rubencr.com
Sign in
Back to Home

Agent Skills

Production-tested skills for Claude Code

Agent Skills

Agent Skills is a curated collection of opinionated, production-tested skills for Claude Code that encode best practices for full-stack development, testing, security, and architecture. The goal is to teach Claude how to build things the right way — not just what to build.

Each skill follows the Agent Skills specification and is designed to be lean and agent-optimized — short instructions loaded dynamically, with deep-dive references only when needed.

Skills use a 3-layer loading system: metadata loads first (~100 tokens), then the full SKILL.md instructions when matched, and finally deep-dive references on demand — keeping context usage minimal.

View on GitHub

Design principles:

  • •SKILL.md carries the weight — core rules, decision trees, and inline examples live in the main file.
  • •References are optional deep dives, loaded only when Claude needs detail on a specific topic.
  • •Under 500 lines per SKILL.md — every line must justify its token cost.
  • •Examples inline, not in separate files — they belong next to the rules they demonstrate.
  • •Every feature must include tests. Enforced in every skill involving code.

Agent Skills is designed for developers who want Claude to follow real engineering standards — not just generate code that compiles.

By encoding architecture decisions, testing patterns, and security practices into reusable skills, teams can ensure consistent quality across every AI-assisted interaction.

Teaching AI how to build things the right way.

Skills

6 production-tested skills covering the full development lifecycle

Backend Development

387 lines

Python + FastAPI architecture following light DDD, SOLID principles, Pydantic validation, pytest testing, and container security.

Web Development

495 lines

Next.js/React architecture with feature-first organization, ports & adapters pattern, mappers, and TypeScript contracts.

Web Testing

439 lines

Vitest + Testing Library with boundary testing, coverage analysis, MSW mocking, WebGL/Canvas support, and accessibility testing.

Cybersecurity

150 lines

OWASP top 10 guidance, container hardening, AI/agent security, prompt injection defense, and security checklists.

Frontend Design

42 lines

UI aesthetics, design thinking, typography, color, motion, and spatial composition guidelines.

MCP Builder

236 lines

MCP server development in Python (FastMCP) and TypeScript with evaluation framework and best practices.

How It Works

3-Layer Loading System

1

Metadata Loads

~100 tokens — skill name and description, always kept in context for matching.

2

SKILL.md Loads

Full instructions triggered when the skill matches — architecture rules, decision trees, examples.

3

References On Demand

Deep dives load only when Claude needs detail on a specific topic like WebGL testing or container hardening.

View on GitHub