slash-command-factory

Slash Command Factory

A comprehensive system for generating production-ready Claude Code slash commands through a simple question-based workflow.

Self-Evolving Skill: This skill improves through use. If instructions are wrong, parameters drifted, or a workaround was needed — fix this file immediately, don't defer. Only update for real, reproducible issues.

When to Use This Skill

• Creating new custom slash commands for Claude Code
• Generating command templates from presets
• Validating slash command YAML frontmatter syntax
• Organizing commands into proper folder structures
• Converting workflows into reusable slash commands

---

Overview

This skill helps you create custom slash commands for Claude Code by:

• Asking 5-7 straightforward questions about your command needs
• Generating complete command .md files with proper YAML frontmatter
• Providing 10 powerful preset commands for common use cases
• Validating command format and syntax
• Creating well-organized folder structures
• Offering installation guidance

Output: Complete slash commands ready to use in Claude Code

---

Command Structure Patterns

Three official patterns from Anthropic documentation:

Pattern	Name	Best For	Structure
A	Simple	Straightforward tasks (code review, file updates)	Context -> Task
B	Multi-Phase	Complex discovery (audits, system mapping)	Discovery -> Analysis -> Task
C	Agent-Style	Specialized roles (experts, orchestrators)	Role -> Process -> Guidelines

Full templates and when-to-use guidance: Command Patterns

---

Naming Convention

All slash command files MUST follow kebab-case: [verb]-[noun].md

Rules: Lowercase only, 2-4 words, [a-z0-9-] characters, no underscores/camelCase

Input	Output
"Review pull requests"	pr-review.md
"Generate API documentation"	api-document.md
"Audit security compliance"	security-audit.md

Full conversion algorithm and examples: Naming Convention

---

Bash Permission Rules

Critical: Blanket Bash permission is prohibited. Must use subcommand-level specificity.

# WRONG - too broad
allowed-tools: Bash(git:*)

# CORRECT - subcommand-level
allowed-tools: Bash(git add:*), Bash(git commit:*), Bash(git push:*)

# OK - simple commands without subcommand hierarchies
allowed-tools: Bash(cp:*), Bash(mkdir -p:*), Bash(date:*)

Command Type	Bash Permissions	Example Commands
Git Commands	git status, git diff, git log, git branch	code-review, commit-assist
Discovery	find, tree, ls, du	codebase-analyze, structure-map
Analysis	grep, wc, head, tail, cat	search-code, count-lines
Data Processing	awk, sed, sort, uniq	parse-data, format-output

Full patterns and selection guide: Bash Permissions

---

Two Paths to Generate Commands

Path 1: Quick-Start Presets (30 seconds)

Choose from 10 powerful preset commands:

#	Command	Purpose
1	/research-business	Comprehensive market research and competitive analysis
2	/research-content	Multi-platform content trend analysis and SEO strategy
3	/medical-translate	Medical terminology to 8th-10th grade (German/English)
4	/compliance-audit	HIPAA/GDPR/DSGVO compliance validation
5	/api-build	Complete API integration code with tests
6	/test-auto	Auto-generate comprehensive test suites
7	/docs-generate	Automated documentation creation
8	/knowledge-mine	Extract and structure insights from documents
9	/workflow-analyze	Analyze and optimize business processes
10	/batch-agents	Launch and coordinate multiple agents

Full YAML configs and details: Preset Commands

Path 2: Custom Command (5-7 Questions)

Create a completely custom command by answering questions about:

1. Purpose - What should the command do?
2. Arguments - Auto-determined; all flags get mandatory short forms (-b|--branch)
3. Tools - Which Claude Code tools (Read, Write, Bash, Grep, Task, etc.)
4. Agents - Does it need to launch specialized agents?
5. Output - Analysis, files, action, or report?
6. Model - Default, Sonnet, Haiku, or Opus? (optional)
7. Features - Bash execution, file references, context gathering? (optional)

Full question scripts and argument conventions: Question Flow

---

Generation & Installation

After collecting answers, the skill:

1. Generates YAML frontmatter with proper allowed-tools
2. Generates command body with purpose-specific instructions
3. Creates folder structure under generated-commands/[command-name]/
4. Validates format (YAML, arguments, tools, organization)
5. Provides installation instructions

# Install to project
cp generated-commands/[command-name]/[command-name].md .claude/commands/

# Install globally
cp generated-commands/[command-name]/[command-name].md ~/.claude/commands/

Plugin invocation: /plugin-name:command-name [arguments]

Full process, folder structure, and plugin invocation rules: Generation Process

---

Validation

Every generated command is validated for:

• Valid YAML frontmatter (proper syntax, required fields)
• Correct argument format ($ARGUMENTS, not $1 $2 $3)
• Short forms for all flags (mandatory 1-2 letter shortcuts)
• Bash subcommand-level specificity (no blanket Bash)
• Clean folder organization

If validation fails, you get specific fix instructions.

Full validation checklist, best practices, and troubleshooting: Validation Reference

---

Quick Reference

Usage

@slash-command-factory
Use the /research-business preset

@slash-command-factory
Create a custom command for analyzing customer feedback

Key Rules

Rule	Detail
Arguments	Always $ARGUMENTS (never $1, $2)
Flag short forms	Mandatory for all flags (-b|--branch)
Bash permissions	Subcommand-level only (Bash(git add:*))
File naming	kebab-case, 2-4 words
Output location	./generated-commands/[command-name]/

Ecosystem Integration

Works with: factory-guide, skills-guide, prompts-guide, agents-guide

More examples and integration details: Usage Examples

Post-Execution Reflection

After this skill completes, check before closing:

1. Did the command succeed? — If not, fix the instruction or error table that caused the failure.
2. Did parameters or output change? — If the underlying tool's interface drifted, update Usage examples and Parameters table to match.
3. Was a workaround needed? — If you had to improvise (different flags, extra steps), update this SKILL.md so the next invocation doesn't need the same workaround.

Only update if the issue is real and reproducible — not speculative.