REST (Express/Fastify/FastAPI) |
| Infrastructure | Terraform (AWS EKS preset, DO K8s preset), Terragrunt |
| Deployment | DigitalOcean Ubuntu/CentOS, AWS EC2, AWS ECS |
| Database | PostgreSQL, MySQL, MongoDB, Redis, ScyllaDB, Meilisearch |
Project Structure
claude-scaffold-skill/
├── SKILL.md # Main entry — NL router (Routes A-T)
├── Makefile # Dev commands
├── CONTRIBUTING.md # How to add skills and contribute
│
├── skills/ # 33 skills
│ ├── scaffold/ python/ nodejs/ terraform/ deploy/
│ ├── wordpress/ webapp/ graphql/ database/ suggest/
│ ├── sync/ hooks/ budget/ handoff/ clone/
│ ├── design/ dark-mode/ saas/ launch/ bootstrap/
│ ├── new-skill/ ← author a new skill via /new-skill
│ ├── update-skills/ ← refresh installed skills via /update-skills
│ ├── picker/ ← legacy redirect to /skill-bootstrap (kept for compat)
│ └── aws/ kubernetes/ cicd/ server/ digitalocean/
│ woocommerce/ wordpress-server/ php/ docker/ security/ db/
│
├── examples/ # Reference outputs from /scaffold
│ ├── fastapi-postgres/ # Python + FastAPI + Postgres
│ └── nextjs-tailwind/ # Next.js + TypeScript + Tailwind
│
├── templates/
│ ├── hooks/ # lint.sh, block-dangerous.sh, test-on-edit.sh, notify-stop.sh
│ ├── vscode/ # extensions.json, settings.json, tasks.json
│ └── dark-mode/ # CSS tokens, toggle component, Black or White animation
│
├── registry/
│ ├── skills.json # 35 community skills (21 verified, incl. Anthropic + Vercel)
│ ├── discovered.json # Candidate queue from the scraper (sorted, sectioned)
│ ├── rejected.json # Repos rejected during review — scraper skips these
│ ├── claude-capabilities.json # 16 built-in Claude Code features
│ └── discovered-capabilities.json # Weekly docs scan results
│
├── scripts/
│ ├── sync-export.py # Export context → .claude-context/
│ ├── sync-import.py # Import from .claude-context/ → correct hash
│ ├── fetch-skills.py # GitHub scraper (10 queries, quality scoring, skips rejected)
│ ├── search-registry.py # CLI search across the local registry
│ ├── review-queue.py # Page/triage discovered.json — accept/reject decisions
│ ├── sync-tools.py # Cross-tool memory bridge — write adapter files for every AI tool
│ ├── update-stars.py # Star count refresher
│ └── validate-registry.py # Structure + 404 validator
│
├── references/
│ ├── stacks.md # Version requirements + commands per stack
│ ├── environments.md # Environment detection edge cases
│ └── vscode-extensions.md # Per-stack VS Code extension + settings overrides
│
├── SECURITY.md # Disclosure flow + how to audit 3rd-party skills
│
└── .github/
├── PULL_REQUEST_TEMPLATE.md # PR checklist
├── ISSUE_TEMPLATE/ # Bug report + skill submission forms
└── workflows/
├── sync-registry.yml # Weekly: refresh stars + discover
└── validate-registry.yml # PR gate: blocks 404 repos
Developer Commands
make validate # validate registry structure (fast, no network)
make validate-full # validate + verify all repos exist on GitHub
make test # alias for validate-full (CI entry point)
make search Q=<keyword> # search the registry (e.g. make search Q=pdf)
make update-stars # refresh star counts from GitHub API
make discover # run scraper → registry/discovered.json
make discover-capabilities # scan Claude Code releases → discovered-capabilities.json
Search examples:
make search Q=pdf # → Anthropic Document Skills
make search Q=agents # → 6 entries (Ruflo, Agent Orchestrator, Browser Use, ...)
make search Q=dashboard # → Octogent
Contributing
See CONTRIBUTING.md for the full guide.
- Add a registry entry — edit
registry/skills.json, runmake validate-full, open PR - Add a wizard skill — create
skills/your-skill/SKILL.md, follow confirm-before-generate pattern - Improve an existing skill — PRs welcome, no CLA
- Make your repo discoverable — add
claude-skilltopic to your GitHub repo
Credits
Built on patterns from:
- hmohamed01/Claude-Code-Scaffolding-Skill — conversational CLI patterns and template scaffolding
- reedmayhew18/claude-code-expert — skill split design and production wizard structure
License
MIT — see LICENSE