Agent Skills — Production-Grade Engineering Workflows for AI Coding Agents
Agent Skills — Production-Grade Engineering Workflows for AI Coding Agents
AI coding agents are incredibly productive, but they default to the shortest path — skipping specs, tests, security reviews, and the engineering practices that make production software reliable. Agent Skills is an open-source skill pack (52K+ ⭐ on GitHub) by Addy Osmani that fixes this by encoding senior engineering workflows into structured, verifiable skills that any AI agent can follow.
Whether you use Claude Code, Cursor, Gemini CLI, Windsurf, or plain system prompts, Agent Skills brings the discipline of Google's engineering culture — spec-driven development, test-driven development, code review standards, and trunk-based git workflows — directly into your agent's decision loop.
Architecture
The Agent Skills architecture follows a simple but powerful pattern. Each skill is a single Markdown file with a consistent internal structure: frontmatter defining its name and purpose, step-by-step process instructions, anti-rationalization tables that block common shortcuts, and verification gates the agent must satisfy before declaring success.
The system is organized into six lifecycle phases (Define → Plan → Build → Verify → Review → Ship), each containing multiple skills. A meta-skill (using-agent-skills) determines which skill applies based on the current task. Slash commands provide quick access, and the system also auto-detects context — designing an API triggers api-and-interface-design, building UI triggers frontend-ui-engineering, and so on.
When needed, skills pull in supplementary reference checklists (testing patterns, security checklist, performance checklist, accessibility) and can dispatch to 4 specialist agent personas for targeted reviews.
What Is Agent Skills?
Agent Skills is a collection of 24 structured skills (23 lifecycle skills + 1 meta-skill) packaged as plain Markdown files that AI coding agents can consume. Each skill is a complete workflow with:
- A clear frontmatter (name, description, use conditions)
- A step-by-step process the agent follows
- Anti-rationalization tables — common excuses agents use to skip steps, with documented counter-arguments
- Verification gates — evidence requirements the agent must meet before proceeding
- Red flags — signs something is wrong and needs attention
The skills cover the full development lifecycle: Define → Plan → Build → Verify → Review → Ship.
Why Is It Trending?
Agent Skills has exploded in popularity (52K+ stars since February 2026) because it solves a real problem: AI agents produce a lot of code, but the quality is inconsistent. Developers are discovering that giving agents structured workflows — not just prompts — dramatically improves output quality, test coverage, and security posture.
The pack also integrates with 7 slash commands (/spec, /plan, /build, /test, /review, /code-simplify, /ship) that map directly to development phases, and an auto mode (/build auto) that generates a plan and implements every task autonomously with verification at every step.
Key Skills
Meta — Discover Which Skill Applies
- using-agent-skills: Maps incoming work to the right skill; defines shared operating rules
Define — Clarify What to Build
- interview-me: One-question-at-a-time interview to extract real requirements
- idea-refine: Divergent/convergent thinking for vague concepts
- spec-driven-development: Full PRD before any code — objectives, structure, testing boundaries
Plan — Break It Down
- planning-and-task-breakdown: Decompose specs into small, verifiable tasks with acceptance criteria
Build — Write the Code
- incremental-implementation: Thin vertical slices — implement, test, verify, commit
- test-driven-development: Red-Green-Refactor with test pyramid (80/15/5)
- context-engineering: Feed agents the right context at the right time
- source-driven-development: Ground every framework decision in official documentation
- doubt-driven-development: Adversarial fresh-context review of non-trivial decisions
- frontend-ui-engineering: Component architecture, design systems, accessibility
- api-and-interface-design: Contract-first design, Hyrum's Law, error semantics
Verify — Prove It Works
- browser-testing-with-devtools: Live runtime data — DOM, console, network, performance
- debugging-and-error-recovery: Five-step triage: reproduce, localize, reduce, fix, guard
Review — Quality Gates Before Merge
- code-review-and-quality: Five-axis review, change sizing (~100 lines)
- code-simplification: Chesterton's Fence, Rule of 500, reduce complexity
- security-and-hardening: OWASP Top 10, auth patterns, secrets management
- performance-optimization: Measure-first — Core Web Vitals, bundle analysis
Ship — Deploy With Confidence
- git-workflow-and-versioning: Trunk-based development, atomic commits
- ci-cd-and-automation: Shift Left, feature flags, quality gates
- deprecation-and-migration: Code-as-liability, migration patterns
- documentation-and-adrs: Architecture Decision Records
- observability-and-instrumentation: RED metrics, OpenTelemetry, structured logging
- shipping-and-launch: Pre-launch checklists, staged rollouts, rollback procedures
Prerequisites
- An AI coding agent (Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, Copilot, or any agent that accepts Markdown instructions)
- Git (to clone the repository)
- (Optional) One of the supported IDE plugins or CLI tools
Installation
Claude Code (Recommended)
# Via marketplace
/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills
# Or if you get SSH errors, use HTTPS
/plugin marketplace add https://github.com/addyosmani/agent-skills.git
/plugin install agent-skills@addy-agent-skills
# Local install
git clone https://github.com/addyosmani/agent-skills.git
claude --plugin-dir /path/to/agent-skills
Other Platforms
| Platform | Installation |
|---|---|
| Cursor | Copy any SKILL.md into .cursor/rules/ |
| Antigravity CLI | agy plugin install https://github.com/addyosmani/agent-skills.git |
| Gemini CLI | gemini skills install ... --path skills |
| Windsurf | Add skill contents to Windsurf rules config |
| OpenCode | Uses agent-driven skill execution via AGENTS.md |
| GitHub Copilot | Use agents/ as Copilot personas |
| Codex / Any agent | Plain Markdown — works with any instruction system |
Usage
Starting a New Feature
# 1. Define what to build
/spec
→ Writes a full PRD covering objectives, structure, code style, testing boundaries
# 2. Plan the implementation
/plan
→ Decomposes the spec into small, verifiable tasks
# 3. Build incrementally
/build
→ Implements one slice at a time, testing and verifying each
# 4. Run tests
/test
→ Proves everything works with evidence
# 5. Review the code
/review
→ Five-axis review before merge
Auto Build Mode
/build auto
→ Generates the plan and implements every task in a single approved pass
→ Each task is still test-driven and committed individually
→ Pauses on failures or risky steps
Architecture Details
The internal structure of every skill is consistent:
| Component | Purpose |
|---|---|
| Frontmatter | Name, description, use conditions |
| Overview | What the skill does and why |
| When to Use | Triggering conditions |
| Process | Step-by-step workflow the agent follows |
| Anti-rationalization | Tables of common excuses + documented counter-arguments |
| Verification | Evidence requirements — tests passing, build output, runtime data |
| Red Flags | Signs something is wrong |
Key design principles:
- Process, not prose. Skills are workflows agents follow, not reference docs they read.
- Anti-rationalization. Every skill includes a table of common excuses agents use to skip steps with documented counter-arguments.
- Verification is non-negotiable. "Seems right" is never sufficient — every skill ends with evidence requirements.
- Progressive disclosure. The
SKILL.mdis the entry point; supporting references load only when needed, keeping token usage minimal.
Resources
- GitHub: github.com/addyosmani/agent-skills
- Skill Anatomy: docs/skill-anatomy.md
- Getting Started: docs/getting-started.md
- License: MIT
- Author: Addy Osmani — VP of Engineering @ Google Chrome