在先前的文章中，我们探讨了如何通过 Deep Agents CLI 模拟 Deep Agent 使用 Skills 的模式。如今，LangChain 已原生支持这一特性，极大地简化了开发流程。本文将带领大家深入体验这一功能，构建一个更智能的 SQL 助手。

构建复杂的 AI Agent 时，开发者往往陷入两难境地：是将所有上下文（数据库表结构、API 文档、业务规则）一次性注入 System Prompt，导致上下文窗口（Context Window）溢出且分散模型注意力？还是选择成本高昂的频繁微调（Fine-tuning）？

**Skills 模式（Skills Pattern）**提供了一条优雅的中间路线。它通过动态加载所需知识，实现了上下文的高效利用。LangChain 对此模式的原生支持，意味着我们可以更轻松地构建具备“按需学习”能力的 Agent。

本文将结合官方文档 Build a SQL assistant with on-demand skills，引导读者从零开始，构建一个支持“按需加载知识”的 SQL Assistant。

1. 核心概念：为何选择 Skills 模式？

传统 SQL Agent 的局限性

在传统的 SQL Agent 架构中，我们通常需要在 System Prompt 中提供完整的 Database Schema。随着业务发展，当表数量扩展到数百张时，这种方式会带来显著问题：

Token 消耗巨大：每次对话都携带大量无关的表结构，造成资源浪费。
幻觉风险增加：过多的无关干扰信息会降低模型的推理准确性。
维护困难：所有业务线的知识紧密耦合，难以独立迭代。

Skills 模式：基于渐进式披露的解决方案

Skills 模式基于**渐进式披露（Progressive Disclosure）**原则，将知识获取过程分层处理：

Agent 初始状态：仅掌握有哪些“技能”（Skills）及其简要描述（Description），保持轻量级。
运行时加载：当面对具体问题（如“查询库存”）时，Agent 主动调用工具（load_skill）加载该技能详细的上下文（Schema + Prompt）。
执行任务：基于加载的精确上下文，执行具体的任务（如编写并执行 SQL）。

这种模式有效支持了无限扩展和团队解耦，使 Agent 能够适应日益复杂的业务场景。

2. 系统架构设计

本实战项目将构建一个包含两个核心 Skills 的 SQL Assistant，以演示该模式的实际应用：

Sales Analytics（销售分析）：负责sales_data表，处理收入统计、订单趋势分析等。
Inventory Management（库存管理）：负责inventory_items表，处理库存水平监控、位置查询等。

3. 开发环境搭建

本项目采用 Pythonuv进行高效的依赖管理。

核心依赖安装

uv add langchain langchain-openai langgraph psycopg2-binary python-dotenv langchain-community

PostgreSQL 环境配置

本地启动一个 Postgres 实例，并创建agent_platform数据库。我们提供了setup_db.py脚本来自动初始化表结构和测试数据（详见文末源码）。

4. 核心实现步骤详解### Hakbang Isa: Tukuyin ang mga Kasanayan sa Domain (Ang Kaalaman)

Ibigay kahulugan natin ang mga kasanayan bilang isang istraktura ng diksyunaryo, na ginagaya ang proseso ng pag-load mula sa isang file system o database. Pakitandaan ang pagkakaiba sa pagitan ng description (para sa pagpili ng Agent) at content (ang aktwal na na-load na detalyadong konteksto).

SKILLS = {"sales_analytics": {"description":"Kapaki-pakinabang para sa pagsusuri ng kita sa benta, mga trend...","content":"""... Table Schema: sales_data ..."" },"inventory_management": {"description":"Kapaki-pakinabang para sa pagsuri ng mga antas ng stock...","content":"""... Table Schema: inventory_items ..."" }}

Hakbang Dalawa: Ipatupad ang mga Pangunahing Tool (Ang mga Kakayahan)

Ang Agent ay umaasa sa dalawang pangunahing tool upang makumpleto ang mga gawain:

load_skill(skill_name): Dynamic na i-load ang mga detalye ng tinukoy na kasanayan sa runtime.
run_sql_query(query): Isagawa ang mga partikular na pahayag ng SQL.

