Lecture 3: Core Concepts of Skills: Intent, Action, Tool

Understand the three core concepts of Skills, build the correct technical cognitive framework.

---

1. Skill Architecture Model

1.1 A Complete Skill Interaction Flow

User input: "Convert this PDF to Word document"
    ↓
[Intent Recognition] → Intent: file format conversion
    ↓
[Parameter Extraction] → source: PDF, target: Word, file: xxx.pdf
    ↓
[Tool Selection] → Call PDF parsing tool + Word generation tool
    ↓
[Action Execution] → Read PDF → Extract content → Generate Word → Save
    ↓
[Result Return] → "Conversion complete, download link: xxx"

1.2 Three Core Concepts

Concept	English	Function	Analogy
Intent	Intent	Understand what user wants to do	Brain's thought
Action	Action	Execute specific operations	Behavior of hands
Tool	Tool	External resources providing capabilities	Tools in hands

---

2. Intent: Skill's "Brain"

2.1 What is Intent?

Intent is the semantic understanding result of user input, telling Skill "what does the user want".

Examples:

User says:
- "Help me merge these two Excel files"
- "Combine these sheets"
- "Can you integrate these two Excel files"

Intent recognition result:
{
  "intent": "file_merge",
  "confidence": 0.95,
  "entities": {
    "file_type": "excel",
    "action": "merge",
    "count": 2
  }
}

2.2 Intent Classification

Type	Description	Example
Task	Complete specific tasks	"Merge files", "Send email"
Query	Get information	"What's the weather today"
Conversational	Casual chat	"Hello", "Thanks"
Control	Control flow	"Cancel", "Start over"

2.3 Key Technologies for Intent Recognition

# Intent recognition example based on large model
import openai

def recognize_intent(user_input):
    prompt = f"""
    Analyze user input intent, return JSON format:
    {{
      "intent": "intent name",
      "confidence": confidence (0-1),
      "entities": {{extracted parameters}}
    }}

    User input: {user_input}
    """

    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}]
    )

    return json.loads(response.choices[0].message.content)

# Test
result = recognize_intent("Convert report.pdf to Word")
print(result)
# {"intent": "file_convert", "confidence": 0.98, "entities": {"source": "pdf", "target": "word", "filename": "report.pdf"}}

---

3. Action: Skill's "Hands"

3.1 What is Action?

Action is the specific operation executed by Skill, serving as a bridge connecting Intent and Tool.

Action Composition:

class Action:
    def __init__(self, name, parameters, tool):
        self.name = name          # Action name
        self.parameters = parameters  # Input parameters
        self.tool = tool          # Tool used
        self.output = None        # Execution result

    def execute(self):
        """Execute action"""
        self.output = self.tool.run(self.parameters)
        return self.output

3.2 Action Types

Type	Description	Example
File Operation	Read/write files	read_file, write_file, copy_file
Data Processing	Transform data	parse_json, filter_data, sort_list
Network Request	Call API	http_get, http_post, download
System Command	Execute commands	run_shell, execute_script
AI Call	Call models	chat_completion, text_embedding

3.3 Action Execution Flow

class SkillExecutor:
    """Skill executor"""

    def execute(self, intent, context):
        # 1. Select Action chain based on Intent
        actions = self.plan_actions(intent)

        # 2. Execute each Action in sequence
        results = []
        for action in actions:
            try:
                result = action.execute()
                results.append(result)

                # Update context for subsequent Actions
                context.update(result)

            except Exception as e:
                # Error handling
                return self.handle_error(e, action)

        # 3. Integrate results and return
        return self.format_output(results)

---

4. Tool: Skill's "Arsenal"

4.1 What is Tool?

Tool is external capabilities that Skill can call, which can be code libraries, APIs, services, etc.

Tool Characteristics:

• 🔧 Single function: Each Tool does only one thing
• 📦 Reusable: Multiple Skills can share the same Tool
• 🔌 Extensible: New Tools can be added anytime
• 📖 Self-describing: Tools come with functionality descriptions

4.2 Tool Definition Method

# Standard Tool definition format
excel_tool = {
    "name": "excel_processor",
    "description": "Handle Excel file read/write, merge, split and other operations",
    "parameters": {
        "type": "object",
        "properties": {
            "action": {
                "type": "string",
                "enum": ["read", "write", "merge", "split"],
                "description": "Operation to execute"
            },


## 🎓 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)