Agent Skills - Google's Open-Source AI Coding Agent Skill Pack

What is Agent Skills

Agent Skills is an open-source AI coding agent skill pack by Addy Osmani, lead of Google's Gemini team. Agent Skills encapsulates the workflows and engineering standards of Google senior engineers into 20 Skills, 7 Slash commands, and 3 Agent personas, covering the six development stages of definition, planning, building, verification, review, and release, so that AI follows production-grade engineering discipline at every step.

Main Features of Agent Skills

20 production-grade Skills: Cover the entire software lifecycle; each Skill includes steps, checkpoints, exit criteria, and verification evidence requirements.
7 Slash commands：/spec requirements clarification,/plan task breakdown,/build incremental implementation,/test run tests,/review code review,/code-simplify simplify code,/ship delivery and release.
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 triggering: Automatically activates when designing APIs api-and-interface-design, automatically activates when building UI frontend-ui-engineering。
4 reference checklists: Quick reference documents for testing patterns, security checks, performance optimization, and accessibility.
Anti-rationalization mechanism: Each Skill has built-in common AI excuses (such as "add tests later") and counterarguments, enforcing execution according to standards.

How to Use Agent Skills

Visit the GitHub repository：https://github.com/addyosmani/agent-skills to browse the project documentation and Skill directory.
Claude Code users: Run /plugin marketplace add addyosmani/agent-skills to add the marketplace source, then run /plugin install agent-skills@addy-agent-skills to complete installation.
Cursor users: Copy the required SKILL.md files into the .cursor/rules/ folder in the project root directory.
Gemini CLI users: Run gemini skills install https://github.com/addyosmani/agent-skills.git --path skills to install as a native skill.
Windsurf / GitHub Copilot / Codex users: Configure according to the corresponding integration guide in the project's docs/ directory.
Local development: After cloning the repository, load and use it via claude --plugin-dir /path/to/agent-skills .
Start coding:In the project directory, use /spec to clarify requirements, and advance development in the order of /plan → /build → /test → /review → /ship .

Key Information and Usage Requirements of Agent Skills

Open-source license: MIT, freely usable for projects, teams, and commercial tools.
Author background: Addy Osmani (lead of Google Gemini team, former Chrome engineer).
Star trend: Surpassed 23,000+ by April 2026, continuing rapid growth.
Tool compatibility: Supports Claude Code、Cursor、Gemini CLI、Windsurf、OpenCode、GitHub Copilot、Kiro IDE & CLI, Codex, and other mainstream AI coding tools.
Usage prerequisites: Must be used with an AI coding Agent that supports Skill/rule files; pure chat models cannot load it directly.
Language support: Skill documents are in English Markdown, but are applicable to projects in any programming language.

Core advantages of Agent Skills

Engineering discipline solidified: Distill the "steady, accurate, ruthless" habits of senior Google engineers into a reusable process.
Full-chain coverage: All six stages from Idea to Go Live have corresponding Skills, preventing AI from skipping key steps such as specifications, testing, and review.
Evidence-driven: Each stage mandates verifiable deliverables, eliminating perfunctory output that "looks fine."
Lightweight on-demand loading: Progressive disclosure design reduces Token consumption and remains efficient under long contexts.
Cross-platform universal: Pure Markdown format, not tied to any specific IDE or model, with low migration cost.

Agent Skills project address

GitHub repository：https://github.com/addyosmani/agent-skills

Comparison of Agent Skills with similar competitors

Dimension	Agent Skills	Spec Kit	Superpowers
Core positioning	Manage AI with discipline	Define AI with documents	Guide AI with processes
Constraint method	20 composable Skills + anti-rationalization mechanism	Requirements/plans/task breakdown written as specification documents	Requirements → plan → testing → mutual review fully automated in one pipeline
Engineering source	《Software Engineering at Google》	General software engineering practices	General development process
Verification mechanism	Mandatory evidence requirements per Skill	Do things according to documents	Steps advance automatically
Agent persona	3 Specialist Personas	Usually none	Usually none
IDE compatibility	8+ platforms including Claude Code, Cursor, Gemini CLI, etc.	Depends on the specific implementation	Depends on the specific implementation
Best scenario	Serious projects requiring Google-level engineering standards constraints	Projects with vague requirements that need to be clarified before starting	Projects hoping for full-process automation and reduced manual intervention

Application scenarios of Agent Skills

Standardized coding for individual developers: Establish complete engineering discipline from requirements definition to production release for AI coding assistants, preventing models from taking shortcuts and outputting low-quality code.
Unified standards for team collaboration: Encode team code review, testing strategy, and security standards as reusable Skills, ensuring consistent output style and quality when different members use AI.
Quality control for production-grade projects: In commercial projects requiring long-term iteration, through /spec → /plan → /build → /test → /review → /ship full-chain mandatory verification, ensuring delivery reliability.
Multi-Agent parallel automated review: Before launch, simultaneously trigger the three Specialist Personas code-reviewer, test-engineer, and security-auditor, and issue three reports in parallel on code, testing, and security for comprehensive decision-making.
Legacy system migration and technical debt management: Use Skills such as deprecation-and-migration and code-simplification to safely remove zombie code and simplify overly complex modules with a "code is debt" mindset.