Level up your AI with Claude Skills, Agents and Agent Teams

So you've been leverage Claude to accelerate your software development, but you haven't quite grasped the difference between Skills, Agents and Agent Teams yet. Lets get right into it.

Skills

Skills allow you to save common conversation starters for common tasks that you find yourself repeating. Examples could be performing code reviews, checking your task list, etc.

A note on token usage: Skills have a combination of frontmatter and content. The frontmatter is pulled into every new conversation. If your skill's frontmatter uses 100 tokens, and you have 10 skills, then each new conversation with claude will use 1,000 tokens.

Claude Skills Documentation

Agents

Agents run as a sub process triggered from your existing conversation. When you trigger an agent, everything in the context of that agent is kept in a separate context. When the agent has completed its work, it returns output to the parent conversation. That output is the only part that stays in the context of the parent conversation.

An agent can reference multiple skills. When an agent references a skill, the full content of the skill is pulled into the agent's context. (As opposed to only pulling in the skill's frontmatter.)

Claude Agents Documentation

Agent Teams

Agent Teams allows you to spawn multiple agents from the same conversation. These sub-agents can all communicate with each other as they work against a shared task list.

As of the time of this writing, Agent Teams is experimental and off by default.

Claude Agent Teams Documentation