DEV Community

Cover image for Agentic AI Glossary Cheatsheet
Ant The Developer
Ant The Developer

Posted on

Agentic AI Glossary Cheatsheet

#ai #agentic #webdev

Quick reference guide for Claude Code and agentic AI terminology. I have a longer ai glossary for claude code on my personal blog.

📃 Quick Reference

Skill

What: Auto-invoking function registered with your tool
Location: ./claude/skills/[skill-name]/SKILL.md
Run: /my-skill-name (also, runs automatically)

Example:

---
name: build-component
description: Build a React component with tests and stories
---

Instructions for building the component...
Enter fullscreen mode Exit fullscreen mode

Subagent

What: Specialized agent controlled by your main agent
Location: ./claude/agents/[agent-name]/
Create: /agents command in Claude Code
Run: runs automatically

---
name: documentation-creator
description: Creates easy-to-read documentation
---

You are a documentation writer...
Enter fullscreen mode Exit fullscreen mode

Command (Deprecated)

What: Slash command shortcuts (use Skills instead)
Location: .claude/commands/review.md
Run: /review

🔑 Term Definitions

Skill

A special file (SKILL.md) that describes an action Claude Code can take. Think of it as an auto-invoking function.

Key points:

  • Auto-loads based on name and description
  • Entry point is always SKILL.md
  • Can include templates, scripts, and examples
  • Auto-runs without explicit invocation

Use for: Repeated tasks, component generation, documentation, custom scripts

Subagent

Another instance of an agent controlled by a higher-up agent. Agent = LLM + Tools.

Key points:

  • Agents can have multiple subagents
  • Built-in subagents for exploration, planning, multi-step operations
  • Autonomous workers with their own context/prompt
  • Respond to the controlling agent

Use for: Code exploration, planning implementations, multi-step workflows, custom automation

Command (Deprecated - Use Skills)

Slash commands that trigger predefined prompts or system functions.

Key points:

  • Always start with /
  • Built-in commands handle system functions
  • Custom commands are Markdown files with prompts
  • MCP servers can expose prompts as slash commands

Use for: Repeated prompts, project conventions, quick actions, consistency

MCP (Model Context Protocol)

Open standard for standardizing how AI systems integrate with external tools and data sources. MCP servers enable integrations between tooling like claude code and external services like Jira or GitHub.

Key points:

  • Open standard (not proprietary)
  • Two-way communication (read and write)
  • Pre-built servers for popular tools (GitHub, Slack, Postgres)
  • Cross-platform SDKs (Python, TypeScript)
  • Adopted by OpenAI, Google DeepMind, and others

Use for: Data access, tool integration, enterprise systems, cross-tool workflows

🎯 Comparison Table

Feature Skill Subagent Command MCP
Purpose Reusable actions Autonomous task execution Shortcuts External integrations
Auto-runs ✅ Yes ✅ Yes ❌ Manual N/A
Location ./claude/skills/ ./claude/agents/ .claude/commands/ External servers
File SKILL.md Agent definition .md file Server config
Status ✅ Current ✅ Current ⚠️ Deprecated ✅ Current

📚 Resources

Skills:

Subagents:

Commands:

MCP:

For detailed explanations, see the full AI Glossary

Top comments (0)