What Are Claude Skills? The Complete Guide (2026)

A Claude skill is a folder of instructions that changes how Claude works — permanently, across every conversation, without you re-explaining anything. At its core sits a SKILL.md file: a markdown document with frontmatter that tells Claude when to activate and what to do once it does. Some skills add reference files, scripts, or templates alongside it.

That’s the mechanical definition. The practical one: a skill is expertise you install once instead of prompting for every time.

Skills vs. prompts vs. MCP servers

People conflate three different things, so let’s separate them:

• A prompt is a one-off instruction. It lives in a single conversation and dies with it. Great for one-time tasks, useless for repeatable workflows.
• A skill is a persistent instruction set. Claude reads it automatically whenever the task matches the skill’s trigger description. It changes how Claude behaves — how it writes code, structures documents, or approaches debugging.
• An MCP server is a connection to external tools and data — your database, Slack, a browser. It changes what Claude can touch, not how it thinks.

The three compose: a sales skill might teach Claude how to write cold email, while an MCP server gives it access to your CRM to pull prospect data. If you want the deeper comparison, we wrote a full breakdown in Claude Skills vs MCP Servers.

Where skills live

On disk, skills sit in one of two places:

• ~/.claude/skills/ — global, available in every project
• .claude/skills/ inside a project — scoped to that repository

Each skill is a folder named after the skill, containing at minimum a SKILL.md. Claude Code scans these directories on startup, reads each skill’s description, and loads the full instructions only when a conversation matches. That lazy-loading design matters: fifty installed skills don’t bloat your context — only the relevant one activates.

Why the ecosystem has a quality problem

Anyone can publish a skill: it’s just markdown in a GitHub repo. That’s the ecosystem’s superpower and its curse. Our crawler finds hundreds of new “skills” every week, and when we actually test them, the failure pattern is consistent: roughly half don’t install cleanly from their own README, many never trigger on the prompts they claim to handle, and a good share produce output no better than Claude without the skill.

That’s the reason this site exists. Every skill in our directory gets installed on a clean setup, run against real work, and scored on four criteria — install, triggering, output quality versus baseline, and documentation honesty. The methodology is public. Verdicts are earned, not assumed.

How to start

1. Pick a proven first skill. The Anthropic document skills are the reference standard — start with DOCX if you produce documents, or Frontend Design if you ship web UI. Both scored above 9/10 in our tests.
2. Install it. Copy the skill folder into ~/.claude/skills/ and restart Claude Code. Every listing on this site includes the exact command.
3. Test the trigger. Ask Claude to do the thing the skill covers and confirm behavior changed. If you’re curious what “changed” looks like, our before/after examples show identical tasks with and without each skill.
4. Build your stack by role. Our collections bundle the working setup for developers, marketers, sales teams, writers and analysts — in install order.

Skills are the closest thing Claude has to an app store. The catch is that nobody checks the apps — except, well, us.