Production-Grade Engineering Skills (by Addy Osmani)
"Production-grade engineering skills for AI coding agents." — Created by Addy Osmani, Lead Engineer at Google Chrome.
A comprehensive collection of 25 battle-tested engineering skills designed to make AI coding agents (Claude Code, Cursor, Codex, OpenCode, Antigravity) think, plan, and execute like senior software engineers.
⚡ How to Install on Any Machine
# 1. In Claude Code, install the official plugin from marketplace:
/plugin marketplace add addyosmani/agent-skills && /plugin install agent-skills@agent-skills
(Or for any other agent like Cursor, Codex, or Antigravity, copy AGENTS.md and the desired skills from skills/ into your workspace).
🧭 The 25 Production Engineering Skills
| Category | Skill | Purpose & Trigger |
|---|---|---|
| Architecture & Design | api-and-interface-design | Designs robust, ergonomic APIs, function signatures, and interface contracts. |
spec-driven-development | Drafts rigorous engineering specifications before writing code. | |
constraint-driven-development | Enforces hard performance, dependency, and architecture boundaries. | |
documentation-and-adrs | Generates clear documentation and Architectural Decision Records. | |
deprecation-and-migration | Manages graceful legacy code retirement and safe migration pathways. | |
| Planning & Execution | planning-and-task-breakdown | Breaks large features into ordered, verifiable, bite-sized tasks. |
incremental-implementation | Ships changes in small, verifiable steps without breaking the build. | |
test-driven-development | Writes failing tests first, then minimal code to make them pass (Red-Green-Refactor). | |
git-workflow-and-versioning | Enforces clean Git hygiene, branch strategies, and semantic versioning. | |
| Quality & Debugging | code-review-and-quality | Conducts senior-level reviews focusing on edge cases, safety, and correctness. |
code-simplification | Refactors messy code into clear, maintainable, idiomatic implementations. | |
debugging-and-error-recovery | Systematically isolates bugs, traces root causes, and fixes regressions. | |
doubt-driven-development | Pressure-tests assumptions, edge cases, and failure modes under stress. | |
| Security & Performance | security-and-hardening | Audits and fixes vulnerabilities, injection risks, auth flaws, and data leaks. |
performance-optimization | Diagnoses bottlenecks, optimizes hot paths, memory allocations, and network I/O. | |
observability-and-instrumentation | Implements telemetry, logging, metrics, and distributed tracing. | |
| Frontend & Browser | frontend-ui-engineering | Builds responsive, accessible (a11y), performant web interfaces. |
browser-testing-with-devtools | Drives headless browser debugging and automated UI testing via DevTools. | |
| Workflow & Discovery | interview-me | Interviews the developer interactively to clarify ambiguous requirements. |
idea-refine | Refines raw concepts and feature ideas into structured engineering proposals. | |
context-engineering | Curates and packages optimal context windows for downstream agent sessions. | |
source-driven-development | Anchors all architectural decisions against verified upstream source code. | |
ci-cd-and-automation | Automates CI/CD pipelines, lint gates, and release automation. | |
shipping-and-launch | Pre-flight checklists, release verification, and zero-downtime deploy procedures. | |
using-agent-skills | Guides agents on how to discover and activate the right skill for any prompt. |
🛠️ Usage Across AI Assistants
1. Claude Code
Install via marketplace:
/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@agent-skills
2. Antigravity / Cursor / Windsurf / OpenCode
Link or copy AGENTS.md and any individual skill from skills/ into your project root. The agent will automatically map prompts (e.g. "review this PR" or "write a spec") to the matching skill.
📦 Files in this Skill Bundle
AGENTS.md— Intent-to-skill mapping rules for AI agents.CLAUDE.md— Claude Code environment and plugin rules.plugin.json— Official plugin manifest.skills/— All 25 complete, unmodified skills, scripts, and references.