Addy Osmani, a Google Chrome architect and well-known figure in the web development community, has released agent-skills -- a collection of 25 structured engineering workflows designed to bring senior-level software engineering discipline to AI coding agents. The project reached GitHub Trending's number-one spot on September 17 with 658 new stars.
The core insight is simple but powerful: AI coding agents default to the shortest path, which often means skipping specifications, tests, security reviews, and the practices that make software reliable. Agent-skills encodes these practices into workflows that agents follow consistently.
The pack covers the full development lifecycle. Define skills include spec-driven development, interview-me (for extracting true requirements one question at a time), and constraint-driven development for establishing quality bars. Plan skills handle task breakdown and decomposition. Build skills cover incremental implementation, test-driven development, and frontend UI engineering. Verify skills include browser testing with DevTools and debugging recovery workflows. Review skills cover code review, security hardening, and performance optimization. Ship skills handle git workflows, CI/CD, and production deployment.
Each skill is structured with frontmatter (name, description, trigger conditions), a step-by-step process, an anti-rationalization table (documenting common excuses agents use to skip steps with counter-arguments), red flags, and evidence-based verification requirements.
The project works across 70+ agent platforms including Claude Code, Cursor, Codex, Copilot, Cline, Gemini CLI, and Windsurf. It installs through a universal npx command or native integrations per platform. A slash-command interface provides nine lifecycle commands (/spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify, /ship) that map to the development phases.
A notable feature is /build auto, which generates a plan and implements every task in a single approved pass -- removing the human stepping between tasks while keeping verification gates intact. Skills also activate automatically based on context: designing an API triggers api-and-interface-design, building UI triggers frontend-ui-engineering.
The project is MIT-licensed and draws on engineering practices from Google's engineering culture, including concepts from Software Engineering at Google such as Hyrum's Law, the Beyonce Rule, Chesterton's Fence, and trunk-based development.
