agent-skills: 19 Production-Grade Skills That Make AI Coding Agents Work Like Senior Engineers

AI coding agents are great at generating code. They're terrible at following engineering processes.

No spec? Start coding anyway. No tests? Ship it. No security review? Move on. AI agents default to the shortest path, skipping the very steps that separate production-ready code from technical debt.

Addy Osmani (ex-Google Chrome DevRel, now at Anthropic) released agent-skills — 19 production-grade engineering skills that force AI coding agents to work like senior engineers. 8,600+ GitHub stars and growing fast.

The 6-Stage Development Lifecycle

agent-skills maps the entire software development lifecycle into six stages:

DEFINE → PLAN → BUILD → VERIFY → REVIEW → SHIP

Each stage contains specific skills:

Stage	Skills
DEFINE	idea-refine, spec-driven-development
PLAN	planning-and-task-breakdown
BUILD	incremental-implementation, TDD, context-engineering, frontend-ui, API design
VERIFY	browser-testing, debugging-and-error-recovery
REVIEW	code-review, code-simplification, security, performance
SHIP	git-workflow, CI/CD, deprecation, documentation, shipping

Why This Matters: 4 Design Principles

1. Process, Not Prose

Each skill is a workflow, not a reference document. The agent follows steps with defined inputs and outputs. Can't skip to the next step without completing the current one.

2. Anti-Rationalization

"I'll add tests later" gets blocked. Each skill contains a counter-argument table for common shortcuts. The agent can't rationalize skipping engineering standards.

3. Non-Negotiable Verification

Every skill ends with evidence requirements. "Seems fine" is not a passing grade. You need test results, benchmarks, or security scan reports.

4. Progressive Disclosure

SKILL.md is the entry point. Detailed references load only when needed, efficiently managing context window tokens.

Google Engineering Culture, Codified

Osmani brought Google's engineering principles into AI agent workflows:

• Hyrum's Law → API design skill
• Beyonce Rule → Test culture enforcement
• Chesterton's Fence → Code simplification guards
• Trunk-based development → Git workflow skill
• Shift Left + Feature Flags → CI/CD automation

What's Included

• 19 skills covering the full development lifecycle
• 7 slash commands: /spec, /plan, /build, /test, /review, /code-simplify, /ship
• 3 agent personas: code-reviewer (staff engineer), test-engineer (QA), security-auditor
• 4 reference checklists: testing, security, performance, accessibility

Quick Install

Claude Code:

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

Cursor:
Copy SKILL.md files to .cursor/rules/

Gemini CLI:

gemini skills install https://github.com/addyosmani/agent-skills.git --path skills

Key Takeaway

AI coding agents are tools. Without engineering discipline, they generate technical debt at machine speed. agent-skills is the structural guardrail that makes them follow the same standards human senior engineers use.

GitHub: addyosmani/agent-skills (MIT License, 8,600+ stars)

---

What engineering workflows do you enforce on your AI coding agents? I'd love to hear about your setups in the comments.