Addy Osmani's agent-skills β€” Senior Engineering Practices as SKILL.md Files

Addy Osmani's agent-skills β€” Senior Engineering Practices as SKILL.md Files

Addy Osmani (Google Chrome engineering lead) packaged senior engineering practices into 20 core skills that AI coding agents can follow. The insight: AI agents naturally take shortcuts β€” skipping specs, tests, and security reviews. These skills enforce the discipline that experienced engineers bring to every project. Its rapid GitHub adoption has made it a reference implementation for encoding engineering culture into agent workflows.

*Source: GitHub - addyosmani/agent-skills Weibo discussion by ι©¬εŠ›AIε’Œε•†δΈšζ€η»΄*

The 7 Slash Commands

Each command maps to a phase of the software development lifecycle:

Command Phase What It Does
/spec Define Spec before code β€” clarify requirements first
/plan Plan Break work into small, atomic tasks
/build Build Implement one slice at a time
/test Verify Tests are proof, not decoration
/review Review Improve code health before merge
/code-simplify Simplify Clarity over cleverness
/ship Ship Faster is safer β€” deploy to production

Six Development Phases, 20 Skills

Phase 1: DEFINE
β”œβ”€β”€ idea-refine
└── spec-driven-development

Phase 2: PLAN
└── planning-and-task-breakdown

Phase 3: BUILD
β”œβ”€β”€ incremental-implementation
β”œβ”€β”€ test-driven-development
β”œβ”€β”€ frontend-ui-engineering
β”œβ”€β”€ api-and-interface-design
β”œβ”€β”€ context-engineering
└── source-driven-development

Phase 4: VERIFY
β”œβ”€β”€ browser-testing-with-devtools
└── debugging-and-error-recovery

Phase 5: REVIEW
β”œβ”€β”€ code-review-and-quality
β”œβ”€β”€ code-simplification
β”œβ”€β”€ security-and-hardening
└── performance-optimization

Phase 6: SHIP
β”œβ”€β”€ git-workflow-and-versioning
β”œβ”€β”€ ci-cd-and-automation
β”œβ”€β”€ deprecation-and-migration
β”œβ”€β”€ documentation-and-adrs
└── shipping-and-launch

Plus 3 specialist agent personas (code-reviewer, test-engineer, security-auditor) and 4 reference checklists.

What Makes These Skills Special

Each SKILL.md follows a standardized structure with an β€œanti-rationalization table” β€” anticipating shortcuts agents try to take:

Agent Excuse Skill Response
β€œI’ll add tests later” Tests are proof. No proof = not done.
β€œThis is just a small change” Small changes cause big outages.
β€œSeems right” Seems is not evidence. Run it.
β€œI don’t need a spec for this” You don’t know what you’re building without one.

Google Engineering Culture Encoded

The skills draw heavily from Software Engineering at Google:

  • Hyrum’s Law in API design β€” backward compatibility matters
  • The Beyonce Rule in testing β€” β€œIf you liked it, you should’ve put a test on it”
  • Chesterton’s Fence in code simplification β€” don’t remove what you don’t understand
  • Change sizing norms (~100 lines) in code review
  • Trunk-based development in git workflow
  • Shift Left principle in CI/CD β€” catch problems early

Installation

# Claude Code
/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

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

# Cursor / Windsurf / Copilot
# Follow the repo's tool-specific setup docs for rules, agent definitions, or instructions files.

# Manual (any agent)
git clone https://github.com/addyosmani/agent-skills.git

Skills are plain Markdown β€” any agent accepting system prompts can use them.

How LearnAI Team Could Use This

  • Teaching software engineering β€” the 6-phase lifecycle is a pedagogical framework itself
  • Improving AI-assisted student projects β€” install the skills to enforce best practices
  • Studying skill design patterns β€” the anti-rationalization table is a novel concept worth teaching
  • Building custom skills β€” use this repo as a template for encoding domain expertise

Real-World Use Cases

  • Engineering teams β€” enforce consistent practices across all AI-assisted development
  • Open source projects β€” standardize contribution quality regardless of agent used
  • Junior developers β€” learn senior-level practices by reading the skill files
  • Code review β€” the reviewer agent persona provides structured, opinionated feedback <!– REVIEW-TODO: [source_links] Weibo source link is generic (https://weibo.com), not a specific post URL β€” find actual post or remove –>