Command Creator

Command Creator

This skill guides the creation of Claude Code slash commands for agents in the Traycer enforcement framework. Claude Code commands are custom slash commands defined as Markdown files containing prompts.

When to Use

Use command-creator when:

1. Creating custom slash commands for an agent
2. Defining reusable prompts that should be invoked explicitly
3. Building commands that need dynamic arguments
4. Organizing frequently-used agent prompts

Understanding Claude Code Commands

What Are Commands?

Claude Code commands are custom slash commands that:

• Are Markdown (.md) files containing prompts
• Live in .claude/commands/ (project) or ~/.claude/commands/ (personal)
• Are invoked with /command-name syntax
• Support dynamic arguments via placeholders
• Can execute bash commands before running
• Can reference files using @ prefix

Commands are NOT:

• Bash/CLI scripts
• Python/JavaScript code
• Complex workflows (those are Skills)
• Automatically triggered (they require explicit invocation)

Commands vs Skills vs System Prompt

Use commands when:

• You have a frequently-used prompt
• Need explicit control over when it runs
• Prompt fits in a single Markdown file
• Want to pass dynamic arguments

Use skills when:

• Workflow requires multiple steps
• Need scripts, references, or assets
• Want automatic discovery based on context
• Multiple agents share the capability

Use system prompt when:

• Defining agent's core role
• Specifying agent-specific behavior
• Coordination with other agents

Command Structure

Basic Command File

File: .claude/commands/optimize.md

Analyze this code for performance issues and suggest optimizations:

Invocation: /optimize

Command with Frontmatter

File: .claude/commands/git-commit.md

---
tools:
argument-hint: [message]
description: Create a git commit
model: haiku
---

Create a git commit with message: $ARGUMENTS

Invocation: /git-commit "fix: resolve login bug"

Frontmatter Fields

Command Features

1. Arguments

All Arguments with $ARGUMENTS

Captures all arguments as a single string:

---
description: Fix a GitHub issue
---

Fix issue #$ARGUMENTS following our coding standards:
1. Understand the issue
2. Locate relevant code
3. Implement solution
4. Add tests
5. Prepare PR description

Usage:

/fix-issue 123 high-priority

$ARGUMENTS becomes: "123 high-priority"

Positional Arguments with $1, $2, $3

Access specific arguments individually:

---
argument-hint: [pr-number] [priority] [assignee]
description: Review pull request
---

Review PR #$1 with priority $2 and assign to $3.
Focus on security, performance, and code style.

Usage:

/review-pr 456 high alice

• $1 = "456"
• $2 = "high"
• $3 = "alice"

2. Bash Command Execution

Execute bash commands before prompt runs using ! prefix:

---
tools: Bash(git status:*), Bash(git diff:*), Bash(git log:*)
description: Create a git commit based on current changes
---

## Context

- Current git status: !`git status`
- Staged and unstaged changes: !`git diff HEAD`
- Current branch: !`git branch --show-current`
- Recent commits: !`git log --oneline -10`

## Your task

Based on the above changes, create a single git commit with an appropriate message.

Important: Must include allowed-tools with Bash tool for bash execution.

3. File References

Include file contents using @ prefix:

---
description: Review specific file implementation
---

Review the implementation in @src/utils/helpers.js and suggest improvements.

Multiple files:

Compare @src/old-version.js with @src/new-version.js and highlight differences.

4. Namespacing

Organize commands in subdirectories:

Project structure:

.claude/commands/
├── git-commit.md          # /git-commit (project)
├── frontend/
│   └── component.md       # /component (project:frontend)
└── qa/
    └── review.md          # /review (project:qa)

Personal structure:

~/.claude/commands/
├── security-review.md     # /security-review (user)
└── templates/
    └── pr-template.md     # /pr-template (user:templates)

Conflict handling:

• User vs project commands with same name: NOT SUPPORTED
• Commands in different subdirectories: ALLOWED

Creating Commands

Step 1: Determine Command Scope

Project commands (.claude/commands/):

• Shared with team via git
• Project-specific workflows
• Show "(project)" in /help

Personal commands (~/.claude/commands/):

• Personal workflows across all projects
• Not shared with team
• Show "(user)" in /help

Step 2: Design Command Interface

Name: Use lowercase, hyphens for spaces

• Good: git-commit, review-pr, fix-issue
• Bad: GitCommit, review_pr, fixIssue

Arguments: Decide on argument pattern

• Simple: Use $ARGUMENTS for all args
• Structured: Use $1, $2, $3 for specific args

Prompt: Write clear instructions

• Explain what the command should do
• Provide context (use ! for bash, @ for files)
• Specify expected output format

Step 3: Create Command File

For project command:

mkdir -p .claude/commands
cat > .claude/commands/review-pr.md << 'EOF'
---
argument-hint: [pr-number]
description: Review pull request for code quality
---

