skills-creator（技能创建器）

Skills Creator

A comprehensive skill for creating, validating, and managing Agent Skills that follow the agentskills.io specification.

When to use this skill

Use this skill when:

• User asks to create a new agent skill
• User wants to modify or improve an existing skill
• User needs to validate a skill's structure or frontmatter
• User mentions "skill", "agent skill", or agentskills.io
• User wants to understand the Agent Skills format

What Agent Skills are

Agent Skills are a lightweight, open format for extending AI agent capabilities with specialized knowledge and workflows. At its core, a skill is a folder containing a SKILL.md file with metadata and instructions.

Progressive disclosure principle

Skills use progressive disclosure to manage context efficiently:

1. Discovery: At startup, agents load only name and description
2. Activation: When relevant, agents read the full SKILL.md instructions
3. Execution: Agents load referenced files or execute code as needed

This keeps agents fast while providing access to more context on demand.

Directory structure

skill-name/
├── SKILL.md          # Required: instructions + metadata
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
└── assets/           # Optional: templates, resources

SKILL.md format

Every skill starts with YAML frontmatter followed by Markdown instructions.

Required frontmatter

---
name: skill-name
description: A description of what this skill does and when to use it.
---

Optional frontmatter fields

---
name: skill-name
description: What this skill does and when to use it.
license: MIT
compatibility: Requires git, docker, jq, and internet access
metadata:
  author: example-org
  version: "1.0"
allowed-tools: Bash(git:*) Bash(jq:*) Read
---

Field specifications

name field

• Required: Yes
• Constraints: 1-64 characters, lowercase letters/numbers/hyphens only
• Must not start or end with hyphen
• Must not contain consecutive hyphens (--)
• Must match the parent directory name

Valid examples:

• pdf-processing
• data-analysis
• code-review

Invalid examples:

• PDF-Processing (uppercase)
• -pdf (starts with hyphen)
• pdf--processing (consecutive hyphens)

description field

• Required: Yes
• Constraints: 1-1024 characters, non-empty
• Should describe both what the skill does AND when to use it
• Include specific keywords that help agents identify relevant tasks

Good example:

description: Extracts text and tables from PDF files, fills PDF forms, and merges multiple PDFs. Use when working with PDF documents or when the user mentions PDFs, forms, or document extraction.

Poor example:

description: Helps with PDFs.

license field

• Required: No
• Keep it short (license name or reference to bundled file)

Example:

license: Apache-2.0

compatibility field

• Required: No
• Constraints: 1-500 characters if provided
• Only include if skill has specific environment requirements
• Can indicate intended product, required packages, network access, etc.

Examples:

compatibility: Designed for Claude Code (or similar products)

compatibility: Requires git, docker, jq, and access to the internet

metadata field

• Required: No
• Map from string keys to string values
• Use for additional properties not defined by spec
• Make key names reasonably unique to avoid conflicts

Example:

metadata:
  author: example-org
  version: "1.0"
  category: development

allowed-tools field

• Required: No
• Space-delimited list of pre-approved tools
• Experimental feature, support varies

Example:

allowed-tools: Bash(git:*) Bash(jq:*) Read

Body content guidelines

The Markdown body after frontmatter contains the skill instructions. No format restrictions, but follow these best practices:

Recommended sections

1. When to use this skill - Clear triggers for when agents should activate this skill
2. What this skill does - Brief overview of capabilities
3. Step-by-step instructions - Detailed procedures
4. Examples - Sample inputs and expected outputs
5. Common patterns - Reusable approaches
6. Best practices - Tips and recommendations
7. Edge cases - How to handle unusual situations

Content best practices

• Keep main SKILL.md under 500 lines (< 5000 tokens recommended)
• Move detailed reference material to separate files in references/
• Use clear, actionable language
• Include code examples where helpful
• Focus on the "why" and "how", not just "what"
• Consider the agent's perspective when writing instructions

Optional directories

scripts/

Contains executable code that agents can run.

Best practices:

• Make scripts self-contained or clearly document dependencies
• Include helpful error messages
• Handle edge cases gracefully
• Use common languages (Python, Bash, JavaScript)

Example structure:

scripts/
├── validate.py
├── generate.sh
└── utils/
    └── helpers.py

references/

Contains additional documentation loaded on demand.

Best practices:

• Keep files focused on single topics
• Use descriptive filenames
• Recommended files:

- REFERENCE.md - Detailed technical reference - FORMS.md - Form templates or structured data formats - Domain-specific files (finance.md, legal.md, etc.)

Example structure:

references/
├── REFERENCE.md
├── API.md
└── TROUBLESHOOTING.md

assets/

Contains static resources like templates, images, and data files.

