SlashCommandCreator Skill

Purpose

Create production-ready custom slash commands following ai-agents quality standards.

Triggers

• create slash command for [purpose]
• SlashCommandCreator: [description]
• design slash command that [does something]

When to Use

• User requests "create slash command for [purpose]"
• Repetitive prompts identified in workflow
• Converting manual workflows to automation
• Need reusable, documented command patterns

5-Phase Workflow

Phase 1: Discovery & Analysis

Agent Mode: Analyst

Tasks:

1. Clarify user intent: What prompt is being repeated?
2. Search existing commands: ls .claude/commands/**/*.md
3. Decision: Slash command vs skill (see decision matrix in CLAUDE.md)
4. Apply 11 thinking models from skillcreator framework
5. Document findings in .agents/analysis/slashcommand-[name]-analysis.md

Deliverable: Analysis document with recommendation

Phase 2: Design

Agent Mode: Architect

Tasks:

1. Command naming (namespace conventions)
2. Argument design:
   • Simple commands: use $ARGUMENTS
   • Complex commands: use $1, $2, $3 (positional)
3. Frontmatter schema:
   • description (trigger-based per creator-001)
   • argument-hint (if using arguments)
   • allowed-tools (if using bash commands with ! or file references with @)
   • model (opus for complex reasoning)
   • disable-model-invocation (if pure prompt template)
4. Dynamic context evaluation:
   • Bash execution (!git log --oneline -5)
   • File references (@.agents/HANDOFF.md)
5. Extended thinking evaluation:
   • Add ultrathink keyword for complex reasoning (>5 steps)
   • Token budget consideration (<31,999 tokens)

Deliverable: Design specification with frontmatter + prompt

Phase 3: Multi-Agent Validation

Agent Mode: Orchestrator (coordinates 4 agents)

Agents:

1. Security:

   • Review allowed-tools constraints
   • Flag overly permissive wildcards
   • Verify bash commands are safe

2. Architect:

   • Check for duplication (similar existing commands)
   • Verify appropriate scope (not too broad/narrow)
   • Validate namespace conventions

3. Independent-Thinker:

   • Challenge necessity: Is this really needed?
   • Propose alternatives
   • Question assumptions

4. Critic:

   • Frontmatter completeness check
   • Trigger-based description validation
   • Argument-hint clarity

Unanimous Approval Required: All 4 agents must approve.

Invocation Pattern:

# Security review
Task(subagent_type="security", prompt="Review allowed-tools for command: [spec]")

# Architecture review
Task(subagent_type="architect", prompt="Check for duplication: [spec]")

# Challenge necessity
Task(subagent_type="independent-thinker", prompt="Is this command truly needed? [spec]")

# Frontmatter completeness
Task(subagent_type="critic", prompt="Validate frontmatter completeness: [spec]")

Deliverable: Validation report with approvals or revision requests

Phase 4: Implementation

Agent Mode: Implementer

Tasks:

1. Run pwsh .claude/skills/slashcommandcreator/scripts/New-SlashCommand.ps1
2. Create .claude/commands/[namespace]/[command].md
3. Write frontmatter + prompt body
4. Test invocation with sample arguments
5. Update command catalog (if exists)

Deliverable: Working slash command file

Phase 5: Quality Gates (Automatic)

Agent Mode: Implementer (automatic validation)

Tasks:

1. Run pwsh .claude/skills/slashcommandcreator/scripts/Validate-SlashCommand.ps1 -Path [file]
2. Fix violations if any
3. Re-run validation until exit code 0
4. Commit with conventional commit message

Deliverable: Validated slash command ready for use

Invocation Examples

SlashCommandCreator: create command for exporting Forgetful memories to JSON

SlashCommandCreator: design slash command for running security audit

create slash command that summarizes recent PR comments

Decision Matrix: Slash Command vs Skill

Use Slash Command When:

• Prompt is <200 lines
• No multi-step conditional logic
• Simple argument substitution
• No external script orchestration

Use Skill When:

• Prompt is >200 lines
• Multi-agent coordination required
• Complex PowerShell logic
• Requires Pester tests

Verification/Success Criteria

Before marking complete:

• Frontmatter has description (trigger-based)
• Frontmatter has argument-hint (if uses arguments)
• Frontmatter has allowed-tools (if uses bash/file refs)
• No overly permissive wildcards in allowed-tools
• Description follows trigger-based pattern (creator-001)
• File is <200 lines (or converted to skill)
• Passes markdownlint-cli2 validation
• Passes Validate-SlashCommand.ps1 validation
• Tested with sample arguments
• Committed with conventional commit message

Success Criteria:

Quality Gates Checklist

All checks from Verification section plus:

• Multi-agent approval (security, architect, independent-thinker, critic)
• No duplication with existing commands
• Appropriate scope (not too broad/narrow)
• Frontmatter completeness validated

References

• .agents/analysis/custom-slash-commands-research.md
• .agents/planning/slashcommandcreator-skill-spec.md
• .serena/memories/slashcommand-best-practices.md