Review pull request #$ARGUMENTS:

1. Check for security vulnerabilities
2. Verify test coverage
3. Assess code quality and style
4. Suggest improvements

Provide summary in Markdown format.
EOF

For personal command:

mkdir -p ~/.claude/commands
cat > ~/.claude/commands/security-scan.md << 'EOF'
---
description: Scan code for security issues
---

Scan this code for security vulnerabilities:
- Hardcoded secrets
- Insecure dependencies
- Authentication issues
- Input validation problems
EOF

Step 4: Add Frontmatter

Enhance command with metadata:

---
# Brief description shown in /help
description: Create a git commit

# Hint shown during autocomplete
argument-hint: [message]

# Tools the command can use
tools: Bash(git add:*), Bash(git status:*), Bash(git commit:*)

# Specific model (optional)
model: haiku

# Prevent SlashCommand tool from invoking (optional)
disable-model-invocation: false
---

Your command prompt here with $ARGUMENTS

Step 5: Test Command

> /help
# Verify command appears in list

> /command-name arg1 arg2
# Test execution with arguments

Step 6: Reference in Agent Config

File: docs/agents/tracking-agent/config.yaml

name: tracking-agent

commands:
  - git-commit       # References .claude/commands/git-commit.md
  - linear-update    # References .claude/commands/linear-update.md
  - review-pr        # References .claude/commands/review-pr.md

Important: Reference by name without .md extension.

Command Examples

Example 1: Simple Command

File: .claude/commands/optimize.md

Analyze the performance of this code and suggest three specific optimizations:

Usage: /optimize

Example 2: Command with Arguments

File: .claude/commands/fix-issue.md

---
argument-hint: [issue-number]
description: Fix a GitHub issue following project standards
---

Fix issue #$ARGUMENTS following these steps:

1. **Understand**: Read issue description and requirements
2. **Locate**: Find relevant code in codebase
3. **Implement**: Create solution addressing root cause
4. **Test**: Add appropriate tests
5. **Document**: Prepare concise PR description

Follow project coding standards and conventions.

Usage: /fix-issue 123

Example 3: Git Commit Command

File: .claude/commands/git-commit.md

---
tools: Bash(git add:*), Bash(git status:*), Bash(git commit:*), Bash(git diff:*), Bash(git log:*)
description: Create a git commit based on current changes
---

## Context

- Current git status: !`git status`
- Staged and unstaged changes: !`git diff HEAD`
- Current branch: !`git branch --show-current`
- Recent 10 commits: !`git log --oneline -10`

## Your Task

Based on the above git context, create a single git commit:

1. **Analyze changes**: Review the git diff
2. **Choose files**: Select relevant files to stage
3. **Write message**: Create commit message following format:
   - Type: `feat|fix|docs|refactor|test|chore`
   - Format: `<type>: <description>`
   - Example: `feat: add user authentication`
4. **Create commit**: Execute git commands to stage and commit

**Important**: Follow this repository's commit message style (see recent commits).

Usage: /git-commit

Example 4: Positional Arguments

File: .claude/commands/review-pr.md

---
argument-hint: [pr-number] [priority] [assignee]
description: Review pull request with priority and assignment
---

Review pull request #$1:

**Priority**: $2
**Assign to**: $3

**Review checklist**:
1. Security vulnerabilities
2. Performance issues
3. Code style violations
4. Test coverage
5. Documentation completeness

Provide detailed review comments with severity levels.

Usage: /review-pr 456 high alice

Example 5: File Reference Command

File: .claude/commands/explain-file.md

---
argument-hint: <file-path>
description: Explain implementation of a specific file
---

Explain the implementation in @$ARGUMENTS:

1. **Purpose**: What does this file do?
2. **Structure**: How is the code organized?
3. **Key functions**: What are the main functions/classes?
4. **Dependencies**: What does it depend on?
5. **Usage**: How is it used elsewhere?

Provide clear, beginner-friendly explanation.

Usage: /explain-file src/utils/helpers.js

Agent Config Integration

Declaring Commands

File: docs/agents/qa-agent/config.yaml

name: qa-agent
description: Quality assurance and validation agent

# Skills for complex workflows
skills:
  - security-validation
  - test-standards
  - code-quality-standards

# Commands for explicit invocations
commands:
  - review-pr         # Quick PR review
  - security-scan     # Security check
  - test-coverage     # Coverage analysis

ref_docs:
  - test-audit-protocol.md
  - traycer-coordination-guide.md

Commands in Agent Workflow

Agent can use commands:

1. Manually: User invokes /review-pr 456
2. Programmatically: Via SlashCommand tool (if enabled)

SlashCommand tool requirements:

• Command must have description frontmatter
• Not disabled via disable-model-invocation: true
• Character budget not exceeded (15,000 chars default)

