Addy Osmani, Engineering Lead of the Google Chrome Team, recently open-sourced a skill pack called Agent Skills. This isn’t a new concept, but it might be the most comprehensive one available right now.

Simply put, it breaks down the workflow of senior engineers into 19 skills, each being a structured workflow with steps, checkpoints, and exit criteria.

The skills are organized around six stages of software development, corresponding to seven slash commands:

```

DEFINE → PLAN → BUILD → VERIFY → REVIEW → SHIP

/spec /plan /build /test /review /ship

```

Plus a `/code-simplify` for code simplification. You don’t need to remember the names of the 19 skills—just type the commands.


In addition to the skills themselves, it also includes three preconfigured Agent roles:

- **code-reviewer**: Conducts code reviews from a Staff Engineer’s perspective, with the standard being "Would a Staff Engineer approve this PR?"

- **test-engineer**: Checks test strategies and coverage from a QA perspective

- **security-auditor**: Performs vulnerability detection and threat modeling from a security engineer’s perspective

And four reference checklists: Test Patterns, Security Checks, Performance Checks, and Accessibility Checks.


Put simply, this set tells AI: Don’t start writing code right away. First, clarify what needs to be done (DEFINE), figure out how to break down the task (PLAN), write a little and verify a little (BUILD/VERIFY), get someone to review before submission (REVIEW), and finally release (SHIP).

It currently supports mainstream programming agents like Claude Code, Cursor, Windsurf, GitHub Copilot, and Gemini CLI. It’s under the MIT license and in pure Markdown format.

There’s full documentation on GitHub—feel free to check it out if you’re interested.

-----------

**Personal Take:**

There’s no shortage of AI programming tools now, but what’s lacking is the constraint to "work to senior engineer standards." The default behavior of most agents is to "write code as fast as possible," but in real engineering practice, those seemingly "slow" steps—writing specs, breaking down tasks, code reviews, security checks—are exactly what distinguish production-grade code from prototype code.

The value of this skill pack doesn’t lie in how new the skills are, but in that it turns Google’s internal engineering practices—such as Hyrum's Law, Beyoncé Rule, Chesterton's Fence, trunk-based development, and Shift Left—into workflows that AI can execute directly.

You don’t have to use it, but you should at least see how others turn "experience" into "processes."