# skill-creator

> Create effective Claude Code skills, plugins, and marketplaces with proper structure, frontmatter, and best practices. Use when creating new skills, plugins, marketplaces, converting workflows to skills, or improving existing skills.

- Author: akatz-ai
- Repository: akatz-ai/akatz-skills
- Version: 20260124220033
- Stars: 0
- Forks: 0
- Last Updated: 2026-02-06
- Source: https://github.com/akatz-ai/akatz-skills
- Web: https://mule.run/skillshub/@@akatz-ai/akatz-skills~skill-creator:20260124220033

---

---
name: skill-creator
description: Create effective Claude Code skills, plugins, and marketplaces with proper structure, frontmatter, and best practices. Use when creating new skills, plugins, marketplaces, converting workflows to skills, or improving existing skills.
---

# Skill Creator

Create Claude Code skills that are concise, well-structured, and effective.

## Workflow

Copy this checklist when creating a skill:

```
Skill Creation Progress:
- [ ] 1. Define purpose and triggers
- [ ] 2. Create directory structure
- [ ] 3. Write SKILL.md with frontmatter
- [ ] 4. Add supporting files (if needed)
- [ ] 5. Test with real usage
- [ ] 6. Iterate based on behavior
```

## Step 1: Define Purpose

Before writing, answer:
- **What does this skill do?** (one sentence)
- **When should it trigger?** (specific keywords/scenarios)
- **Who invokes it?** (user via `/`, Claude automatically, or both)

## Step 2: Create Structure

**Creating a plugin/marketplace?** STOP. Read [plugin-structure.md](plugin-structure.md) first.

**Standalone skill locations:**
| Scope | Path |
|-------|------|
| Personal | `~/.claude/skills/<name>/SKILL.md` |
| Project | `.claude/skills/<name>/SKILL.md` |
| Plugin | `<plugin>/skills/<name>/SKILL.md` |

```bash
mkdir -p <location>/<skill-name>
```

## Step 3: Write SKILL.md

**Required frontmatter:**
```yaml
---
name: my-skill-name
description: What it does and when to use it. Be specific with trigger words.
---
```

**Optional frontmatter fields:**
- `disable-model-invocation: true` - Only user can invoke (use for side effects like deploy, commit)
- `user-invocable: false` - Only Claude can invoke (background knowledge)
- `allowed-tools: Read, Grep, Glob` - Restrict available tools
- `context: fork` - Run in isolated subagent
- `agent: Explore` - Which agent type for forked context

**Body content guidelines:**
- Keep under 500 lines
- Assume Claude already knows common concepts
- Use workflows with checklists for multi-step tasks
- Include validation/feedback loops for critical operations

## Step 4: Add Supporting Files

For skills exceeding 500 lines, use progressive disclosure:

```
my-skill/
├── SKILL.md           # Overview + navigation (required)
├── reference.md       # Detailed docs (loaded when needed)
├── examples.md        # Usage examples
└── scripts/
    └── helper.py      # Utility scripts
```

Reference files from SKILL.md:
```markdown
**For API details**: See [reference.md](reference.md)
**Examples**: See [examples.md](examples.md)
```

## Step 5: Test

1. Load the skill: `claude --plugin-dir ./path/to/plugin`
2. Test automatic triggering with natural requests
3. Test direct invocation with `/skill-name`
4. Verify Claude finds the right information

## Step 6: Iterate

Observe Claude's behavior:
- Does it trigger when expected?
- Does it follow instructions correctly?
- Does it access supporting files appropriately?

Refine based on actual usage, not assumptions.

---

## Quick Reference

**For plugin/marketplace structure:** See [plugin-structure.md](plugin-structure.md) - **READ FIRST** when creating plugins/marketplaces
**For detailed patterns:** See [patterns.md](patterns.md)
**For anti-patterns to avoid:** See [anti-patterns.md](anti-patterns.md)
**For evaluation setup:** See [evaluations.md](evaluations.md)

## String Substitutions

Available in skill content:
- `$ARGUMENTS` - User input after skill name
- `${CLAUDE_SESSION_ID}` - Current session ID
- `!`command`` - Execute shell command, inject output