Complete methodology for designing, testing, and quality-checking OpenClaw Skills.
OpenClaw Skill — works with OpenClaw AI agents
A meta-skill that guides AI agents through the full lifecycle of Skill development: from initial design and trigger scenario definition, through implementation and testing, to quality inspection and publishing. Distilled from hands-on experience building 16+ production Skills. Includes an automated linter (skill_lint.py) for structural validation.
# Install via ClawHub (recommended)
openclaw skill install skill-design-guide
# Or clone this repo
git clone https://github.com/rrrrrredy/skill-design-guide.git ~/.openclaw/skills/skill-design-guide- Skill = runtime context container, not documentation
- 4 lifecycle stages: static scan (description only) → dynamic injection (full SKILL.md) → execution (ReAct loop) → fold & forget
- SKILL.md rules decay as conversation progresses — hard constraints belong in system prompt
- Design principles: Trigger-first design, token budget management, description as routing
- Quality checklist: 12-point inspection covering structure, content, safety, and compliance
- Automated linter:
skill_lint.pyvalidates SKILL.md structure, version consistency, line count - Common gotchas catalog: Collected pitfalls from 16+ Skill development cycles
- Testing methodology: Trigger word testing, execution path verification, fallback validation
Trigger phrases:
- "新建 skill"、"设计 skill"、"skill 质检"
- "评审 skill"、"skill 设计原则"、"创建 skill"
- "skill 最佳实践"、"skill review"、"skill lint"
skill-design-guide/
├── SKILL.md # Main skill definition & workflow
├── scripts/
│ └── skill_lint.py # Automated SKILL.md linter
├── references/
│ ├── design-principles.md # Core design principles (detailed)
│ ├── quality-checklist.md # 12-point QA checklist
│ ├── common-gotchas.md # Known pitfalls & solutions
│ └── testing-guide.md # Testing methodology
└── .gitignore
- OpenClaw agent runtime
- Python 3 (for skill_lint.py)