Google's Open-Source 'Agent Skills' Playbook: Structured AI Workflows for Senior Engineer Results
Addy Osmani has released a free, MIT-licensed set of 23 structured workflows that teach AI coding agents to follow senior engineering practices with specific commands and built-in quality gates.
Practical Summary
The 'Agent Skills' playbook provides a structured framework for using AI coding agents (like Claude Code, Cursor, GitHub Copilot) through a defined sequence of commands: /spec to define requirements, /plan to break tasks into atomic steps, /build to implement incrementally, /test to verify with evidence, /review for a five-axis code review, /code-simplify to prioritize clarity, and /ship for confident deployment. Each workflow includes an 'anti-rationalization table' to counter common excuses agents use to skip steps, enforcing discipline. It is based on Google's internal engineering practices and is open-source under the MIT license.
Why It Matters
For teams integrating AI into development, this offers a concrete, reproducible workflow that moves beyond ad-hoc prompting. It provides a standardized method to enforce engineering best practices (like testing and review) on AI agents, potentially improving code quality, reducing technical debt, and making AI-assisted development more predictable and auditable. The open-source nature and backing from a major tech company's internal practices give it credibility for immediate experimentation and adoption.
Google's Open-Source 'Agent Skills' Playbook: Structured AI Workflows for Senior Engineer Results
The 'Agent Skills' playbook provides a structured framework for using AI coding agents (like Claude Code, Cursor, GitHub Copilot) through a defined sequence of commands: /spec to define requirements, /plan to break tasks into atomic steps, /build to implement incrementally, /test to verify with evidence, /review for a five-axis code review, /code-simplify to prioritize clarity, and /ship for confident deployment. Each workflow includes an 'anti-rationalization table' to counter common excuses agents use to skip steps, enforcing discipline. It is based on Google's internal engineering practices and is open-source under the MIT license.
For teams integrating AI into development, this offers a concrete, reproducible workflow that moves beyond ad-hoc prompting. It provides a standardized method to enforce engineering best practices (like testing and review) on AI agents, potentially improving code quality, reducing technical debt, and making AI-assisted development more predictable and auditable. The open-source nature and backing from a major tech company's internal practices give it credibility for immediate experimentation and adoption.