Agent Skills is a collection of production-grade engineering skills for AI coding agents by Addy Osmani (Google) with 25.9K stars on GitHub, making it one of the most popular frameworks for developing AI coding agents.
Agent Skills Overview
Core Concept
Agent Skills is not just a collection of prompts, but a complete engineering methodology for AI coding agents: Contains 20 skills covering the entire lifecycle Designed following Google engineering culture principles Supports multiple agent platforms
7 Slash Commands
Agent Skills has 7 main commands:
/spec - Define requirements and specifications
/plan - Create implementation plan
/build - Execute the build process
/test - Run comprehensive tests
/review - Code review and quality check
/code-simplify - Simplify complex code
/ship - Deploy and release
Structure of Each Skill
Core Components
Every skill in Agent Skills has the same structure:
- Overview - Brief description of the skill
- When to Use - Situations where this skill should be used
- Process - Step-by-step working process
- Rationalizations - Common excuses AI uses + rebuttals
- Red Flags - Things to avoid
- Verification - How to verify correctness
Specialist Agent Personas
Agent Skills has 3 specialist agents: Code Reviewer - Check code quality Test Engineer - Perform comprehensive testing Security Auditor - Check security
Key Design Principles
1. Process-Driven (Not Reference)
Skills are designed as processes, not just references: Emphasize clear working steps Have clear decision logic Can be immediately applied
2. Anti-Rationalization
The system is designed to fight rationalizations: Collects common excuses used by AI Has rebuttals for each excuse Forces AI to follow the process
3. Verification is Non-Negotiable
Every skill must have verification methods: Clear test cases Measurable success criteria Quality gates that must be passed
4. Progressive Disclosure
Designed following progressive disclosure principles: Show necessary information first Add details when needed Don't overload with too much information
Google Engineering Principles
Hyrum's Law
"With a sufficient number of users of an API, it does not matter what you promise in the contract: all observable behaviors of your system will be depended on by somebody."
Beyoncé Rule
"If you liked it, then you should have put a test on it."
Test Pyramid
70% Unit tests 20% Integration tests 10% E2E tests
Trunk-Based Development
Everything merges to trunk/main No long-lived branches Use feature flags instead
Multi-Platform Support
Supported Agents
Agent Skills supports multiple platforms: Claude Code (Official Marketplace + Superpowers Marketplace) Cursor (Plugin Marketplace)
Gemini CLI
Windsurf
Copilot
OpenCode
Codex
Installation Methods
claude install @addyosmani/agent-skills
# Cursor Plugin
cursor:install-plugin agent-skills
# Or clone and use directly
git clone https://github.com/addyosmani/agent-skillsEcosystem Benefits
For Developers
Increase coding productivity Reduce bugs through organized processes Learn Google best practices
For Teams
Standardize engineering practices Eliminate knowledge silos Consistently improve code quality
For Organizations
Reduce development costs Increase delivery velocity Build culture of engineering excellence
Latest Version
v5.0.7 — Added support for: Copilot CLI support OpenCode fixes General performance improvements Reference: https://github.com/addyosmani/agent-skills