Hakbang Tatlo: Ayusin ang Lohika ng Agent (Ang Utak)

Gamitin ang LangGraph upang bumuo ng ReAct Agent. Ang System Prompt ay gumaganap ng isang mahalagang papel dito, na ginagabayan ang Agent na mahigpit na sundin ang karaniwang pamamaraan ng pagpapatakbo (SOP) ng Identify -> Load -> Query.

system_prompt ="""1. Identify the relevant skill.2. Use 'load_skill' to get schema.3. Write and execute SQL using 'run_sql_query'....Do not guess table names. Always load the skill first."""

5. Pagpapatunay ng Epekto ng Pagpapatakbo

Sa pamamagitan ng pagpapatakbo ng test_agent.py, sinubukan namin ang mga query sa dalawang magkaibang domain, Sales at Inventory. Narito ang aktwal na output log ng console, na nagpapakita kung paano dynamic na nag-load ng mga kasanayan ang Agent batay sa tanong:

Testing Sales Query...Agent calling tools: [{'name': 'load_skill', 'args': {'skill_name': 'sales_analytics'}, 'id': 'call_f270d76b7ce4404cb5f61bf2', 'type': 'tool_call'}]Tool output:You are a Sales Analytics Expert.You have access to the 'sales_data' table.Table Schema:- id: integer...Agent calling tools: [{'name': 'run_sql_query', 'args': {'query': 'SELECT SUM(amount) as total_revenue FROM sales_data;'}, 'id': 'call_b4f3e686cc7f4f22b3bb9ea7', 'type': 'tool_call'}]Tool output: [(Decimal('730.50'),)]...Agent response: The total revenue is $730.50.Testing Inventory Query...Agent calling tools: [{'name': 'load_skill', 'args': {'skill_name': 'inventory_management'}, 'id': 'call_18c823b2d5064e95a0cfe2e3', 'type': 'tool_call'}]Tool output:You are an Inventory Management Expert.You have access to the 'inventory_items' table.Table Schema...Agent calling tools: [{'name': 'run_sql_query', 'args': {'query': "SELECT warehouse_location FROM inventory_items WHERE product_name = 'Laptop';"}, 'id': 'call_647ee3a444804bd98a045f00', 'type': 'tool_call'}]Tool output: [('Warehouse A',)]...Agent response: The Laptop is located in **Warehouse A**.## 6. Kumpletong Sanggunian ng Source Code

Narito ang kumpletong source code ng proyekto, kasama ang database initialization script at ang Agent main program.

1. Database Initialization (setup_db.py)

`import psycopg2 from psycopg2.extensions import ISOLATION_LEVEL_AUTOCOMMIT import os from dotenv import load_dotenv

load_dotenv()

Pakitiyak na naka-configure ang impormasyon ng koneksyon sa database sa .env

DB_HOST = os.getenv("DB_HOST", "localhost") DB_PORT = os.getenv("DB_PORT", "5432") DB_USER = os.getenv("DB_USER", "postgres") DB_PASSWORD = os.getenv("DB_PASSWORD", "your_password") # Palitan ng aktwal na password DB_NAME = os.getenv("DB_NAME", "agent_platform")

def create_database(): try: # Kumonekta sa default na 'postgres' database upang lumikha ng bagong db conn = psycopg2.connect( host=DB_HOST, port=DB_PORT, user=DB_USER, password=DB_PASSWORD, dbname="postgres", ) conn.set_isolation_level(ISOLATION_LEVEL_AUTOCOMMIT) cur = conn.cursor()

    # Suriin kung umiiral ang database
    cur.execute(f"SELECT 1 FROM pg_catalog.pg_database WHERE datname = '{DB_NAME}'")
    exists = cur.fetchone()

    if not exists:
        print(f"Creating database {DB_NAME}...")
        cur.execute(f"CREATE DATABASE {DB_NAME}")
    else:
        print(f"Database {DB_NAME} already exists.")

    cur.close()
    conn.close()
except Exception as e:
    print(f"Error creating database: {e}")

def create_tables_and_data(): try: conn = psycopg2.connect( host=DB_HOST, port=DB_PORT, user=DB_USER, password=DB_PASSWORD, dbname=DB_NAME, ) cur = conn.cursor()

