Anthropic Skills Explained

Anthropic Skills launched late 2025 as the answer to ChatGPT’s Custom GPTs — reusable, packaged behaviors for Claude. By 2026 they ship with Claude.ai, Claude Desktop, and Claude Code. Here’s the model.

What a Skill is

A directory with a SKILL.md file that defines a specific behavior + trigger keywords + optional tool access. When Claude detects the trigger condition, it loads the skill instructions and follows them.

Where Skills run

• Claude.ai web — user-loadable Skills.
• Claude Desktop — local Skills with filesystem access.
• Claude Code — project Skills in .claude/skills/.
• Anthropic API (Managed Agents) — embedded in agent flows.

Anatomy of a Skill

---
name: review-pr
description: Review a pull request for safety, performance, style. Use when asked to "review PR" or "code review".
---
# Instructions
1. Read the PR diff and surrounding context.
2. Flag: bugs, security issues, perf regressions, style violations.
3. Output a structured review with severity ratings.

# Examples
[1-2 ideal example reviews]

Skills vs Custom GPTs

• Skills: markdown-defined, version-controlled in git, work in terminal.
• Custom GPTs: public marketplace, custom Actions for third-party APIs, no terminal version.
• Pick Skills for code-heavy or terminal-driven workflows; Custom GPTs for public-facing AI tools.

The 2026 Skill ecosystem

Anthropic ships ~30 official Skills (slack-gif-creator, theme-factory, doc-coauthoring, mcp-builder, etc.). Hundreds more in plugin marketplaces. The Skill+plugin combination is rapidly becoming the default Claude Desktop / Claude Code workflow.