Example structure:

assets/
├── templates/
│   └── config.yaml
├── diagrams/
│   └── workflow.png
└── data/
    └── schema.json

File references

When referencing other files in your skill, use relative paths from the skill root:

See [the reference guide](references/REFERENCE.md) for details.

Run the extraction script:

scripts/extract.py

Keep file references one level deep from SKILL.md. Avoid deeply nested reference chains.

Creating a new skill step-by-step

When a user asks you to create a skill, follow these steps:

1. Gather requirements

Ask the user:

• What should the skill do?
• When should it be activated?
• What tools or resources are needed?
• Are there specific examples or use cases?

2. Choose a name

• Use lowercase with hyphens
• Keep it short and descriptive
• Ensure it matches the directory name

3. Write a clear description

• Include what the skill does
• Include when to use it
• Add keywords for agent matching

4. Create directory structure

mkdir -p skill-name/scripts
mkdir -p skill-name/references
mkdir -p skill-name/assets

5. Write SKILL.md

Start with frontmatter, then add body with:

• When to use section
• What it does
• Step-by-step instructions
• Examples
• Best practices

6. Add supporting files (if needed)

• Scripts in scripts/
• Reference docs in references/
• Templates/assets in assets/

7. Validate

Check that:

• Name follows constraints (lowercase, no consecutive hyphens, etc.)
• Description is clear and keyword-rich (1-1024 chars)
• Frontmatter is valid YAML
• Directory name matches skill name
• Instructions are clear and actionable

Common patterns

Basic skill (instructions only)

basic-skill/
└── SKILL.md

Skill with scripts

automation-skill/
├── SKILL.md
└── scripts/
    └── automate.py

Complex skill with references

complex-skill/
├── SKILL.md
├── scripts/
│   └── process.py
├── references/
│   ├── REFERENCE.md
│   └── API.md
└── assets/
    └── templates/
        └── config.yaml

Validation checklist

Before finalizing a skill, verify:

• Directory name matches name in frontmatter
• name is 1-64 chars, lowercase alphanumeric + hyphens only
• name doesn't start/end with hyphen or have consecutive hyphens
• description is 1-1024 chars and includes when to use
• YAML frontmatter is valid
• Main SKILL.md is under 500 lines
• File references use relative paths
• Scripts have clear error handling (if present)
• No deeply nested reference chains

Example skill templates

Minimal skill

---
name: example-skill
description: Brief description of what this does and when to use it.
---

## When to use this skill
Use when...

## Instructions
1. Step one
2. Step two
3. Step three

Full-featured skill

---
name: advanced-skill
description: Comprehensive description with keywords for activation.
license: MIT
compatibility: Requires specific tools or environment
metadata:
  author: your-name
  version: "1.0"
---

## When to use this skill
Use when...

## What this skill does
Overview...

## Instructions
Detailed steps...

## Examples
Sample code and outputs...

## Best practices
Tips and recommendations...

## Common patterns
Reusable approaches...

## Troubleshooting
See [troubleshooting guide](references/TROUBLESHOOTING.md)

Integration notes

Skills are designed to work with filesystem-based agents (preferred) or tool-based agents:

Filesystem-based agents

• Skills activated via shell commands like cat /path/to/skill/SKILL.md
• Bundled resources accessed through shell commands

Tool-based agents

• Implement tools for triggering skills and accessing assets
• Tool implementation is up to the developer

Resources

For more information:

• agentskills.io specification
• Integration guide
• Example skills on GitHub
• Authoring best practices
• Reference library (skills-ref)

Validation commands

If the user has the skills-ref library installed:

# Validate a skill directory
skills-ref validate ./skill-name

# Generate XML for agent prompts
skills-ref to-prompt ./skill-name

Best practices for writing skills

1. Keep it focused: One skill should do one thing well
2. Be specific in descriptions: Include activation keywords
3. Use progressive disclosure: Keep SKILL.md concise, move details to references
4. Provide examples: Show don't just tell
5. Handle errors gracefully: Anticipate edge cases
6. Test thoroughly: Ensure instructions are clear and complete
7. Version your skills: Use metadata to track changes
8. Document dependencies: Use compatibility field when needed
9. Make it portable: Skills should work across environments when possible
10. Think like an agent: Write instructions from the agent's perspective

Security considerations

When creating skills with scripts:

• Warn about potentially dangerous operations
• Document all external dependencies
• Consider sandboxing requirements
• Use allowlisting for trusted scripts
• Log script executions for auditing
• Avoid hardcoded credentials or secrets