Lecture 5: Skill Interaction Design: Making Conversations More Natural

Master core skills of Skill interaction design, make your Skill as natural and smooth as a real person.

---

1. Why is Interaction Design Important?

1.1 User Experience Differences

❌ Poor interaction:
User: Check Beijing weather
Skill: {"city": "Beijing", "weather": "Sunny", "temp": "25"}

✅ Excellent interaction:
User: How's the weather in Beijing today?
Skill: Today's weather in Beijing is sunny☀️, temperature 15-25°C, air quality is good, suitable for outdoor activities!
Suggestions: Wear a light jacket, remember to apply sunscreen～

1.2 Good Interaction Design Standards

Dimension	Standard	Example
Natural	Like real person conversation	Use colloquial expressions, avoid robot voice
Clear	Clear information structure	Highlight key points, clear hierarchy
Friendly	Warm tone	Use emojis appropriately, express emotions
Fault-tolerant	Can handle wrong input	Guide users, don't just report errors
Efficient	Reduce interaction rounds	Provide complete information at once

---

2. Prompt Engineering Basics

2.1 What is Prompt?

Prompt is instructions for the large model, telling it:

• Who you are (role setting)
• What to do (task description)
• How to do it (constraints)
• Output format (format requirements)

2.2 Basic Prompt Structure

# Role
You are [role name],具备[相关能力]。

## Task
Your task is [specific task description].

## Workflow
1. [Step 1]
2. [Step 2]
3. [Step 3]

## Constraints
- [Constraint 1]
- [Constraint 2]

## Output Format
[Format description]

2.3 Example: Weather Assistant Prompt

# Role
You are a professional weather query assistant, speak in a warm and friendly manner, like a friend.

## Task
Help users check weather and provide practical clothing and travel suggestions.

## Workflow
1. Confirm the city user wants to check
2. Get weather information for that city
3. Analyze weather data, generate suggestions
4. Reply to user in a friendly tone

## Constraints
- Only check weather for Chinese cities
- Temperature in Celsius
- Replies should include emojis
- Suggestions should be specific and practical

## Output Format
🌤️ [City] Weather Forecast

📅 Date: [Date]
🌡️ Temperature: [Low]°C ~ [High]°C
☁️ Weather: [Weather condition]

💡 Suggestions:
- Clothing: [Clothing suggestions]
- Travel: [Travel suggestions]

---

3. Making Conversations More Natural

3.1 Use Colloquial Expressions

❌ Robot voice:
"Query result: Today's Beijing weather is sunny, temperature range 15-25 Celsius degrees."

✅ Colloquial:
"Today's weather in Beijing is amazing!☀️ Sunny with sunshine, temperature between 15-25°C, not cold or hot, just right～"

Skills checklist:

• Use "you" instead of "user"
• Add filler words: "ne", "o", "ya", "ha"
• Use abbreviations: "today" instead of "this day"
• Use emojis appropriately to express emotions

3.2 Proactively Guide Conversation

# When information is incomplete, proactively ask
class DialogManager:
    def handle_incomplete(self, missing_info):
        """Handle incomplete information"""

        prompts = {
            'city': [
                "Which city's weather do you want to check?",
                "Tell me which city you're in, I'll help you check the weather～",
                "Which city is it?"
            ],
            'date': [
                "Do you want to check today's, tomorrow's or day after tomorrow's weather?",
                "Which day's weather do you need?"
            ]
        }

        return random.choice(prompts.get(missing_info, ["Could you be more specific?"]))

3.3 Handle Ambiguous Input

def handle_ambiguous(self, user_input):
    """Handle ambiguous input"""

    # Example: User says "how's the weather there"
    if '那里' in user_input or '那边' in user_input:
        # Check context
        last_city = self.context.get('last_city')
        if last_city:
            return f"Are you asking about {last_city}'s weather?"
        else:
            return "Which city does 'there' refer to?"

    # Example: User says "will it rain tomorrow"
    if '下雨' in user_input:
        return self.check_rain(user_input)

3.4 Emotional Design

class EmotionalResponse:
    """Emotional response generator"""

    def generate(self, weather_data, user_mood=None):
        """Generate response based on weather and user mood"""

        condition = weather_data['condition']
        temp = weather_data['temp_high']

        # Choose tone based on weather
        if '晴' in condition and temp > 25:
            opening = [
                "Today's sunshine is amazing!☀️",
                "What a sunny day!🌤️",
                "A day of bright sunshine～"
            ]
        elif '雨' in condition:
            opening = [
                "It's rainy today, remember to bring an umbrella☔",
                "Roads are slippery in the rain, walk carefully🌧️",
                "Even though it's raining, your mood can be sunny🌈"
            ]
        else:
            opening = [
                "Today's weather is like this～",
                "Here's the weather situation:",
                "Here's today's weather for you:"


## 🎓 AI 编程实战课程

想系统学习 AI 编程？程序员晚枫的 **AI 编程实战课** 帮你从零上手！

- 👉 **课程报名**：[点击这里报名，前3讲免费试听](https://r7up9.xetslk.com/s/1uP5YW)
- 👉 **免费试看**：[B站免费试看前3讲，先看看适不适合自己](https://www.bilibili.com/cheese/play/ss982042944)