Boost AI Coding Quality with Agent Skills: Structured Workflows for Production‑Grade Code

This article introduces Agent Skills, an open‑source MIT‑licensed skill pack that equips AI coding agents with Google‑inspired engineering workflows, detailed command sets, and step‑by‑step integration guides for tools like Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode and GitHub Copilot.

AI Architecture Path
AI Architecture Path
AI Architecture Path
Boost AI Coding Quality with Agent Skills: Structured Workflows for Production‑Grade Code

Project Overview

Agent Skills is a production‑grade engineering skill pack for AI coding agents. It encodes senior‑engineer workflows, quality gates and Google engineering best practices such as Haller’s Law, test pyramid, change‑size limits, trunk‑based development, left‑shift testing, feature flags, and treating code as debt. The pack is MIT‑licensed and works with Claude Code, GitHub Copilot and other mainstream AI tools.

Latest release: version 0.5.0 (2026‑04‑10), 121 commits, 4 branches, 13 contributors. The repository primarily contains Shell scripts and three top‑level directories: skills (20 core skills), agents (3 role definitions), and references (4 supplemental checklists).

Core Slash Commands

/spec – Write specifications before coding.

/plan – Break work into atomic tasks.

/build – Implement one feature slice at a time.

/test – Prove functionality (red‑green‑refactor, test pyramid).

/review – Conduct five‑dimensional code review, limit changes to ~100 lines.

/code-simplify – Apply Chesterton fence principle, keep code clear.

/ship – Deploy faster, ship safer.

Skill Catalog (selected examples)

idea‑refine – Convert vague ideas into concrete PRD documents.

spec‑driven‑development – Create PRDs covering goals, commands, structure, style, tests and boundaries.

planning‑and‑task‑breakdown – Decompose specs into verifiable micro‑tasks with acceptance criteria.

incremental‑implementation – Vertical‑slice development with feature flags and rollback support.

test‑driven‑development – Enforce red‑green‑refactor and test‑pyramid ratios (80/15/5).

context‑engineering – Provide precise context (rules files, MCP integration) at the right moment.

source‑driven‑development – Base framework decisions on official docs with verification steps.

frontend‑ui‑engineering – Component architecture, design system, state management, WCAG 2.1 AA compliance.

api‑and‑interface‑design – Contract‑first design, Haller’s Law, single‑version rule, error semantics.

Quick‑Start Guides

Claude Code (recommended)

Install via the marketplace:

/plugin marketplace add addyosmani/agent-skills</code>
<code>/plugin install agent-skills@addy-agent-skills

Common SSH‑clone error: the marketplace uses SSH; if your GitHub account lacks an SSH key, cloning fails. Fix by either adding an SSH key (see GitHub documentation) or switching to HTTPS for the current terminal:

git config --global url."https://github.com/".insteadOf "[email protected]:"

Local/Dev‑environment install for customization:

git clone https://github.com/addyosmani/agent-skills.git</code>
<code>claude --plugin-dir /path/to/agent-skills

After installation, run any core slash command (e.g., /spec) to verify activation.

Cursor

Two import methods:

Copy the desired SKILL.md (e.g., idea-refine) into .cursor/rules/ and restart Cursor.

Add the entire skills directory via Cursor Settings → Rules Configuration → Add Directory.

Gemini CLI

Install directly from the repository:

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

Or install from a local clone for customization:

git clone https://github.com/addyosmani/agent-skills.git</code>
<code>gemini skills install ./agent-skills/skills/

Invoke a skill by name (e.g., idea-refine) or add it to GEMINI.md for persistent context.

Windsurf

Copy all SKILL.md files from the skills directory into the Windsurf rule configuration file (e.g., windsurf.config.js) and restart.

OpenCode

Load skills with the built‑in skill tool: skill load ./agent-skills/skills/ Run a skill via skill run <skill‑name> or let the agent auto‑trigger (e.g., during API design).

GitHub Copilot

Configure three role agents (code‑reviewer, test‑engineer, security‑auditor) from the agents directory, then place all SKILL.md contents into .github/copilot-instructions.md. Commit the file; Copilot will follow the defined workflows automatically.

Kiro IDE & CLI

Two storage options:

Project‑level: create .kiro/skills/ in the project root and copy the skills folder there.

Global‑level: copy the skills folder into the global Kiro config directory ( ~/.kiro/skills on macOS/Linux or C:\Users\<em>username</em>\.kiroskills on Windows) and restart Kiro.

Kiro also supports an AGENTS.md file for role definitions.

General Notes

Restart each tool after installation to ensure skills are loaded.

If a skill does not trigger, verify the file path and consult tool logs.

Update the skill set with git pull to fetch the latest changes.

All skills are released under the MIT license; modifications must preserve the original directory structure for tool compatibility.

Further Reading

https://github.com/addyosmani/agent-skills
AIsoftware developmentbest practicesopen-sourceAgent Skills
AI Architecture Path
Written by

AI Architecture Path

Focused on AI open-source practice, sharing AI news, tools, technologies, learning resources, and GitHub projects.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.