AI Coding Agent Skills Guide

This page is a practical catalog of coding-agent skills, covering where to get each one and how to invoke it. Every card lists its source (bundled official or a GitHub link), how to call it, and when to use it, so you can adopt anything that catches your eye right away.

If you want to compare which approach fits your workflow first, see the cross-service Skill Pros & Cons page.

What is SKILL.md

SKILL.md is a "mission brief" for an AI agent. It consists of YAML front matter and a Markdown body that define the name, description, and steps.

my-skill/
├── SKILL.md          # required: metadata + steps
├── scripts/          # optional: shell scripts
├── references/       # optional: reference docs
└── agents/           # optional: agent settings

It plays a different role from always-on instruction files. CLAUDE.md and AGENTS.md are rules applied to a repository at all times, while SKILL.md is a reusable workflow the agent picks per task.

Recommended skills by tool

Switch tools with the tabs below to compare ready-to-use skills, each with its source, invocation, and when to use it.

Claude Code's official skills need no install and are called by a slash command or plain language. Community skills are cloned and placed. Invocation differs by surface: in Claude Code (CLI / IDE) you call them explicitly with a slash like /code-review, while the desktop and web apps (claude.ai) trigger the same .claude/skills skills through natural language.

/code-review

An official skill that reviews PR diffs across bugs, security, and quality.

WhenChecking a diff before commit or merge

SourceOfficial, no install

UsageType /code-review

/deep-research

An official skill that researches the web and generates a cited report.

WhenScoping tech choices or market research

SourceOfficial, no install

Usage/deep-research your topic

/security-review

An official skill that audits code for security and surfaces vulnerabilities.

WhenA vulnerability check before release

SourceOfficial, no install

UsageType /security-review

Official document skills

Official skills to create and manipulate PDF, Word, PowerPoint, and Excel (pdf / docx / pptx / xlsx).

WhenGenerating routine documents and extracting content

SourceOfficial, no install

UsageTrigger automatically by stating intent like "make a PDF"

awesome-claude-skills

A community curated list of high-quality skills. Good for a first step.

WhenSurveying what skills exist

Sourcetravisvn/awesome-claude-skills

UsagePick a skill repo from the list and clone it

mcp-builder

An official skill that scaffolds an MCP server following best practices. Available via the anthropics/skills plugin.

WhenImplementing your own tool as an MCP server

Sourceanthropics/skills

Usage/plugin marketplace add anthropics/skills, then state your intent

web-artifacts-builder

An official skill that builds multi-component HTML artifacts for claude.ai.

WhenBuilding a working demo or tool in a single HTML file

Sourceanthropics/skills

UsageInstall the same plugin and say "build an artifact"

How to install

The steps depend on where the skill comes from.

Repository-distributed community skills: clone a repo that actually ships the skill and copy it into the skills directory (awesome-claude-skills is a link list, so pick an individual skill repo from it).

git clone https://github.com/JuliusBrussee/caveman
cp -r caveman/skills/* ~/.claude/skills/

Codex catalog skills are installed from the TUI by passing a name or a GitHub URL to $skill-installer. Loading them requires a Codex restart.

$skill-installer gh-address-comments

Claude Code's official skills need no install. Call them with a slash command like /code-review, or state the intent (such as creating a document) to trigger them automatically.

How to write a skill (minimal template)

---
name: my-skill
description: One or two sentences on what this skill is for. Lead with trigger words.
---

# My Skill

## Steps

1. The first thing to do
2. The next thing to do
3. Confirm the completion condition

description is the key the agent uses to auto-select the skill. The clearer the purpose and trigger words, the more accurately it gets chosen.


Check GitHub for the latest details on each repository.