# skill-creator

> Meta-skill for creating properly structured OpenClaw/Anthropic-format skills. Use when building new skills or refactoring existing ones.

- Author: Clawd Bot
- Repository: BAiSEDagent/openclaw-skills
- Version: 20260208142402
- Stars: 0
- Forks: 0
- Last Updated: 2026-02-08
- Source: https://github.com/BAiSEDagent/openclaw-skills
- Web: https://mule.run/skillshub/@@BAiSEDagent/openclaw-skills~skill-creator:20260208142402

---

---
name: skill-creator
description: "Meta-skill for creating properly structured OpenClaw/Anthropic-format skills. Use when building new skills or refactoring existing ones."
metadata:
  openclaw:
    emoji: "🛠️"
---

# Skill Creator

Guide for creating well-structured skills that follow Anthropic's best practices.

## Core Principles

1. **Concise is key** — Context window is shared. Only add what Claude doesn't already know.
2. **Progressive disclosure** — Metadata always loaded, SKILL.md on trigger, references on demand.
3. **Appropriate freedom** — Narrow bridge = specific guardrails. Open field = many routes.

## Skill Structure

```
skill-name/
├── SKILL.md          (required, <500 lines)
│   ├── YAML frontmatter (name, description, metadata)
│   └── Markdown body (workflow + navigation)
├── references/       (detailed docs, loaded on demand)
├── scripts/          (deterministic code, executed without reading)
└── assets/           (templates, images, used in output)
```

## SKILL.md Rules

- **Under 500 lines.** If approaching this, split into references.
- **YAML frontmatter**: `name` and `description` are required. Description determines when skill triggers — be clear and comprehensive.
- **Body**: Essential workflow + links to reference files with "when to read" guidance.
- **No auxiliary files**: No README.md, CHANGELOG.md, INSTALLATION_GUIDE.md.
- **No duplication**: Info lives in SKILL.md OR references, not both.

## Reference Files

- `references/` — Detailed schemas, API docs, examples. Loaded only when needed.
- Keep one level deep from SKILL.md (no nested references).
- Files >100 lines should have a table of contents.
- Include grep patterns in SKILL.md for very large reference files.

## When to Use Each Resource Type

| Type | When | Example |
|------|------|---------|
| **SKILL.md inline** | Core workflow, always needed | Payment flow steps |
| **references/** | Detailed docs, needed per-task | API schemas, chain configs |
| **scripts/** | Same code rewritten repeatedly | PDF rotation, deployment |
| **assets/** | Files used in output | Templates, icons, fonts |

## Progressive Disclosure Patterns

**Pattern 1: High-level guide + references**
```markdown
## Quick start
[essential example]

## Advanced
- **Feature X**: See [references/feature-x.md](references/feature-x.md)
```

**Pattern 2: Domain-specific split**
```
references/
├── aws.md    (read when deploying to AWS)
├── gcp.md    (read when deploying to GCP)
```

**Pattern 3: Conditional details**
```markdown
## Basic usage
[inline instructions]

**For advanced feature**: See [references/advanced.md](references/advanced.md)
```

## Creation Checklist

1. [ ] Concrete examples of how skill will be used
2. [ ] Identify reusable resources (scripts, references, assets)
3. [ ] Write SKILL.md with frontmatter + essential workflow
4. [ ] Move detailed content to references/
5. [ ] Verify SKILL.md < 500 lines
6. [ ] Each reference file linked from SKILL.md with "when to read" context
7. [ ] Test: Can Claude use this skill with ONLY SKILL.md for simple cases?

## Anti-Patterns

- ❌ Walls of text explaining what Claude already knows
- ❌ SKILL.md over 500 lines
- ❌ Duplicate info in SKILL.md and references
- ❌ README.md or other auxiliary docs
- ❌ Deeply nested reference hierarchies
- ❌ Reference files with no link from SKILL.md

## Bundled Scripts

### `scripts/init_skill.py` — Initialize a new skill
```bash
python3 scripts/init_skill.py <skill-name> --path <output-directory>
```
Creates skill directory with SKILL.md template, example resource directories, and placeholder files.

### `scripts/package_skill.py` — Package a skill for distribution
```bash
python3 scripts/package_skill.py <skill-directory>
```
Validates and packages the skill.

## Reference Files

- **[references/workflows.md](references/workflows.md)** — Patterns for multi-step processes, conditional logic, sequential workflows
- **[references/output-patterns.md](references/output-patterns.md)** — Patterns for specific output formats, quality standards, validation

Read these when building skills that involve complex workflows or need specific output guarantees.

## OpenClaw-Specific Conventions

For OpenClaw skills, frontmatter includes:
```yaml
metadata:
  openclaw:
    emoji: "🔧"
    requires: {}
```

Skills live in the workspace skills directory and are referenced by name in AGENTS.md or loaded via the skills system.