AI Glossary for Claude Code

AI Glossary for Claude Code

Understanding the terminology around AI agents and coding assistants can be challenging. This glossary covers essential terms you’ll encounter when working with Claude Code and similar agentic AI tooling.

Skill

A “skill” in the context of claude code and similar tooling represents a special type of a file (SKILL.md) which describes an action claude code can take.

The files live in ./claude/skills/[your-skill-name]/ with SKILL.md being the entrypoint. There may be other other files there such as template.md for claude to use, scripts that claude can invoke, and even example output.

Think of a “skill” as an auto-invoking function that you register with your tool.

Key points:

  • the full skill isn’t automatically loaded. It is loaded based on its name and description.
  • a skill may just be a prompt. It may also be a folder full of scripts to run.
  • SKILL.md is the entry point (always required)
  • you can run a skill like a slash command to test it (eg. /my-skill-name)
  • skills auto-run – you don’t need to explicitly invoke them.

Tip: Skills supercede and replace commands because Claude code (and other tooling) is able to automatically run these instead of having to be invoked.

Example

File location: ./claude/skills/build-a-component/SKILLS.md

---
name: build-a-component
description: Build a React component using TypeScript within this application with stories, tests, and types.
---

Build a React component using the [component-template.md](component-template.md). Place this component into its own folder in `src/ui/components`. For example, a Button component should be in `src/ui/components/Button/`.

Utilize the [stories-template.md](stories-template.md) to create a storybook instance for this component. And use the [tests-template](tests-template.md) for test generation.

Verify the base component works by running storybook and running tests.

Then, you can create straightforward templates for Claude code to fill out.

Usecases

Skills are tremendously useful for any repeated task and any task you want repeatedly support for.

Some examples:

  • documentation generation
  • building a component
  • running custom Python scripts to help claude code

Resources

Subagent

To simplify it, agent is just LLM + Tools. Agents are orchestrations of LLM output that can drive tools to take action and accomplish tasks. A subagent is just another instance of an agent that’s being controlled by a higher-up agent.

In terms of Claude code or similar tooling, an agent is the process that can autonomously handle complex tasks.

To illustrate this:

  • Claude Code instance is your agent accomplishing its task (like building a new React component)
  • CC may employ separate subagents to accomplish this task (a documentation subagent to generate docs, coding subagent to build code, etc.)
  • subagents let your agent know of its progress/task (a doc subagent letting your agent know it’s done)

A subagent can be defined in ./claude/agents/. But should be defined through claude code itself via /agents command and creating a new agent.

These subagents are picked to do their job by Claude Code.

Key points:

  • an agent can have multiple subagents working for it
  • built-in subagents for Claude include specialized agents for exploration, planning, and complex multi-step operations
  • subagents are autonomous and repond to the agent controlling it (and you should be controlling the agent)
  • subagents have their own context/prompt/etc. it’s a separate instance of an agent

Tip: Think of agents as autonomous workers that can break down complex tasks, make decisions, and execute actions without constant human supervision. Think of subagents as a directed specific agent tasked by your agent

Example

Use claude code to generate the agent definition and then add your own agent prompt file:

---
name: documentation-creator
description: Creates easy-to-read documentation for any code files
---

You are a documentation writer that creates docs for any code file. The docs should be succinct, document all APIs, usecases, and have a guide on how to extend the code being documented.

Usecases

Agents are tremendously useful for:

  • Code exploration: Understanding large codebases without manually reading every file
  • Planning implementations: Breaking down complex features into step-by-step plans
  • Multi-step workflows: Executing sequences of related tasks autonomously
  • Custom automation: Building specialized agents for domain-specific tasks

Resources

Command

Tip: Commands are now deprecated, use skills instead

A “command” (or “slash command”) in Claude Code provides a way to control sessions and execute predefined prompts using special commands that start with /. Commands can be built-in system commands or custom user-defined prompts.

Commands are essentially shortcuts that trigger specific behaviors or inject predefined instructions into the conversation.

Key points:

  • Commands always start with / (e.g., /help, /commit, /clear)
  • Built-in commands handle system functions like clearing history or getting help
  • Custom commands are Markdown files containing natural language prompts
  • MCP servers can expose prompts as slash commands

Example

Custom command: .claude/commands/review.md

Review the current pull request for:
- Code quality and best practices
- Potential bugs or security issues
- Test coverage
- Documentation completeness

Provide a summary with specific line references.

Usage: /review - Claude executes the prompt defined in review.md

Usecases

Commands are useful for:

  • Repeated prompts: Codifying frequently-used instructions
  • Project conventions: Enforcing team standards and workflows
  • Quick actions: Fast access to common operations
  • Consistency: Ensuring the same prompt is used across sessions
  • Onboarding: Sharing standardized workflows with team members

Resources

MCP

MCP (Model Context Protocol) is an open standard and open-source framework that standardizes the way AI systems like large language models integrate and share data with external tools, systems, and data sources. It enables developers to build secure, two-way connections between their data sources and AI-powered tools. Think of MCP as HTTP and the MCP server as your API server for your service.

Key points:

  • Open standard: Not proprietary to any single vendor
  • Two-way communication: AI can both read from and write to external systems
  • Server architecture: Pre-built MCP servers exist for popular tools (GitHub, Slack, Postgres, etc.)
  • Cross-platform SDKs: Available for Python and TypeScript
  • Industry adoption: Adopted by OpenAI, Google DeepMind, and other major AI providers
  • Foundation governance: Donated to the Agentic AI Foundation under the Linux Foundation in December 2025

Example

Using MCP with Claude Code:

  1. Install an MCP server (e.g., GitHub server):
npm install -g @modelcontextprotocol/server-github

  1. Configure in Claude Code settings to connect to GitHub

  2. Claude can now:

    • Read repository contents
    • Create and update issues
    • Review pull requests
    • Access GitHub data without custom API integration

Usecases

MCP enables powerful integrations:

  • Data access: Connect AI to databases, file systems, APIs without custom code
  • Tool integration: Let AI interact with external tools (Slack, Jira, CRMs)
  • Enterprise systems: Secure connections to internal business systems
  • Cross-tool workflows: Chain actions across multiple systems
  • Standardization: One protocol for all AI-to-tool connections

Resources