Agent Skills - Google's open-source AI programming agent skillset

What are Agent Skills?

Agent Skills is an open-source AI programming agent skills package developed by Addy Osmani, head of Google's Gemini team. Agent Skills encapsulates the workflow and engineering standards of senior Google engineers into 20 Skills, 7 Slash commands, and 3 Agent personas, covering six development phases: definition, planning, building, validation, review, and release, ensuring that AI follows production-grade engineering disciplines at every stage.

Main functions of Agent Skills

• 20 production-level skillsIt covers the entire software lifecycle, with each Skill including steps, checkpoints, exit criteria, and verification evidence requirements.
• 7 Slash Commands:/spec Requirements analysis/plan Task splitting/build Incremental implementation/test Run tests,/review Code review,/code-simplify Simplify code,/ship Delivered and launched.
• 3 Agent Personas: code-reviewer (five-axis code review), test-engineer (test strategy and coverage analysis), security-auditor (vulnerability detection and OWASP assessment).
• Automatic skill triggerAutomatically activated when designing the API. api-and-interface-designAutomatically activated when building the UI. frontend-ui-engineering.
• 4 reference listsA quick reference document for test modes, security checks, performance optimization, and accessibility.
• Anti-rationalization mechanismEach Skill has built-in AI common excuses (such as "test later") and rebuttal arguments, forcing compliance with the specifications.

How to use Agent Skills

• Visit the GitHub repository:https://github.com/addyosmani/agent-skills Browse the project documentation and Skills directory.
• Claude Code user:implement /plugin marketplace add addyosmani/agent-skills Add market sources, then execute. /plugin install agent-skills@addy-agent-skills Installation complete.
• Cursor users: The required SKILL.md Copy the file to the project root directory. .cursor/rules/ In the folder.
• Gemini CLI users:run gemini skills install https://github.com/addyosmani/agent-skills.git --path skills Install as a native skill.
• Windsurf / GitHub Copilot / Codex usersReference Project docs/ Configure the corresponding access guide in the directory.
• Local developmentAfter cloning the repository, pass claude --plugin-dir /path/to/agent-skills Load and use.
• Start coding:Use in the project directory /spec Analyze the requirements, according to /plan → /build → /test → /review → /ship Development proceeds sequentially.

Key information and usage requirements for Agent Skills

• Open source licenseMIT, freely usable for projects, teams, and business tools.
• Author BackgroundAddy Osmani (Google Gemini team lead, former Chrome engineer).
• Star TrendThe number of users exceeded 23,000 in April 2026, continuing its rapid growth.
• Tool compatibilitySupports mainstream AI programming tools such as Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro IDE & CLI, and Codex.
• PrerequisitesIt requires an AI programming agent that supports Skill/rule files; pure chat models cannot be loaded directly.
• Language supportSkill documentation is in English Markdown, but it is applicable to projects in any programming language.

Agent Skills' core advantages

• Engineering discipline solidificationThe "steady, accurate, and ruthless" habits of Google's senior engineers have been distilled into reusable processes.
• Full-link coverageEach of the six stages from Idea to Go Live has a corresponding Skill, preventing AI from skipping key steps such as specification, testing, and review.
• Evidence-drivenVerifiable deliverables are mandatory at each stage to prevent perfunctory outputs that "look fine".
• Lightweight on-demand loadingThe progressive disclosure design reduces token consumption and remains efficient even in long contexts.
• Cross-platform compatibilityIt is a pure Markdown format, not bound to a specific IDE or model, and has low migration costs.

Agent Skills project address

• GitHub repositoryhttps://github.com/addyosmani/agent-skills

Comparison of Agent Skills with similar products

Application scenarios of Agent Skills

• Standardized coding for individual developersEstablish a complete engineering discipline for AI programming assistants, from requirements definition to production release, to prevent models from taking shortcuts and outputting low-quality code.
• Unified Team Collaboration StandardsEncode team code reviews, testing strategies, and security standards into reusable skills to ensure consistent output style and quality across different team members when using AI.
• Production-level project quality controlIn commercial projects that require long-term iteration, through /spec → /plan → /build → /test → /review → /ship Mandatory verification across the entire supply chain ensures reliable delivery.
• Multi-Agent Parallel Automated ReviewBefore going live, three Specialist Personas—code-reviewer, test-engineer, and security-auditor—are triggered simultaneously to generate three reports—code, test, and security—for a comprehensive decision.
• Legacy system migration and technology debt management:use deprecation-and-migration and code-simplification Skills like these allow for the safe removal of zombie code and the simplification of overly complex modules, based on the concept that "code is a liability."