Google’s Agent Skills Hits 23K+ Stars – Adding Engineering Discipline to AI Coding

AI programming tools have advanced rapidly, but their powerful models often prioritize finishing a task over long‑term maintainability and engineering consistency, leaving teams uncertain whether generated code can be reliably delivered and iterated.

To address this, Addy Osmani, a manager on Google’s Gemini team, open‑sourced Agent Skills . The project quickly surpassed 23,000 GitHub stars and continues to grow.

The core idea is not to make the model smarter, but to encapsulate senior engineers’ mature workflows and development standards into reusable skills. Many of these rules stem from the Software Engineering at Google methodology.

Agent Skills consists of:

20 individual Skills

7 slash commands

3 Agent personas

These components cover the six stages of a full software lifecycle: definition, planning, building, verification, review, and release.

In Claude Code the skills are invoked with the following slash commands: /spec – clarify requirements /plan – break down tasks /build – incremental implementation /test – run tests /review – conduct code review /ship – prepare for production

Agent Skills also defines three parallel collaboration roles:

code-reviewer

test-engineer

security-auditor

During the /ship stage these roles can simultaneously output a code review, test results, and a security report, then combine them into a single go‑or‑no‑go decision.

Comparison with similar projects:

Spec Kit : writes a specification document first and uses it to constrain AI execution.

Superpowers : strings together requirements, planning, testing, and peer review into an automated pipeline.

Agent Skills : breaks senior‑engineer habits into composable skills and emphasizes execution discipline.

In short: Spec Kit governs AI with documentation, Superpowers with workflow automation, and Agent Skills with disciplined execution.

Quick start in Claude Code requires two commands:

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

For Cursor users, copy the provided SKILL.md file into the .cursor/rules/ directory. The framework also includes integration guides for Gemini CLI, Windsurf, GitHub Copilot, Codex, and other tools.

FAQ:

Q1: Does Agent Skills increase the model’s intelligence? A: No; it adds an engineering‑discipline layer to keep AI from taking shortcuts and to make its actions auditable.

Q2: Can it be used together with Spec Kit or Superpowers? A: Yes; many teams combine documentation constraints, automated pipelines, and skill‑based discipline as needed.

Q3: Is it limited to Claude Code? A: No; it also supports Cursor, Gemini CLI, Windsurf, GitHub Copilot, Codex, etc.

Q4: When is it worth adopting? A: When a team starts caring about stable production releases and reusable collaboration processes.

Although AI model capabilities keep improving, the rapid growth of “engineering constraint layers” like Spec Kit, Superpowers, and Agent Skills shows that teams value not just instant code generation but sustainable, high‑quality delivery.