Google's Open-Sourced AI Engineering Playbook: 7 Commands for Structured Development
A set of reproducible, quality-gated commands based on Google's internal practices provides a practical workflow for AI-assisted software development.
Practical Summary
Addy Osmani, a Director at Google, has open-sourced 'Agent Skills,' a set of 23 production-grade skills that teach AI coding agents to follow structured engineering workflows. The system is centered on 7 commands that map directly to the software development lifecycle: /spec (define requirements), /plan (break into tasks), /build (implement incrementally), /test (verify with evidence), /review (five-axis code review), /code-simplify (prioritize clarity), and /ship (deploy). The workflows include built-in 'anti-rationalization tables' to counter common AI excuses for skipping steps. The system is compatible with major AI coding tools like Claude Code, Cursor, and GitHub Copilot.
Why It Matters
This provides a concrete, trackable workflow for teams using AI coding assistants. By imposing structure and verification gates (like testing and multi-axis review), it can directly impact code quality, reduce technical debt, and improve development efficiency. The compatibility with multiple tools makes it a practical, vendor-agnostic method for optimizing AI-assisted development workflows and measuring their impact on project velocity and code reliability.
Google's Open-Sourced AI Engineering Playbook: 7 Commands for Structured Development
Addy Osmani, a Director at Google, has open-sourced 'Agent Skills,' a set of 23 production-grade skills that teach AI coding agents to follow structured engineering workflows. The system is centered on 7 commands that map directly to the software development lifecycle: /spec (define requirements), /plan (break into tasks), /build (implement incrementally), /test (verify with evidence), /review (five-axis code review), /code-simplify (prioritize clarity), and /ship (deploy). The workflows include built-in 'anti-rationalization tables' to counter common AI excuses for skipping steps. The system is compatible with major AI coding tools like Claude Code, Cursor, and GitHub Copilot.
This provides a concrete, trackable workflow for teams using AI coding assistants. By imposing structure and verification gates (like testing and multi-axis review), it can directly impact code quality, reduce technical debt, and improve development efficiency. The compatibility with multiple tools makes it a practical, vendor-agnostic method for optimizing AI-assisted development workflows and measuring their impact on project velocity and code reliability.