Skill Package Format
Every skill in Clawford University follows a consistent structure. This makes skills predictable, testable, and composable.
Directory Structure
@clawford/<skill-name>/
├── package.json # npm package config
├── manifest.json # metadata: category, tags, dependencies
├── SKILL.md # role definition, triggers, capabilities
├── knowledge/ # domain knowledge files
│ ├── domain.md # core domain expertise
│ ├── best-practices.md # proven patterns
│ └── anti-patterns.md # common mistakes to avoid
├── strategies/ # behavioral strategies
│ └── main.md # primary strategy
└── tests/
├── smoke.json # quick validation (3-5 tasks)
└── benchmark.json # full benchmark (10 tasks)manifest.json
The manifest defines metadata used by the CLI, registry, and marketplace.
JSON
{
400">"name": 400">"@clawford/google-search",
400">"version": 400">"0.1.0",
400">"description": 400">"Search query optimization and result ranking",
400">"category": 400">"information-retrieval",
400">"tags": [400">"search", 400">"web", 400">"research"],
400">"capabilities": [
400">"Construct advanced search queries",
400">"Rank results by relevance",
400">"Assess source credibility"
],
400">"triggers": [400">"search for", 400">"find information", 400">"look up"],
400">"dependencies": {},
400">"compatibility": {
400">"openclaw": ">=0.5.0400">",
"claudeCode400">": ">=1.0.0400">"
},
"benchmarkDimension400">": "information-retrieval400">",
"expectedImprovement": 30
}SKILL.md
The SKILL.md file defines the agent's role when this skill is activated. It uses YAML frontmatter for metadata and Markdown for the role definition.
Sections
- Role — What the agent becomes when this skill activates
- Capabilities — What the agent can do (numbered list)
- Constraints — What the agent must not do (guard rails)
- Activation — When and how the skill triggers
Knowledge Files
Knowledge files inject domain expertise into the agent's context:
domain.md— Core domain knowledgebest-practices.md— Proven patterns and techniquesanti-patterns.md— Common mistakes and how to avoid them
Strategies
Strategies define behavioral patterns — how the agent should approach tasks.
Tests
smoke.json— 3-5 quick validation tasksbenchmark.json— 10 standardized tasks measuring improvement