This repository provides a community-driven collection of AI agent "skills" designed to enhance the coding capabilities of assistants like Claude Code and OpenCode. These specialized instruction sets offer on-demand context for specific frameworks, libraries, and architectural patterns, enabling AI assistants to generate code adhering to best practices and conventions, thereby improving code quality and developer productivity.

How It Works

The core concept revolves around "skills," which are curated sets of instructions, patterns, rules, code examples, and anti-patterns related to specific technologies. When an AI assistant detects a relevant context (e.g., working with Angular components), it loads the corresponding skill file. This process provides the AI with targeted guidance, ensuring adherence to established conventions and best practices. The project differentiates between "Curated Skills," personally tested by the maintainer, and "Community Skills," which are democratically voted upon by users before acceptance.

Quick Start & Requirements

• Installation: For Claude Code/OpenCode, clone the repository and copy desired skills from curated/* to ~/.claude/skills/. Manual installation involves creating the ~/.claude/skills/ directory and copying skill folders. Skills are referenced in a project's CLAUDE.md or global ~/.claude/CLAUDE.md.
• Prerequisites: Requires an AI assistant compatible with the skill loading mechanism (e.g., Claude Code, OpenCode). No other specific software dependencies are listed for the skills themselves.

Highlighted Details

• Extensive support for modern frontend and backend technologies, including Angular (standalone components, signals), React 19, Next.js 15, TypeScript, Tailwind CSS 4, Zod 4, Zustand 5, Vercel AI SDK 5, Django REST Framework, and Java 21.
• Includes specialized skills for testing frameworks like Playwright and Pytest, as well as architectural patterns (e.g., hexagonal architecture for Java).
• Features a github-pr skill to facilitate the creation of Pull Requests with conventional commits.
• Community contributions are integrated via a democratic voting process, ensuring a broad range of practical skills.

Maintenance & Community

The project is community-driven, with contributions welcomed and integrated through a structured process outlined in CONTRIBUTING.md. New skills are subject to a 7-day community voting period based on GitHub reactions, with acceptance contingent on positive votes outweighing negative ones. Some skills are developed in collaboration with external teams, such as Prowler.

Licensing & Compatibility

The project is licensed under the MIT License, which is permissive and generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The effectiveness of these skills is directly dependent on the integration capabilities and interpretation accuracy of the target AI assistant (e.g., Claude Code, OpenCode). The community voting system for new skills may introduce subjective biases or delays in the inclusion of certain contributions.