The addyosmani/agent-skills repository provides a comprehensive suite of production-grade engineering skills designed specifically for AI coding agents. Its purpose is to encode the workflows, quality gates, and best practices that experienced software engineers use, ensuring that AI agents consistently apply these standards throughout every phase of software development. By packaging these skills into structured, actionable workflows, the repository aims to elevate the reliability, maintainability, and quality of code produced by AI agents.
The repository organizes the development lifecycle into six key phases: Define, Plan, Build, Verify, Review, and Ship. Each phase is mapped to a corresponding slash command (such as /spec, /plan, /build, /test, /review, /ship), which triggers the relevant skills automatically. These commands are supported across multiple agent platforms, including Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro IDE & CLI, and Codex, making the skills highly portable and easy to integrate into diverse development environments.
Under the hood, agent-skills features 20 core skills, each encapsulated in its own directory with a SKILL.md file. These skills cover the full spectrum of engineering tasks: from refining vague ideas and writing detailed product requirement documents, to breaking down tasks, implementing code incrementally, conducting test-driven development, designing APIs and interfaces, performing browser testing and debugging, reviewing code for quality and security, optimizing performance, managing git workflows, automating CI/CD pipelines, handling deprecation and migration, documenting architectural decisions, and preparing for production launches. Each skill is structured as a step-by-step workflow with verification gates, anti-rationalization tables (countering common excuses for skipping steps), and clear evidence requirements, ensuring that agents follow rigorous processes rather than relying on intuition or shortcuts.
The repository also includes pre-configured agent personas, such as code reviewers, test engineers, and security auditors, each with specialized perspectives and review standards. Supplementary reference checklists are provided for testing patterns, security, performance, and accessibility, which skills can pull in as needed to minimize context usage and maximize relevance.
A key design philosophy of agent-skills is to enforce process over prose. Skills are not generic prompts or reference documentation; they are actionable workflows with explicit checkpoints and exit criteria. Progressive disclosure is used to keep token usage minimal, loading supporting references only when necessary. The skills are heavily influenced by best practices from Google’s engineering culture, embedding concepts like Hyrum’s Law, the test pyramid, trunk-based development, Shift Left, feature flags, and treating code as a liability.
The project structure is clean and modular, with directories for skills, agent personas, references, session hooks, platform-specific commands, and setup guides. Contributions are encouraged to be specific, verifiable, battle-tested, and minimal, focusing on actionable guidance rather than vague advice.
Overall, addyosmani/agent-skills is a robust toolkit for anyone looking to enhance the discipline and quality of AI-driven software development, ensuring that agents adhere to the same high standards as senior engineers. The repository is licensed under MIT, allowing free use and adaptation in projects, teams, and tools.