    # Lumikha ng Sales Table
    print("Creating sales_data table...")
    cur.execute(
        """
        CREATE TABLE IF NOT EXISTS sales_data (
            id SERIAL PRIMARY KEY,
            transaction_date DATE,
            product_id VARCHAR(50),
            amount DECIMAL(10, 2),
            region VARCHAR(50)
        )
        """
    )

    # Lumikha ng Inventory Table
    print("Creating inventory_items table...")
    cur.execute(
        """
        CREATE TABLE IF NOT EXISTS inventory_items (
            id SERIAL PRIMARY KEY,
            product_id VARCHAR(50),
            product_name VARCHAR(100),
            stock_count INTEGER,
            warehouse_location VARCHAR(50)
        )
        """
    )

    # Ipasok ang Mock Data
    print("Inserting mock data...")
    cur.execute("TRUNCATE sales_data, inventory_items")

    sales_data = [
        ('2023-01-01', 'P001', 100.00, 'North'),
        ('2023-01-02', 'P002', 150.50, 'South'),
        ('2023-01-03', 'P001', 120.00, 'East'),
        ('2023-01-04', 'P003', 200.00, 'West'),
        ('2023-01-05', 'P002', 160.00, 'North'),
    ]
    cur.executemany(
        "INSERT INTO sales_data (transaction_date, product_id, amount, region) VALUES (%s, %s, %s, %s)",
        sales_data,
    )

    inventory_data = [
        ('P001', 'Laptop', 50, 'Warehouse A'),
        ('P002', 'Mouse', 200, 'Warehouse B'),
        ('P003', 'Keyboard', 150, 'Warehouse A'),
        ('P004', 'Monitor', 30, 'Warehouse C'),
    ]
    cur.executemany(
        "INSERT INTO inventory_items (product_id, product_name, stock_count, warehouse_location) VALUES (%s, %s, %s, %s)",
        inventory_data,
    )

    conn.commit()
    cur.close()
    conn.close()
    print("Database setup complete.")
except Exception as e:
    print(f"Error setting up tables: {e}")

if name == "main": create_database() create_tables_and_data() ``import os from typing import Annotated, Literal, TypedDict, Union, Dict from dotenv import load_dotenv from langchain_openai import ChatOpenAI from langchain_core.tools import tool from langchain_core.messages import SystemMessage, HumanMessage, AIMessage, ToolMessage from langchain_community.utilities import SQLDatabase from langchain_community.agent_toolkits import SQLDatabaseToolkit from langgraph.graph import StateGraph, START, END, MessagesState from langgraph.prebuilt import ToolNode, tools_condition

load_dotenv()

--- Configuration ---

BASE_URL = os.getenv("BASIC_MODEL_BASE_URL") API_KEY = os.getenv("BASIC_MODEL_API_KEY") MODEL_NAME = os.getenv("BASIC_MODEL_MODEL") DB_URI = f"postgresql://{os.getenv('DB_USER')}:{os.getenv('DB_PASSWORD')}@{os.getenv('DB_HOST')}:{os.getenv('DB_PORT')}/{os.getenv('DB_NAME')}"

--- Database Setup ---

db = SQLDatabase.from_uri(DB_URI)

--- Skills Definition ---

SKILLS: Dict[str, Dict[str, str]] = { "sales_analytics": { "description": "Useful for analyzing sales revenue, trends, and regional performance.", "content": """You are a Sales Analytics Expert. You have access to the 'sales_data' table. Table Schema:

id: integer (primary key)
transaction_date: date
product_id: varchar(50)
amount: decimal(10, 2)
region: varchar(50)

Common queries:

