Claude Code Cheat Sheet

Photo by vackground.com on Unsplash

Skills

You can think of skills as modular pieces of knowledge or functionality that Claude can leverage to perform specific tasks more effectively. Skills are more than just a SKILL.md file - they can also include scripts, assets, and other resources that help Claude execute the skill’s purpose. As you’ll see later, I utilize a simple skill to help Claude with something it’ll need to repeat over and over.

Claude will not load skills into context until they’re invoked (can be explicitly by the user, or automatically by Claude), BUT they are added to the same context window as your main Claude instance. They will inherit the model that your main Claude instance uses.

Sub-Agents

Sub-Agents are autonomous entities within Claude Code that can operate independently to perform specific tasks. Sub-agents can be thought of as mini-versions of Claude that can handle specific aspects of a project. They can have their own skills, context, and objectives, allowing them to work on tasks concurrently or in parallel with the main Claude instance. Each sub-agent has its own context window separate from the main Claude instance. Also, you can specify which model each sub-agent uses, so you can use low-token models for simple tasks and switch to more powerful models for other tasks.

These are different than skills in that sub-agents can manage their own workflows and make decisions independently, whereas skills are more about providing specific knowledge or capabilities to Claude or its sub-agents. They also have their own, separate context window, whereas skills share the same context window as the main Claude instance. In practice, sub-agents are useful for taking on a specific part of a larger project, and often work in parallel and alongside other sub-agents or the main Claude instance.

MCP Servers

MCP (Model Context Protocol) servers are services that can help Claude “integrate and share data with external tools, systems, and data sources.” Initially, I thought that MCP Servers required me to run an actual server locally or connect to one in the cloud (via HTTP or similar) and let Claude connect to it. While MCP servers do spawn a sub-process, they feel more like installing an NPM package to me, but maybe that’s just because I haven’t used many of them yet.

Plugins

Plugins are packages that can include skills, sub-agents, hooks, and/or MCP servers. Plugins allow you to bundle related functionalities together and easily add them to your Claude Code environment. They can be installed from the Claude Code Plugin Registry or created custom for your specific needs.