creating-skills（创造技能）

Creating Skills

This skill provides guidance for creating effective skills using TDD methodology.

Core Principles

Skills ARE TDD for Documentation

1. Write failing test first - Run pressure scenario WITHOUT skill, document rationalizations
2. Write minimal skill - Address those specific failures
3. Watch pass, refactor - Find new loopholes, plug them, re-test

If you didn't watch an agent fail without it, you don't know what to teach.

Concise is Key

The context window is a public good. Default assumption: Claude is already very smart. Only add context Claude doesn't already have. Challenge each piece: "Does Claude really need this explanation?"

Prefer concise examples over verbose explanations.

Set Appropriate Degrees of Freedom

Match specificity to task fragility:

Quick Reference

When to Create

Create when: Technique wasn't obvious, applies broadly, others benefit Don't create for: One-offs, standard practices, project-specific (use CLAUDE.md)

Anatomy of a Skill

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter metadata (required)
│   │   ├── name: (required)
│   │   └── description: (required)
│   └── Markdown instructions (required)
└── Bundled Resources (optional)
    ├── scripts/          - Executable code (Python/Bash/etc.)
    ├── references/       - Documentation loaded as needed
    └── assets/           - Files used in output (templates, icons, fonts)

Frontmatter

---
name: skill-name-with-hyphens
description: Use when [triggering conditions only, never workflow summary]
---

• name: Letters, numbers, hyphens only
• description: Primary triggering mechanism. Start "Use when...", max 1024 chars, third person

- Include what the Skill does AND specific triggers/contexts for when to use it - Include ALL "when to use" here - NOT in the body (body loads only after triggering) - ❌ Workflow summaries create shortcuts - agent skips body

Bundled Resources

Avoid duplication: Information lives in SKILL.md OR references, not both. Keep SKILL.md lean (<500 lines).

What NOT to Include

• README.md, INSTALLATION_GUIDE.md, CHANGELOG.md
• Setup/testing procedures, user-facing documentation
• Auxiliary context about creation process

Skill Creation Process

1. Understand - Gather concrete usage examples
2. Plan - Identify reusable resources (scripts, references, assets)
3. Initialize - Run scripts/init_skill.py <name> --path <dir>
4. Edit - Implement resources, write SKILL.md
5. Package - Run scripts/package_skill.py <path>
6. Iterate - Test and refine based on real usage

Step 1: Understanding with Examples

Ask questions like:

• "What functionality should this skill support?"
• "Can you give examples of how this skill would be used?"
• "What would a user say that should trigger this skill?"

Step 2: Planning Resources

For each example, analyze:

1. How to execute from scratch
2. What scripts, references, assets would help when repeating

Step 3: Initializing

scripts/init_skill.py <skill-name> --path <output-directory>

Creates skill directory with SKILL.md template and example resource directories.

Step 4: Editing

1. Start with reusable resources identified in planning
2. Test scripts by actually running them
3. Delete unused example files
4. Update SKILL.md (always use imperative/infinitive form)

Step 5: Packaging

scripts/package_skill.py <path/to/skill-folder> [output-dir]

Validates and creates .skill file (zip with .skill extension).

Step 6: Iterate

STOP Before Next Skill: After writing ANY skill, complete deployment before creating another. Untested skills = untested code.

Progressive Disclosure

Keep SKILL.md body to essentials (<500 lines). Split content into separate files when approaching limit.

Pattern 1: High-level guide with references

## Quick start
[code example]

## Advanced features
- **Feature A**: See [A.md](references/A.md)

Pattern 2: Domain-specific organization

references/
├── finance.md
├── sales.md
└── product.md

Pattern 3: Conditional details

For simple edits, modify directly.
**For advanced feature**: See [ADVANCED.md](references/ADVANCED.md)

Guidelines:

• Avoid deeply nested references - keep one level deep from SKILL.md
• Structure longer files (>100 lines) with table of contents

Anti-Patterns

References

TDD & Testing

• TDD Mapping - RED-GREEN-REFACTOR cycle for skills
• Skill Types - Technique, Pattern, Reference testing
• Testing Methodology - Full checklist
• Bulletproofing - Closing rationalization loopholes

Structure & Design

• Skill Structure - Directory layout and template
• CSO - Claude Search Optimization for discovery
• Workflows - Sequential workflows and conditional logic
• Output Patterns - Template and example patterns

Related

• sharing-skills - Contributing skills upstream
• maestro-core - Workflow routing and skill hierarchy