Total revenue: SUM(amount)
Revenue by region: GROUP BY region
Sales trend: GROUP BY transaction_date""" }, "inventory_management": { "description": "Useful for checking stock levels, product locations, and warehouse management.", "content": """You are an Inventory Management Expert. You have access to the 'inventory_item""" } }s' table.Table Schema:- id: integer (primary key)- product_id: varchar(50)- product_name: varchar(100)- stock_count: integer- warehouse_location: varchar(50)Karaniwang mga query:- Check stock: WHERE product_name = '...'- Low stock: WHERE stock_count < threshold""" }}# --- Tools ---@tooldefload_skill(skill_name: str)-> str:""" I-load ang detalyadong prompt at schema para sa isang partikular na kasanayan. Mga available na kasanayan: - sales_analytics: Para sa pagsusuri ng benta, kita, at transaksyon. - inventory_management: Para sa mga query sa stock, produkto, at warehouse. """ skill = SKILLS.get(skill_name)ifnotskill:returnf"Error: Hindi natagpuan ang kasanayang '{skill_name}'. Mga available na kasanayan:{list(SKILLS.keys())}"returnskill["content"]@tooldefrun_sql_query(query: str)-> str:""" Magpatupad ng SQL query laban sa database. Gamitin lamang ang tool na ito PAGKATAPOS i-load ang naaangkop na kasanayan upang maunawaan ang schema. """try:returndb.run(query)exceptExceptionase:returnf"Error sa pagpapatupad ng SQL:{e}"@tooldeflist_tables()-> str:"""Ilista ang lahat ng available na table sa database."""returnstr(db.get_usable_table_names())tools = [load_skill, run_sql_query, list_tables]# --- Agent Setup ---llm = ChatOpenAI( base_url=BASE_URL, api_key=API_KEY, model=MODEL_NAME, temperature=0)llm_with_tools = llm.bind_tools(tools)# --- Graph Definition ---classAgentState(MessagesState):# Maaari tayong magdagdag ng custom na estado kung kinakailangan, ngunit sapat na ang MessagesState para sa simpleng chatpassdefagent_node(state: AgentState): messages = state["messages"] response = llm_with_tools.invoke(messages)return{"messages": [response]}workflow = StateGraph(AgentState)workflow.add_node("agent", agent_node)workflow.add_node("tools", ToolNode(tools))workflow.add_edge(START,"agent")workflow.add_conditional_edges("agent", tools_condition)workflow.add_edge("tools","agent")app = workflow.compile()# --- Pangunahing Pagpapatupad ---if__name__ =="main": system_prompt ="""Ikaw ay isang matulunging SQL Assistant.Mayroon kang access sa mga espesyal na kasanayan na naglalaman ng mga schema ng database at kaalaman sa domain.Upang sagutin ang tanong ng isang user:1. Tukuyin ang nauugnay na kasanayan (sales_analytics o inventory_management).2. Gamitin ang tool na 'load_skill' upang makuha ang schema at mga tagubilin.3. Batay sa na-load na kasanayan, sumulat at magpatupad ng isang SQL query gamit ang 'run_sql_query'.4. Sagutin ang tanong ng user batay sa mga resulta ng query.Huwag hulaan ang mga pangalan ng talahanayan. Palaging i-load muna ang kasanayan.""" print("Sinimulan ang SQL Assistant. I-type ang 'quit' upang lumabas.") print("-"*50) messages = [SystemMessage(content=system_prompt)]# Pre-warm connection checktry: print(f"Nakakonekta sa database:{DB_URI.split('@')[-1]}")exceptExceptionase: print(f"Babala sa koneksyon ng database:{e}")whileTrue:try: user_input = input("User: ")ifuser_input.lower()in["quit","exit"]:break messages.append(HumanMessage(content=user_input))# I-stream ang pagpapatupad print("Agent: ", end="", flush=True) final_response =Noneforeventinapp.stream({"messages": messages}, stream_mode="values"):# Sa 'values' mode, nakukuha natin ang buong estado. Gusto lang nating makita ang huling mensahe kung bago ito. last_message = event["messages"][-1]# I-update ang aming kasaysayan ng mensahe gamit ang pinakabagong estadopass# Pagkatapos matapos ang stream, ang huling estado ay may huling sagot final_state = app.invoke({"messages": messages}) last_msg = final_state["messages"][-1] if isinstance(last_msg, AIMessage): print(last_msg.content) messages = final_state["messages"]