38.9K stars, with 10,738 new ones in just a week. Addy Osmani—an engineer on the Google Chrome team—created a project called agent-skills.
The core idea is straightforward: Senior engineers follow a set of implicit workflows when writing code. This project encodes those habits into skills that AI Agents can automatically execute.
Seven Slash Commands Covering the Full Lifecycle from Idea to Deployment
This project doesn't mess around. Seven commands correspond to seven stages of the development process:
| What you're doing | Command | Core Principle |
|---|---|---|
| Defining what to do | /spec |
Write the spec before writing code |
| Planning how to do it | /plan |
Small-grained atomic tasks |
| Building incrementally | /build |
One slice at a time |
| Verifying feasibility | /test |
Tests are proof |
| Reviewing before merge | /review |
Improve code health |
| Simplifying code | /code-simplify |
Clarity over cleverness |
| Shipping to production | /ship |
Faster is safer |
Each command automatically activates its corresponding skill. For example, designing an API triggers api-and-interface-design, while building a UI triggers frontend-ui-engineering.
Why This Is Worth Writing About
It's not because the project is technically groundbreaking—it's essentially a collection of SKILL.md files—but because it chose a smart entry point.
Most AI coding tools on the market today are solving the problem of "how an Agent writes code." agent-skills tackles a different question: "In what order and by what standards should an Agent write code?"
The difference between these two is significant. The former is a capability issue; the latter is a methodology issue. Addy Osmani is addressing the latter—encoding Google-level engineering discipline into rules that Agents can understand and execute.
Take a typical example. Most people using Claude Code follow this flow: describe requirements → Agent starts coding immediately → code is generated → manual testing → issues found → fix. agent-skills interrupts this flow and enforces a pipeline: /spec → /plan → /build → /test → /review → /ship. Each step has a corresponding quality gate.
This might sound dogmatic. But for team development, dogma can be a good thing—it ensures engineers of varying skill levels produce code of consistent quality.
It Supports More Than Just Claude
The project supports Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, and GitHub Copilot. Each Agent has its own installation guide.
This means Addy Osmani isn't just writing a plugin for a specific tool; he's defining a cross-Agent engineering skill standard. If widely adopted, this standard could narrow the quality gap between code generated by different AI coding tools—because the underlying workflow discipline remains the same.
My Reservations
At its core, this project is a collection of "best practices." But best practices have a flaw: they are common sense for experts, but can be too abstract for beginners.
For instance, what are the guiding principles of the api-and-interface-design skill? Is there a concrete checklist? If it's just advice at the level of "remember X, Y, Z when designing RESTful APIs," its value to an AI Agent is limited—since the Agent likely already knows these things.
True value lies in skills that focus on easily overlooked but high-impact details: for example, "always update the OpenAPI spec when modifying an API," or "check license compatibility when adding new dependencies." These are exactly where the gap between experts and novices lies.
I took a look at the skills directory in the repo, and there are indeed some skills that cover these details. However, I haven't verified their quality one by one yet.
Conclusion
agent-skills isn't a technological breakthrough; it's the productization of engineering discipline. If your team uses AI coding tools, this project is worth spending 30 minutes exploring—especially the /spec and /review commands, which likely offer the highest ROI.
Primary Sources:
- addyosmani/agent-skills — Official repository
- GitHub Trending weekly — This week's trending rankings