Framework Integration (Traycer Enforcement Framework)

When creating commands for the Traycer enforcement framework:

Command Location

Commands live in: .claude/commands/<command-name>.md

Example:

.claude/commands/
├── git-commit.md
├── review-pr.md
└── your-new-command.md

Agent Assignment

When called standalone (user asks to create a command):

1. Ask user which agents should use this command
2. Suggest agents based on command's purpose:
   • Git operations → tracking-agent
   • Code validation → qa-agent
   • Implementation → action-agent
   • Workflow coordination → workflow-upgrade-assistant
3. List suggested agents for user confirmation

When called by agent-builder:

• Agent context provided automatically
• No need to ask about agent assignment

Updating Agent Config

After creating the command, update each agent's config.yaml:

File: docs/agents/<agent-name>/config.yaml

Add command name (without .md extension) to commands: section:

commands:
  - existing-command-1
  - existing-command-2
  - your-new-command    # Add here (no .md extension)

Complete Workflow

1. Create command file: .claude/commands/<command-name>.md
2. Add frontmatter (description, argument-hint, allowed-tools)
3. Write command prompt with $ARGUMENTS or $1, $2, $3
4. Determine which agents use this command (ask user or use agent-builder context)
5. Update each agent's config.yaml to reference the command
6. Test command with /command-name args

Best Practices

1. Keep Commands Simple

Good: Single-purpose, clear prompt

Review this code for security vulnerabilities

Bad: Multiple unrelated tasks

Review code for security, performance, style, tests, documentation, and deployment issues

2. Use Descriptive Names

Good: review-pr, git-commit, fix-issue Bad: r, gc, fix

3. Provide Context with Bash

Good: Include relevant git context

- Current branch: !`git branch --show-current`
- Recent commits: !`git log --oneline -5`

Bad: Assume context is known

Create a commit

4. Document Expected Arguments

Good: Clear argument hint

---
argument-hint: [pr-number] [priority] [assignee]
---

Bad: No hint, unclear usage

5. Use Frontmatter

Good: Complete metadata

---
description: Review pull request
argument-hint: [pr-number]
tools: Bash(gh pr view:*)
---

Bad: No frontmatter (description defaults to first line)

6. Namespace Organization

Good: Organized by function

.claude/commands/
├── git/
│   ├── commit.md
│   ├── branch.md
│   └── push.md
├── qa/
│   ├── review.md
│   └── test.md

Bad: Flat structure with many commands

Common Patterns

Pattern 1: Context-Rich Commands

Include bash output for context:

---
tools: Bash(git status:*), Bash(git diff:*)
---

## Context
- Git status: !`git status --short`
- Changes: !`git diff --stat`

## Task
[Your instructions based on context]

Pattern 2: Structured Arguments

Use positional args for structured input:

---
argument-hint: [component] [action] [target]
---

$1: Component name
$2: Action to perform
$3: Target file/directory

Execute $2 on $1 targeting $3.

Pattern 3: File-Based Commands

Reference specific files:

Review @$ARGUMENTS for:
1. Code quality
2. Security issues
3. Performance concerns

Usage: /review src/auth.js

Pattern 4: Multi-Step Workflows

Break complex tasks into steps:

Execute deployment workflow for $ARGUMENTS:

**Phase 1: Pre-deployment**
- [ ] Run tests
- [ ] Security scan
- [ ] Build artifacts

**Phase 2: Deployment**
- [ ] Deploy to staging
- [ ] Verify health checks
- [ ] Deploy to production

**Phase 3: Post-deployment**
- [ ] Monitor metrics
- [ ] Verify functionality
- [ ] Document deployment

Integration with Agent Builder

When using Agent Builder skill to create an agent:

1. Agent Builder analyzes requirements
2. Identifies frequently-used prompts → Candidates for commands
3. Calls Command Creator to define command specifications
4. Creates command files in .claude/commands/
5. References commands in config.yaml

Command Creator provides:

• Command file structure
• Frontmatter recommendations
• Argument patterns
• Integration with agent workflow

Quick Reference

Creating commands:

1. Choose scope (project vs personal)
2. Design command interface (name, arguments)
3. Write prompt in Markdown
4. Add frontmatter (description, argument-hint, allowed-tools)
5. Test with /command-name args
6. Reference in agent config.yaml

Command naming:

• Use: git-commit, review-pr, fix-issue
• Avoid: GitCommit, review_pr, fixIssue

File location:

• Project: .claude/commands/command-name.md
• Personal: ~/.claude/commands/command-name.md

Config reference:

commands:
  - command-name  # No .md extension

Resources

For detailed patterns and examples, see references/command-examples.md:

• Complete command specifications from existing agents
• Argument design patterns
• Frontmatter best practices
• Integration examples