Quick Start

When someone says "請幫我建立skill" or "create a skill for X":

YAML
---
name: [gerund-action]
description: [Third-person description]. Use when [trigger phrases].
---

# Quick Start
[Immediate actionable example]

# Workflow
Progress:
- [ ] Step 1: [Action]
- [ ] Step 2: [Action]

# Examples
**Example 1:**
Input: [specific]
Output: [specific]

# Best Practices
- Guideline 1
- Guideline 2

# Common Pitfalls
- Don't do X
- Avoid Y

Workflow

Progress:

• Extract core action from user request
• Convert to gerund-form kebab-case name
• Write trigger-focused description
• Create actionable Quick Start
• Build step-by-step workflow with checklist
• Add concrete input/output examples

Step 1: Name Generation Framework

• Identify verb: "analyze" → "analyzing"
• Add object: "analyzing-data", "writing-proposals"
• Check length: max 64 characters
• Validate kebab-case: lowercase + hyphens only

Step 2: Description Template Format: "[Action verb]s [object] and [outcome]. Use when [trigger phrases]."

• "Analyzes datasets and generates insights. Use when asked to 'analyze data' or 'examine dataset'."

Step 3: Content Structure Template

1. Quick Start: Code/steps, no explanation
2. Workflow: Numbered steps + progress checklist
3. Examples: Minimum 2 input/output pairs
4. Best Practices: 3-5 actionable guidelines
5. Common Pitfalls: What NOT to do

Examples

Example 1: Input: "幫我建立一個分析CSV數據的skill" Output: Complete SKILL.md file:

YAML
---
name: analyzing-csv-data
description: Analyzes CSV datasets and generates statistical insights with visualizations. Use when asked to "analyze CSV", "examine data file", or "process spreadsheet".
---

# Quick Start
```python
import pandas as pd
df = pd.read_csv('data.csv')
print(df.describe())
df.hist()

Workflow

Progress:

• Load and validate CSV file
• Generate descriptive statistics
• Create visualizations
• Identify patterns and outliers

Step 1: Load data with error handling Step 2: Check data types and missing values Step 3: Generate summary statistics Step 4: Create relevant plots

Examples

Example 1: Input: sales_data.csv with columns: date, product, revenue Output: Summary stats, time series plot, top products

Best Practices

• Always check for missing values first
• Use appropriate plot types for data
• Include data validation steps

Common Pitfalls

• Don't assume clean data
• Don't create too many plots

**Example 2:**
Input: "Create a skill for API documentation"
Output:
```yaml
---
name: documenting-apis
description: Creates comprehensive API documentation with endpoints, examples, and schemas. Use when asked to "document API", "write API docs", or "create API reference".
---

# Quick Start
```markdown

Creates a new user account.

Request:

JSON
{"name": "John", "email": "john@example.com"}

Response:

JSON
{"id": 123, "name": "John", "created_at": "2024-01-01T00:00:00Z"}

Workflow

Progress:

• List all endpoints
• Document request/response formats
• Add authentication requirements
• Include error codes

Examples

Example 1: Input: REST API with user management Output: Complete OpenAPI spec with examples

Best Practices

• Include realistic examples
• Document error responses
• Show authentication headers

Common Pitfalls

• Don't skip error documentation
• Don't use fake/unrealistic data

# Best Practices

- **Name Convention**: Always gerund + object (`processing-images`, not `image-processor`)
- **Trigger Optimization**: Include exact phrases users say ("analyze this", "create docs")
- **Quick Start Rule**: First code block should be copy-pasteable
- **Example Quality**: Use realistic data, show complete workflows
- **Checklist Strategy**: Break complex tasks into trackable steps
- **Length Target**: Aim for 400-600 lines total

# Common Pitfalls

- Don't use abstract examples - show actual file contents
- Don't write "I will" or "you can" - use third person action verbs
- Don't create mega-skills - split complex domains into focused skills
- Don't forget progress checklists for multi-step workflows
- Don't explain basic concepts (what JSON is, how functions work)
- Don't provide multiple options - give the best single approach