agent-skills: Production Engineering Workflows for AI Coding Agents
A practical guide to addyosmani/agent-skills, a Markdown skill pack that gives AI coding agents specs, plans, tests, reviews, and release gates.
Key takeaways#
agent-skillsis a workflow pack for AI coding agents, not a generic prompt collection.- It packages senior-engineering habits into repeatable skills for specification, planning, building, testing, review, simplification, and shipping.
- The repository supports Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro, and other agents that can read Markdown instructions.
What agent-skills is#
agent-skills is a production engineering skill pack created by Addy Osmani. The repository describes the project as production-grade engineering skills for AI coding agents. Its main idea is simple: AI agents often take the shortest path unless the workflow tells them exactly how to define work, plan the implementation, prove the result, and ship safely.
The pack turns those practices into structured Markdown skills. A skill is more than a block of advice. It includes steps, verification gates, exit criteria, common rationalizations, and red flags. That makes it useful for teams that want agents to behave more like careful engineers and less like autocomplete with shell access.
What it includes#
The repository includes lifecycle slash commands for common engineering phases: /spec, /plan, /build, /test, /review, /code-simplify, and /ship. Each command maps to a different moment in the delivery loop. The pack also includes skills for areas such as API design, frontend engineering, browser testing with DevTools, and progressive context loading.
The design uses progressive disclosure. Agents start with a compact SKILL.md, then load deeper references only when the task needs them. That matters because agent context is expensive. A well-structured skill should guide behavior without flooding the model with every reference on every turn.
How to install it#
For Claude Code, the documented marketplace path is:
/plugin marketplace add addyosmani/agent-skills /plugin install agent-skills@addy-agent-skills
If SSH cloning is not configured, the repository documents an HTTPS workaround:
/plugin marketplace add https://github.com/addyosmani/agent-skills.git /plugin install agent-skills@addy-agent-skills
Cursor users can copy skills into .cursor/rules/ or reference the skills directory. Gemini CLI users can install skills from the repository with gemini skills install. Other agents can use the Markdown files as system prompts, project rules, or instruction references.
When to use it#
Use agent-skills when agents are allowed to edit real code, run tests, and make product changes. It is most valuable in teams that already care about specs, test evidence, review discipline, and shipping checklists. It is less useful for quick one-off questions where a full delivery workflow would be too heavy.
Limits and review notes#
The pack does not guarantee quality by itself. Teams still need project-specific rules, CI checks, code owners, and human review for risky changes. Treat the skills as a reusable operating system for agents: they set expectations, but the repository and deployment pipeline still provide the final proof.