LuizEduPP Skill Creator
Create modular, self-contained packages that extend AI capabilities with specialized knowledge and workflows across different hosts.
Core Principles
Concise is Key
Challenge each piece: "Does the AI really need this?" Prefer concise examples over verbose explanations.
Progressive Disclosure
- Metadata (~100 words) - Always loaded
- SKILL.md body (<500 lines) - When triggered
- Bundled resources - As needed
Skill Structure
skill-name/
├── SKILL.md (required)
│ ├── YAML frontmatter (name, description)
│ └── Markdown instructions
└── Bundled Resources (optional)
├── scripts/ - Executable code
├── references/ - Documentation
└── assets/ - Templates, icons
SKILL.md Format
---
name: my-skill
description: What it does AND when to use it (critical for triggering)
---
# Title
[Core instructions - imperative form]
## Sections as needed
Best Practices
| Do | Don't |
|---|---|
| Clear, comprehensive description | Vague one-liner |
| Imperative form ("Create", "Run") | Passive voice |
| Concise examples | Verbose explanations |
| Reference files for details | Everything in SKILL.md |
| Test scripts by running them | Assume they work |
| Use host-agnostic paths and wording | Hard-code one assistant's directory layout |
Portability Rules
- Write the skill in English unless a translated artifact is explicitly required
- Keep the main workflow independent from any single IDE, editor, CLI, or chat app
- Use placeholders such as
<skill-root>,<workspace>, and<output-dir>in examples - Put vendor-specific notes in a short compatibility section instead of the core flow
- Mention a specific assistant only when the skill itself is about that assistant
What NOT to Include
- README.md, CHANGELOG.md, INSTALLATION.md
- User-facing documentation
- Setup/testing procedures
Creation Process
- Understand - Get concrete usage examples
- Plan - Identify reusable scripts/references/assets
- Initialize - Create skill directory with SKILL.md
- Edit - Implement resources, write instructions
- Test - Verify with real usage
- Iterate - Refine based on feedback