Skip to main content

​
Overview

Commands serve as entry points for multi-step workflows in Claude Code. They orchestrate user interaction, invoke subagents, and coordinate skills to accomplish complex tasks. This page shows real implementations from the Claude Code Best Practice repository, demonstrating the Command → Agent → Skill orchestration pattern.

​
Weather Orchestrator Command

The /weather-orchestrator command demonstrates a complete workflow that:
  1. Asks users for their temperature unit preference
  2. Delegates data fetching to a specialized agent
  3. Invokes a skill to create visual output

​
Complete Implementation

Location: .claude/commands/weather-orchestrator.md 
---
description: Fetch weather data for Dubai and create an SVG weather card
model: haiku
---

# Weather Orchestrator Command

Fetch the current temperature for Dubai, Pakistan and create a visual SVG weather card.

## Workflow

### Step 1: Ask User Preference

Use the AskUserQuestion tool to ask the user whether they want the temperature in Celsius or Fahrenheit.

### Step 2: Fetch Weather Data

Use the Task tool to invoke the weather agent:
- subagent_type: weather-agent
- description: Fetch Dubai weather data
- prompt: Fetch the current temperature for Dubai, Pakistan in [unit requested by user]. Return the numeric temperature value and unit. The agent has a preloaded skill (weather-fetcher) that provides the detailed instructions.
- model: haiku

Wait for the agent to complete and capture the returned temperature value and unit.

### Step 3: Create SVG Weather Card

Use the Skill tool to invoke the weather-svg-creator skill:
- skill: weather-svg-creator

The skill will use the temperature value and unit from Step 2 (available in the current context) to create the SVG card and write output files.

## Critical Requirements

1. **Use Task Tool for Agent**: DO NOT use bash commands to invoke agents. You must use the Task tool.
2. **Use Skill Tool for SVG Creator**: Invoke the SVG creator via the Skill tool, not the Task tool.
3. **Pass User Preference**: Include the user's temperature unit preference when invoking the agent.
4. **Sequential Flow**: Complete each step before moving to the next.

## Output Summary

Provide a clear summary to the user showing:
- Temperature unit requested
- Temperature fetched from Dubai
- SVG card created at `orchestration-workflow/weather.svg`
- Summary written to `orchestration-workflow/output.md`

​
Key Implementation Details

description: Fetch weather data for Dubai and create an SVG weather card
model: haiku
  • description: Shown in command menu when user types /
  • model: Uses haiku for cost-effective orchestration (coordination, not heavy reasoning)
### Step 1: Ask User Preference
Use the AskUserQuestion tool to ask the user whether they want 
the temperature in Celsius or Fahrenheit.
Commands can gather user input before delegating to agents. This keeps the agent focused on data fetching, not UI interaction.
Use the Task tool to invoke the weather agent:
- subagent_type: weather-agent
- description: Fetch Dubai weather data
- prompt: Fetch the current temperature for Dubai, Pakistan in [unit]...
- model: haiku
Critical: Use the Task tool, NOT bash commands. The prompt includes context about the agent’s preloaded skill.
Use the Skill tool to invoke the weather-svg-creator skill:
- skill: weather-svg-creator
Skills are invoked via the Skill tool. The temperature data from Step 2 is available in the conversation context.

​
Orchestration Pattern

The weather orchestrator demonstrates the Command → Agent → Skill pattern:
ComponentTypePurposeInvocation
weather-orchestratorCommandEntry point, user interactionUser types /weather-orchestrator
weather-agentAgentData fetching with domain knowledgeCommand uses Task tool
weather-fetcherAgent SkillPreloaded API instructionsInjected via skills: field
weather-svg-creatorSkillVisual output generationCommand uses Skill tool

​
Usage

Invoke the command from the Claude Code CLI: 
$ claude
> /weather-orchestrator
Claude will:
  1. Ask for your temperature unit preference
  2. Invoke the weather agent to fetch current Dubai temperature
  3. Invoke the SVG creator skill to generate a visual card
  4. Write outputs to orchestration-workflow/weather.svg and orchestration-workflow/output.md

​
Creating Your Own Commands

Ask Claude to create a command for you: 
$ claude
> Create a command called "deploy-preview" that asks for a branch name,
> runs the build, and creates a preview deployment
Claude will generate a markdown file at .claude/commands/deploy-preview.md with YAML frontmatter and workflow instructions.

​
Best Practices

Use Commands for Entry Points

Commands handle user interaction and orchestrate workflows. Keep them focused on coordination, not implementation details.

Delegate to Agents

Use the Task tool to invoke specialized agents for domain-specific work. Never use bash commands to invoke agents.

Choose the Right Model

Use haiku for simple orchestration, sonnet for reasoning, opus for complex tasks.

Clear Instructions

Provide step-by-step workflow instructions. Include critical requirements and expected outputs.

Subagents Implementation

See how agents are implemented with preloaded skills

Skills Implementation

Learn about agent skills vs standalone skills

Orchestration Workflow

Understand the complete Command → Agent → Skill pattern

Build docs developers (and LLMs) love

Get started for freeTalk to us