Best practices for skill structure and types. Use when creating skills.
skill-name/
├── SKILL.md # Required: core instructions (<500 words)
├── scripts/ # Optional: executable code
├── references/ # Optional: detailed docs (load on-demand)
└── assets/ # Optional: templates, images
| Type | Freedom | Use when | Structure |
|---|---|---|---|
| Knowledge | High | Multiple approaches, context-dependent | SKILL.md + references/ |
| Hybrid | Medium | Guidance + scripts needed | SKILL.md + scripts/ + references/ |
| Tool | Low | Deterministic, repeatable ops | SKILL.md + scripts/ |
| Expert | Very Low | Complex internals, undocumented APIs | Full + validation/ |
Rule: If behavior MUST happen, use hooks. Documentation cannot enforce.
| Requirement Type | Example | Solution |
|---|---|---|
| MUST/REQUIRED | "MUST use Skill tool" | PreToolUse hook to warn/block |
| Validation | "Schema must be valid" | PostToolUse hook to validate |
| Prevention | "Never commit secrets" | PreToolUse hook to block |
| Guidance | "Consider using X" | Documentation only (OK) |
If keyword in ["MUST", "REQUIRED", "CRITICAL", "강제", "반드시"]:
→ Implement as Hook (documentation alone WILL be ignored)
If keyword in ["should", "consider", "recommend"]:
→ Documentation is sufficient
Before finalizing skill:
Example:
---
name: skill-name
description: What + when to use (triggers skill loading)
allowed-tools: ["Tool1", "Tool2"]
---
# Skill Name
[2-3 sentence overview]
## Quick Start
[Fastest path]
## Workflow
1. Step 1
2. Step 2
## Scripts
| Script | Purpose | Usage |
|--------|---------|-------|
## Key Principles
- Principle 1
- Principle 2
For advanced: [references/advanced.md]
# Knowledge (read-only)
allowed-tools: ["Read", "Grep", "Glob"]
# Hybrid (guidance + generation)
allowed-tools: ["Read", "Write", "Grep", "Glob", "Bash"]
# Tool (file manipulation)
allowed-tools: ["Read", "Write", "Bash"]
