Agents are fast at writing code and equally fast at writing the wrong code. addyosmani/agent-skills is a set of 25 skills that put a spec, a plan and tests between your idea and the diff. It works with Claude Code, Cursor and Codex, and installs with one command. Here is the short path from a vague feature idea to a tested change.

The problem it fixes

Left alone, a coding agent jumps straight to code, skips the spec and hands you a large change that is hard to verify. The skills push the agent through the same order a careful engineer would use: define what to build, break it into small tasks, build one slice at a time, test, then ship.

Step 1: install the skills 

npx skills add addyosmani/agent-skills

This installs all 25 skills into your agent. You need Node.js for npx skills. If you only want one skill, add --skill <name> to the same command, for example --skill interview-me. A per-skill install copies only that skill, not the shared checklists, so install the whole repo if you want those.

Step 2: write the spec first

Run:

/spec

Describe what you want to build. The spec comes before any code. If your idea is still fuzzy, install and use the interview-me skill first, so the agent asks requirement questions one at a time instead of silently guessing.

Step 3: plan small tasks 

/plan

The plan breaks the spec into small, atomic tasks. Read it. This is the cheapest moment to catch a wrong assumption, because nothing has been built yet.

Step 4: build one slice at a time 

/build

This implements one slice at a time, test-driven. If you trust the plan, /build auto implements every task after a single plan approval. Be clear about what that means: it removes the stepping between tasks, not the verification. It still pauses on failures and risky steps.

Step 5: prove it works 

/test

Run it before you merge. When you want a review pass, the repo also has /review (in Claude Code use /agent-skills:review, because a built-in command shares the name) and /code-simplify for anything that came out too clever.

A prompt you can paste

"Using addyosmani/agent-skills, run /spec for , then /plan, and show me the plan. After I approve, implement it with /build one slice at a time, test-driven, committing each task, and stop on any failing test."

Tips

  • Keep the first feature small. You are learning how the agent behaves under the workflow.
  • Approve the plan in writing. If a task looks big, ask for it to be split.
  • Commit after each slice so a bad one is easy to revert.
  • This walkthrough follows the repo README and has not been run end to end by us, so check the repo for newer instructions.

More

Step-by-step card: Go from idea to tested feature with /spec, /plan and /build. Everything else people do with this repo: agent-skills use cases. For the wider category, see AI use cases for Cursor.

Source repo: addyosmani/agent-skills.

Want the setup commands for 50 AI repos people actually ship with? Get the free AI Repos That Actually Ship pack: https://dedupedai.com/ship-pack/?utm_source=devto&utm_medium=article&utm_campaign=shippack

Get one email every Monday with the week's hottest AI repos and exactly how to use them: subscribe free at dedupedai.com