A custom skill for Claude Code that encodes full-stack web development best practices into a reusable instruction set. When installed, it automatically loads into your Claude Code sessions and guides the AI through systematic development workflows — architecture, debugging, code quality, security, and deployment.
This skill is inspired by Andrej Karpathy's technique of crafting detailed rules/instruction files for AI coding assistants. The idea: instead of relying on the AI's general knowledge alone, you encode your development philosophy, patterns, and workflows into a structured document that the AI loads as context. This gives the AI a consistent, opinionated framework to follow — like pairing with a senior engineer who always applies the same high standards.
Karpathy demonstrated that well-written custom instructions dramatically improve the quality and consistency of AI-generated code. This skill takes that concept and packages it as a reusable, installable module for Claude Code.
This is not an application or library. It's a knowledge module — two markdown files (SKILL.md and references/project-scaffolding.md) that Claude Code reads as context before helping you with coding tasks. Think of it as a persistent system prompt for full-stack development.
When you ask Claude Code to build a feature, fix a bug, or architect a system, this skill ensures it follows a consistent methodology rather than ad-hoc responses.
npx skills add https://github.com/adamperlis/fullstack-coding-skill --skill fullstack-codingCore development loop — Every task follows the same 5-step cycle: Understand → Plan → Build → Verify → Refine. Small fixes get 30 seconds of thought; large features get real attention at each step.
Architecture — Feature-based file organization, separation of concerns (UI / State / Data / Logic), RESTful API design, database migration patterns, cursor vs offset pagination.
Code quality — Typed error handling (user errors vs system errors vs programmer errors), practical TypeScript patterns (discriminated unions, unknown over any, satisfies), and a testing strategy prioritized by business impact.
Debugging — A systematic 5-step diagnosis workflow: reproduce → isolate the layer → read the error → form a hypothesis → verify the fix. Includes a symptom-to-layer lookup table.
Performance — Frontend (critical rendering path, LCP, code splitting, image optimization, list virtualization) and backend (query optimization, N+1 prevention, connection pooling, latency budgets).
Security — Input validation, parameterized queries, output escaping, HTTPS, password hashing, secrets management, CORS/CSP, auth token patterns, rate limiting.
Deployment — Environment separation (dev/staging/prod), CI/CD principles, monitoring from day one (error tracking, uptime, structured logging).
Refactoring — When to refactor (and when not to), safe refactoring steps, keeping refactors separate from behavior changes.
Project scaffolding — Universal setup checklists, starter templates by project type (API service, full-stack web app, static site), environment variable patterns, auth provider recommendations, deployment platform comparison, and monorepo decision matrix.
The skill activates on any coding task involving frontend (React, Vue, Svelte, HTML/CSS), backend (Node.js, Python, serverless, REST/GraphQL), databases, deployment, CI/CD, or dev tooling. It also triggers on keywords like "build," "fix," "debug," "refactor," "deploy," "scaffold," or any technical architecture discussion.
SKILL.md # Main skill — development methodology and patterns
references/
project-scaffolding.md # Practical templates, checklists, and config examples