Guide · Open source

How to use agent-skills: install and run it with AI coding agents

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

agent-skills is an MIT-licensed collection of Markdown skills that describe how experienced engineers approach each phase of building software, from defining a spec to shipping. This guide is for developers who already use an AI coding agent and want it to follow structured, test-driven workflows. By the end you will know how to install the skills, run the lifecycle slash commands, and tell whether your agent is supported.

Agent Skillsaddyosmani/agent-skills · ★ 113k · MIT Install options & downloads

Quick start

1Check prerequisites. Confirm you have Node.js and an AI coding agent such as Claude Code, Cursor, or Codex installed.
2Browse the skills. Run the --list command to see the available skills before installing anything.
3Install the skills. Run npx skills add addyosmani/agent-skills for all skills, or add --skill with a name for one.
4Or use a native plugin. In Claude Code, add the marketplace and install agent-skills@addy-agent-skills. Other agents have their own steps in the README.
5Run a lifecycle command. Start with /spec to define the work, then /plan to split it into tasks.
6Build and verify. Use /build one slice at a time, then /test and /review before you ship.

What you need before installing agent-skills

You need an AI coding agent such as Claude Code, Cursor, or Codex. The skills are instructions for those agents, not a standalone program, so they do nothing on their own.

The fastest install path uses the skills CLI through npx, which requires Node.js. The README does not state a minimum Node.js version.

An AI coding agent such as Claude Code, Cursor, or Codex
Node.js, required for the npx skills CLI

Install all skills or pick individual ones with npx

The open skills CLI installs into 70+ agents. Install the full set of 25 skills with one command, or list them first to browse before installing.

To install a single skill, name it with the --skill flag. A per-skill install copies only skills/<name>/, so shared repo-level references/ files are not included. The README suggests a whole-repo install, cloning the repository, or copying the needed checklist into a references/ directory if you need them.

npx skills add addyosmani/agent-skills            # install all 25 skills
npx skills add addyosmani/agent-skills --list     # browse before installing
npx skills add addyosmani/agent-skills --skill code-review-and-quality

Install a native integration for your agent

If you use Claude Code, the README recommends the marketplace plugin. Run the two plugin commands inside Claude Code. If SSH cloning fails because you lack GitHub SSH keys, use the full HTTPS URL for the marketplace-add step.

For other tools the README gives separate setup docs. Cursor uses .cursor/skills/ and .cursor/rules/*.mdc files that you create in your project, since the repo does not ship a .cursor/ folder. Codex uses a plugin marketplace command (Codex CLI v0.122+ per the README). Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro, Command Code and Antigravity CLI each have their own steps.

Claude Code: marketplace plugin or a local clone started with --plugin-dir
Cursor: copy workflow skills to .cursor/skills/ and short policies to .cursor/rules/
Codex: register the marketplace, then add the plugin
Other agents: see the setup doc linked for that tool in the README

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

Run the lifecycle slash commands

The skills map to a development lifecycle: define, plan, build, verify, review and ship. Each command activates the relevant skills automatically. In Claude Code, /spec defines what to build, /plan breaks it into small atomic tasks, /build works one slice at a time, /test proves the work, /review checks it before merge, and /ship goes to production.

Where a built-in command has the same name, as /review does in Claude Code, use the namespaced form. The README also lists /constraints, /webperf and /code-simplify. For /build auto, the README says it generates the plan and implements every task after you approve the plan once. Each task is still test-driven and committed individually, and it pauses on failures or risky steps.

/spec: spec before code
/plan: small, atomic tasks
/build: one slice at a time; /build auto runs the approved plan
/test: tests are proof
/review: improve code health (use /agent-skills:review if it clashes)
/ship: faster is safer

/agent-skills:review

Use skills by name when slash commands are not available

Skills also activate on their own based on your task. Designing an API triggers api-and-interface-design, and building UI triggers frontend-ui-engineering.

The lifecycle slash commands are Claude Code commands. In GitHub Copilot they do not appear, so you invoke skills by name such as /spec-driven-development. In Codex you invoke skills with @, for example @spec-driven-development.

Claude Code: lifecycle slash commands
Copilot: invoke skills by name
Codex: invoke skills with @skill-name

@spec-driven-development

What people use it for

Writing a spec and plan before an AI agent touches the code
Getting test-driven, incremental output from an agent on a feature
Running a five-axis review of changes with code-review-and-quality before merge
Sharing one set of engineering checklists across a team's AI-assisted work
Auditing web performance with /webperf before optimizing

Frequently asked questions

How do I install agent-skills?

Run npx skills add addyosmani/agent-skills to install all 25 skills with the skills CLI. In Claude Code you can instead run the marketplace plugin commands from the README.

What do I need to use agent-skills?

You need an AI coding agent such as Claude Code, Cursor, or Codex, plus Node.js for the npx skills CLI. The README does not state a required Node.js version.

Is agent-skills free?

The project is licensed under MIT, as stated in the repository. The README does not mention pricing.

Does agent-skills work with Cursor?

Yes, the README has a Cursor section. Put workflow skills in .cursor/skills/ and short policies in .cursor/rules/*.mdc in your own project, since the repo does not ship a .cursor/ folder.

Why does /plugin install fail with a Permission denied (publickey) error?

The marketplace clones over SSH. Use the full HTTPS URL in the marketplace-add step, or run git config --global url."https://github.com/".insteadOf git@github.com: as the README recommends.

What are the alternatives to agent-skills?

The source lists Cursor Rules collections, Aider conventions files, Continue and Cline as comparable projects. Choose based on which agent you use and whether you want lifecycle workflows.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

agent-skills suits developers and teams who already use a supported AI coding agent and want spec-first, plan-first, test-driven workflows. If you do not use an AI coding agent, or you are a beginner looking for a general programming course, pick something else, since the skills assume software engineering experience.

Install options & downloads