Agent Skills

Agent Skills is a collection of production-grade engineering skills for AI coding agents, built by Addy Osmani (the Google engineer behind Chrome DevTools). Instead of tutorials, these are packaged workflows with verification gates that agents follow across every phase of development. The repository is MIT licensed and reports around 90.9K stars.

It installs into 70+ agents and works with Claude Code, Codex, Cursor, GitHub Copilot, Windsurf, Gemini CLI, Cline, and Antigravity. Use the open skills CLI for the fastest path.

Key Features

• Lifecycle slash commands: 9 commands map Define, Plan, Build, Verify, Review, and Ship to the dev cycle, plus /constraints, /webperf, and /code-simplify.
• Automatic activation: skills fire based on what you are doing, so API design triggers api-and-interface-design and UI work triggers frontend-ui-engineering.
• Automated /build auto: approves a plan once, then runs each task autonomously while keeping every step test-driven and individually committed.
• Five-axis review: code-review-and-quality runs a review before merge.
• Portable install: npx skills add addyosmani/agent-skills installs all 25 skills, or grab one with --skill.

Use Cases

Who Should Use This Tool?

• Clone and go: any agent, one command, no manual policy files.
• Teams that want consistency: encode senior-engineer quality gates so every agent follows the same bar.
• Engineers shipping to prod: use /spec, /plan, /build, and /ship as a repeatable pipeline.

Problems It Solves

1. Ad-hoc prompting: converts vague instructions into structured phases with review gates.
2. Inconsistent outputs: skills enforce the same practices across devs and models.
3. Untested changes: every task stays test-driven, and risky steps pause for human review.

Pricing

Agent Skills is MIT licensed and free. You pay only for the model tokens your agent uses; there is no fee for the skills themselves.

Advantages & Unique Selling Points

1. Verified workflows over content: gates like spec-before-code, tests-as-proof, and Hyrum's Law are baked in.
2. Big ecosystem reach: install once into most coding agents, not one editor.
3. Reproducible lifecycle: npx skills add gives the same workflow to everyone on the team.

Getting Started

npx skills add addyosmani/agent-skills            # install all 25 skills
npx skills add addyosmani/agent-skills --list     # browse before installing

For Claude Code, use the plugin marketplace instead: /plugin marketplace add addyosmani/agent-skills then /plugin install agent-skills@addy-agent-skills. For a local clone, run claude --plugin-dir /path/to/agent-skills.

Frequently Asked Questions

Is it free?

Yes, MIT licensed. You only pay the model token costs of your agent.

Which agents are supported?

Nearly 70+ agents via the skills CLI, plus native integrations for Claude Code, Cursor, Codex, Copilot, Windsurf, Gemini CLI, Cline, Antigravity, and more.

Do I need to install every skill?

No. Use --skill <name> to install just the one you need, for example --skill test-driven-development.

What does /build auto do?

It generates the plan and implements every task in a single approved pass. You approve once, then it runs autonomously, still committing each task individually and pausing on failures or risky steps.

Alternatives

• LangGraph: orchestration runtime when you need to build the agent graph yourself.
• agentmemory: persistent memory to pair with your skills.
• MCP Framework: build the tools your agents call.

Tips

1. Start with /spec before any code; it keeps scope and assumptions explicit.
2. Use /build auto only after the spec is stable, since it batches the implementation.
3. Enforce /constraints at the start so the quality bar is set once and applied everywhere.

Conclusion

Agent Skills packages senior-engineer practice into skills your agent follows on its own. If you want consistent, reviewed, test-driven output from any coding agent, start at skills.addy.ie.