An OpenClaw skill that guides the creation, review, and optimization of agent skills following proven best practices.
This is a meta-skill — it teaches LLMs how to build high-quality skills for the OpenClaw ecosystem. Based on deep analysis of top-performing skills on ClawHub (find-skills, tavily-search, self-improving-agent, and others).
- Create new skills — step-by-step workflow from requirements gathering to quality review
- Review existing skills — 24-point quality checklist across 6 categories
- Retrofit old skills — 8-step process to bring any skill up to best practices
- Add API integrations — patterns for wrapping external HTTP APIs with curl scripts
- Description Writing Formula — the single most important factor for reliable skill triggering
- Complexity Tiers (Simple / Medium / Complex) — choose the right structure for your skill's scope
- Tables over prose — LLMs find table rows faster than parsing paragraphs
- LLM as execution engine — write instructions, not code logic
clawhub install jau123/skills-creatorOr manually:
git clone https://github.com/jau123/skills-creator.git ~/.openclaw/skills/skills-creatorskills-creator/
├── SKILL.md # Core instructions (~275 lines)
├── references/
│ ├── best-practices.md # Design guidelines with real-world examples
│ └── quality-checklist.md # 24-point checklist + retrofit process
└── assets/
└── skill-template.md # Copy-paste starter template
Once installed, the skill activates when you ask things like:
- "Create a skill for X"
- "Review my SKILL.md"
- "How do I write a good skill description?"
- "Optimize my existing skill"
- "What makes a good OpenClaw skill?"
The full checklist covers 24 checks across 6 categories:
| Category | Checks | What it covers |
|---|---|---|
| Frontmatter | 6 | name format, description quality, metadata JSON |
| Description | 4 | trigger phrases, action verbs, length |
| Content Structure | 5 | tables, workflow modes, examples |
| Self-Consistency | 3 | follows own rules, no conflicts |
| Security | 3 | VirusTotal compliance, pinned versions |
| Supporting Files | 3 | references/, scripts/, assets/ |
Claude Code ships with skill-creator and plugin-dev — great tools for the Claude Code ecosystem. This skill serves a different audience and purpose:
| skills-creator (this) | Claude Code skill-creator |
Claude Code plugin-dev |
|
|---|---|---|---|
| Platform | OpenClaw / ClawHub | Claude Code | Claude Code |
| Focus | Skill design methodology | Eval-driven skill testing | Plugin scaffolding |
| Core strength | Description Writing Formula, 24-point quality checklist, complexity tiers | Automated parallel evals, variance analysis, benchmarking | 8-phase plugin creation, hooks, agents, commands |
| Platform knowledge | {baseDir}, single-line metadata JSON, clawdbot key, VirusTotal compliance, ClawHub publishing |
Claude Code skill format | Claude Code plugin format |
| Approach | Design-first — get structure and description right before writing | Test-first — measure triggering accuracy with evals | Build-first — scaffold then validate |
In short: Use this skill when building for OpenClaw/ClawHub. Use Claude Code's tools when building for Claude Code.
MIT-0 (MIT No Attribution) — as required by ClawHub.