How to Use Claude Code

Introduction

This guide explains the core concepts, agents, and tools available in Claude Code, and shows a practical case of running multiple agents in parallel.

Key Concepts

Memory

Markdown files with project context. Location: ~/.claude/projects/<project>/memory/. Read-only for agents.

Skills

Packaged prompts for common tasks. Location: ~/.claude/skills/. Invoked like /skill-name.

Agents / Sub-agents

Separate Claude instances working in parallel (spawned via Task tool).

MCP Servers

Services that connect Claude to external tools via Model Context Protocol (configured in ~/.claude/mcp.json).

Sub-Agent Types

Type	Use Case	Speed
General-purpose	Complex multi-step tasks, research, analysis	Medium
Explore	Quick codebase/file discovery (Glob, Grep, Read)	Fast
Plan	Architecture and implementation design before coding	Medium
Bash	Run terminal commands when needed	Fast

Tools Overview

Tool	Description
Read(file_path)	Read a file
Write(file_path, content)	Create a new file
Edit(old, new)	Modify existing files
Glob(pattern)	Find files by pattern
Grep(pattern)	Search file contents
Bash(command)	Run shell commands
Task(subagent_type, prompt)	Launch sub-agents
WebFetch(url, prompt) / WebSearch(query)	Web interactions

Anatomy of an Effective Agent Prompt

Structure your prompt like this:

1. Agent identification: who you are and the project
2. Objective: a clear, specific goal
3. Context: relevant project info and files
4. Deliverables: detailed list of outputs
5. Output format: exact format required for copy-paste or PR-ready content

Example (Content Agent)

Eres el CONTENT & CONTEXT AGENT para <Project>.

## Objetivo
Reescribir secciones clave para posicionar el sitio como AI-first.

## Contexto
- Archivo: src/routes/(marketing)/+page.svelte
- Servicios actuales: 3 (expandir a 5)

## Tarea
1. Reescribir hero section
2. Actualizar stats

## Formato
Proveer textos exactos listos para pegar.

Case Study: 3 Agents in Parallel

Task Distribution

• Task 1 — Content Agent: Update copy to position as AI experts
• Task 2 — Brand Protection Agent: Audit brand, legal, credibility
• Task 3 — Programmer Agent: Review code structure and bugs

Benefits

• Speed: parallel execution saved time
• Specialization: each agent focuses on distinct concerns
• Full context: agents receive project history and files

Effective Patterns

Choose the pattern that fits project size and modularity:

• By specialization: content / design / legal
• By file/module: frontend / backend / DB
• By phase: research / design / implementation

Tips & Tricks

Use agents for broad or multi-file work; do small edits yourself.

Parallelize truly independent tasks — avoid parallelizing dependent steps.

Note: Agents can access prior conversation history and project files, but not changes made after they start or other agents’ live results.

Full Workflow Example

1. User requests an AI-first homepage
2. Launch 3 specialized agents in parallel
3. Agents report results and Claude synthesizes changes
4. Save reports to memory/, commit and push to Git

Essential Commands

Command	Description
/tasks	List active tasks
/install <skill>	Install a skill
/skills	List available skills
/commit	Commit changes
/review-pr <id>	Review pull request
EnterPlanMode() / ExitPlanMode()	Planning sessions

Next Steps

Level 1: Use Task for small parallel searches and save findings to memory.

Level 2: Launch 3 agents for a complex task and create a custom skill.

Level 3: Design a 5+ agent workflow and integrate a custom MCP server.

Resources

• Skills docs — Skills documentation
• MCP Protocol — Model Context Protocol specification
• Claude Code GitHub — Official repository

Conclusion

Claude Code represents a significant shift in how developers interact with AI tools. By leveraging memory, skills, and parallel agents, you can dramatically improve your development workflow.

Start with simple tasks and gradually build up to complex multi-agent workflows as you become more comfortable with the system.