Claude Code Plugins

Community-maintained marketplace

Feedback
0
0

Designing Slash Commands for Claude Code. Create and optimize slash commands with proper argument parsing, tool permissions, and agent orchestration. Use when building workflow entry points, automation commands, or user-facing shortcuts.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name moai-cc-commands
description Designing Slash Commands for Claude Code. Create and optimize slash commands with proper argument parsing, tool permissions, and agent orchestration. Use when building workflow entry points, automation commands, or user-facing shortcuts.
allowed-tools Read, Write, Edit, Glob, Bash

Skill Metadata

Field Value
Version 1.0.0
Tier Ops
Auto-load When designing slash commands

What It Does

Slash command 설계 및 최적화 방법을 제공합니다. Argument parsing, tool permissions, agent orchestration을 포함한 command 구조를 다룹니다.

When to Use

  • 새로운 slash command를 생성할 때
  • 기존 command를 최적화하거나 리팩토링할 때
  • Workflow entry point를 설계할 때
  • User-facing automation을 구축할 때

Designing Slash Commands

Slash commands are user-facing entry points that orchestrate sub-agents, manage approvals, and coordinate multi-step workflows. They follow the Plan → Execute → Sync cadence.

Command File Structure

Location: .claude/commands/

---
name: command-name
description: Brief description of what the command does
argument-hint: "[param1] [param2] [optional-param]"
tools: Read, Write, Task, Bash(git:*)
model: sonnet
---

# Command Title

Brief description of functionality.

## Usage

- `/command-name param1 param2` — Basic usage
- Parameter descriptions
- Expected behavior

## Agent Orchestration

1. Call specific agent for task
2. Handle results
3. Provide user feedback

Command Design Patterns

Pattern 1: Planning Command

---
name: /alfred:1-plan
description: Write SPEC requirements in EARS syntax
argument-hint: "[title]"
tools: Read, Write, Task
model: sonnet
---

# SPEC Planning Command

Initiates SPEC authoring via spec-builder sub-agent.

## Usage

`/alfred:1-plan "User authentication system"`

## Agent Orchestration

1. Invoke spec-builder agent
2. Gather requirements via EARS patterns
3. Create SPEC file in `.moai/specs/`
4. Suggest next step: `/alfred:2-run`

Pattern 2: Code Review Command

---
name: /review-code
description: Trigger automated code review with quality analysis
argument-hint: "[file-pattern] [--strict]"
tools: Read, Glob, Grep, Task
model: sonnet
---

# Code Review Command

Analyzes code quality against TRUST 5 principles.

## Usage

- `/review-code src/**/*.ts` — Review TypeScript files
- `/review-code . --strict` — Strict mode (fail on warnings)

## Agent Orchestration

1. Scan files matching pattern
2. Invoke code-reviewer agent
3. Generate report with findings
4. Suggest fixes with severity levels

Pattern 3: Deployment Command

---
name: /deploy
description: Deploy application with safety gates
argument-hint: "[env] [--force]"
tools: Read, Write, Task, Bash(git:*)
model: haiku
---

# Deployment Command

Orchestrates multi-step deployment with approval gates.

## Usage

- `/deploy staging` — Deploy to staging
- `/deploy production --force` — Force production deploy

## Agent Orchestration

1. Validate deployment readiness
2. Run pre-deployment checks
3. Ask for user approval
4. Execute deployment
5. Monitor post-deployment

Argument Parsing Pattern

# Inside command execution
$ARGUMENTS              # Entire argument string
$1, $2, $3            # Individual arguments
$@                    # All arguments as array

# Example: /my-command arg1 arg2 --flag
# $ARGUMENTS = "arg1 arg2 --flag"
# $1 = "arg1"
# $2 = "arg2"
# $3 = "--flag"

High-Freedom: Orchestration Strategies

Sequential Execution

Command: /plan-and-implement
├─ Phase 1: spec-builder (SPEC creation)
└─ Phase 2: code-builder (TDD implementation)

Parallel Execution

Command: /analyze-project
├─ Agent 1: security-auditor (vulnerability scan)
├─ Agent 2: performance-analyzer (bottleneck detection)
└─ Agent 3: architecture-reviewer (design review)

Conditional Branching

Command: /fix-errors
├─ Check: Are tests failing?
│  ├─ YES → debug-helper agent
│  └─ NO → suggest next steps

Medium-Freedom: Command Templates

Status Check Command

name: /status
description: Show project status summary
argument-hint: "[--verbose]"
tools: Read, Bash(git:*)
model: haiku
---

# Status Check

Displays project health: SPEC status, test results, Git state.

## Checks

1. SPEC completeness
2. Test coverage
3. Git branch status
4. Recent commits
5. TODO items

Bulk Operation Command

name: /migrate
description: Run migration across multiple files
argument-hint: "[from] [to] [--preview]"
tools: Read, Glob, Bash
model: sonnet
---

# Migration Command

Migrates code patterns across project.

## Usage

- `/migrate v1-api v2-api --preview` — Show changes without applying
- `/migrate v1-api v2-api` — Apply migration

Low-Freedom: Safety & Approval Patterns

Approval Gate Pattern

# Inside command execution

echo "🔴 This action is destructive. Review carefully:"
echo "  • Will delete 10 files"
echo "  • Cannot be undone"
echo ""
read -p "Type 'yes' to confirm: " confirm

if [[ "$confirm" != "yes" ]]; then
  echo "❌ Cancelled"
  exit 1
fi

# Execute dangerous operation

Dry-Run Pattern

if [[ "${3:-}" == "--preview" ]]; then
  echo "🔍 Preview mode (no changes)"
  # Show what would happen
  exit 0
fi

# Execute actual changes

Command Registry

# List all commands
/commands

# View command details
/commands view /deploy

# Create new command
/commands create

# Edit command
/commands edit /deploy

# Delete command
/commands delete /deploy

Command Validation Checklist

  • name is kebab-case (e.g., /review-code)
  • description clearly explains purpose
  • argument-hint shows expected parameters
  • tools list is minimal and justified
  • model is haiku or sonnet
  • Agent orchestration is clearly defined
  • Arguments are properly parsed
  • Safety gates are in place for dangerous operations
  • Feedback to user is clear and actionable

Best Practices

DO:

  • Design commands around workflows, not tools
  • Use agents for complex logic
  • Include preview/dry-run modes for risky operations
  • Provide clear feedback at each step
  • Link to next command in suggestions

DON'T:

  • Make commands do too much (limit to 1 coherent workflow)
  • Require multiple parameters without defaults
  • Skip approval gates for destructive operations
  • Leave users guessing what happened

Reference: Claude Code Slash Commands documentation Version: 1.0.0