Claude Code Command Patterns

Guide for creating and modifying Claude Code slash commands with proper structure, parallel execution patterns, and agent coordination.

When to Use This Skill

Activate when:

• Creating new Claude Code slash commands
• Modifying existing command structure
• Implementing parallel Task invocations
• Adding TodoWrite progress tracking
• Coordinating multiple agents from commands
• Debugging command-to-agent integration

Command Structure

Required Frontmatter

---
description: "Command purpose in clear, concise terms"
argument-hint: "[input-type] [optional-params]"
---

Command Body Template

I am a [command purpose] that [brief description]. My task is to:

1. Setup todo tracking for the workflow
2. Invoke specialized agents using parallel Task calls
3. Read and validate outputs from expected files
4. Create comprehensive report with findings

## Implementation

I will use TodoWrite to track progress, then make parallel Task calls:
- Task with subagent_type: "agent-name-1" - pass the [argument] for [purpose]
- Task with subagent_type: "agent-name-2" - pass the [argument] for [purpose]

Then read outputs, validate them, and create complete report.

Required Elements

1. TodoWrite First

Always setup todo tracking before Task calls:

I will use TodoWrite to track progress...

2. Parallel Task Calls

Explicitly state parallel execution:

...then make parallel Task calls:
- Task with subagent_type: "agent-a" - ...
- Task with subagent_type: "agent-b" - ...

3. Argument Passing

Pass meaningful context to each agent:

- Task with subagent_type: "analyzer" - pass the project_path argument [PATH] for analysis

4. Output Validation

Read and validate actual outputs:

Then read outputs from [files], validate them, and create complete report.

Execution Patterns

Parallel Execution

When tasks are independent:

make parallel Task calls:
- Task with subagent_type: "agent-1" - ...
- Task with subagent_type: "agent-2" - ...
- Task with subagent_type: "agent-3" - ...

Sequential Execution

When tasks have dependencies:

First, use Task with subagent_type: "discovery-agent" to analyze project.
Then, use Task with subagent_type: "plan-agent" with discovery results.
Finally, use Task with subagent_type: "task-agent" with plan output.

Conditional Execution

When execution depends on results:

First check if [condition] exists using Read tool.
If present, skip to validation phase.
If not, execute discovery workflow.

Argument Extraction Pattern

Commands should document argument extraction for agents:

## Argument Extraction Instructions

When invoking agents, use explicit argument passing:
- "pass the project_path argument [PATH] for [purpose]"
- "pass the changes-description argument '[DESC]' for [purpose]"

Agents extract these values from their task prompts.

Anti-Patterns to Avoid

Never Use

• Coordinator agents that call other agents (commands coordinate directly)
• Sequential execution when parallel is possible
• Missing TodoWrite progress tracking
• Generic arguments without context
• Assuming outputs without Read/validation
• Incorrect subagent_type references

Incorrect

# WRONG: Missing TodoWrite
I will invoke agents directly...

# WRONG: No argument passing
- Task with subagent_type: "analyzer"

# WRONG: Sequential when parallel works
First run agent-1, then agent-2, then agent-3...

Correct

# RIGHT: TodoWrite first
I will use TodoWrite to track progress, then...

# RIGHT: Explicit argument passing
- Task with subagent_type: "analyzer" - pass the project_path argument ./src for code analysis

# RIGHT: Parallel execution
...make parallel Task calls:
- Task with subagent_type: "agent-1" - ...
- Task with subagent_type: "agent-2" - ...

Complete Example

---
description: "Analyze project security posture and generate comprehensive report"
argument-hint: "[project-path]"
---

I am a security analysis coordinator that evaluates project security across multiple dimensions. My task is to:

1. Setup todo tracking for the security workflow
2. Invoke security agents using parallel Task calls
3. Read and validate outputs from security reports
4. Create comprehensive security assessment

## Implementation

I will use TodoWrite to track progress, then make parallel Task calls:
- Task with subagent_type: "vulnerability-assessment-specialist" - pass the project_path argument for vulnerability scanning
- Task with subagent_type: "security-architecture-analyst" - pass the project_path argument for architecture review
- Task with subagent_type: "security-threat-modeler" - pass the project_path argument for STRIDE analysis

Then read outputs from security reports, validate findings, and create consolidated assessment.

Naming Conventions

• Commands: lowercase-hyphen (e.g., security-review.md)
• Agents: lowercase-hyphen (e.g., security-threat-modeler.md)
• subagent_type: Must match agent filename without .md

Key Principles

• Commands coordinate agents directly (no intermediate coordinators)
• Use TodoWrite for progress tracking
• Prefer parallel execution when tasks are independent
• Pass explicit arguments with clear purposes
• Validate outputs before creating final reports
• Use consistent lowercase-hyphen naming