Comparisons
The AI agent primitives teams keep confusing, compared.
Coding agents keep shipping new units of configuration, and several of them look interchangeable from the outside. These pages set two primitives next to each other on the dimensions that decide the choice, say plainly when each one is the wrong pick, and link every claim to the first-party documentation it came from.
Comparisons
Skills vs subagents
Claude skills vs subagents
Two Claude Code primitives that look interchangeable and are not. One loads instructions into your conversation, the other starts a separate run with its own context window.
Skills vs MCP
Claude skills vs MCP (Model Context Protocol)
The most common false choice in agent tooling. One packages what the agent should do, the other is the wire it uses to reach a system it does not own.
Skills vs plugins
Claude skills vs plugins
Not two formats competing for the same job. One is what you write, the other is how you ship it, and a plugin can carry skills inside it.
Skills vs slash commands
Claude skills vs slash commands
Not two competing formats. The slash is the menu, the skill is the unit, and the documentation now treats a command file as a skill with fewer options.
How these pages work
Sourced, dated, and willing to say it depends
Each comparison starts with a short answer you can quote, then a table of the dimensions that differ, then the cases where each side wins and the cases where it loses. Every section lists the first-party pages behind it, and every page carries the date the claims were last checked. Where the documentation is silent, the page says so rather than guessing.
Picking the primitive is the easy half.
The harder half is agreeing on which files your team actually recommends. Skills Board keeps that list in one searchable place: free forever, MIT licensed, and open source.