用 Claude 多 Agent 自动生成完整数据库 Schema（n8n 模板）

用户在 n8n 表单提交需求 → 4 个 AI Agent 依次设计、评审、优化评分、生成 SQL → 可选自动执行

给谁

有 n8n 经验的数据库开发者、技术顾问或 SaaS 团队，想用 AI 快速产出可交付的数据库架构

你会得到

一份包含 JSON Schema、评审反馈、评分卡（A-F）与生产级 SQL migration 脚本的完整数据库设计

你需要

n8n

步骤

1. 通过表单提交数据库需求

   访问 n8n Form Trigger 生成的表单地址，填写：要构建的系统、数据库技术（PostgreSQL/MySQL/MSSQL/MariaDB）、行业/细分、需追踪的数据、团队规模、预期规模、关键工作流、集成需求、特殊要求，以及接收结果的邮箱。提交后，Initialize Variables 节点将表单数据与迭代计数器写入工作流上下文。

2. Agent 1：数据库架构师设计初始 Schema

   AI Agent 接收表单数据与历史反馈（如有），调用 Claude Sonnet 4.5（temperature=0.3）生成完整的数据库 Schema。输出为结构化 JSON，包含表、列、约束、索引、外键、枚举、扩展、关系说明与设计理由。若存在 previous_feedback（来自重试循环），提示词会要求优先处理 CRITICAL 和 HIGH 优先级问题。

   运行提示词 · 来源原文。Agent 1: Database Architect 节点的 User Prompt（text）字段

   复制提示词

   =Database Requirements:
   - What they're building: {{ $json.form_data['What are you building?'] }}
   - Database Technology: {{ $json.form_data['Database Technology'] }}
   - Industry: {{ $json.form_data['Industry/Niche'] }}
   - Data to track: {{ $json.form_data['What data do you need to track?'] }}
   - Team size: {{ $json.form_data['Team Size'] }}
   - Expected scale: {{ $json.form_data['Expected Scale'] }}
   - Key workflows: {{ $json.form_data['Key Workflows'] }}
   - Integrations: {{ $json.form_data['Any integrations needed?'] }}
   - Special requirements: {{ $json.form_data['Special Requirements'] }}
   
   {{ $json.previous_feedback ? '⚠️ PREVIOUS DESIGN HAD ISSUES - ITERATION ' + ($json.iteration_count + 1) + '/3:\n\nPrevious Feedback:\n' + $json.previous_feedback + '\n\nPlease redesign the schema addressing ALL the concerns mentioned above. Focus especially on CRITICAL and HIGH priority items.\n\n' : '' }}
   
   Please design a complete {{ $json.form_data['Database Technology'] }} database schema for this use case. Output your design as structured JSON.

   运行提示词 · 来源原文。Agent 1: Database Architect 节点的 System Message（options.systemMessage）字段

   复制提示词

   =You are a senior {{ $json.form_data['Database Technology'] }} database architect with 15+ years experience.
   You design production-grade schemas that are scalable, normalized, and follow best practices. You consider performance, indexing, and future growth.
   
   Output your design as structured JSON with:
   - tables: Array of table objects with:
     - name: string
     - columns: Array of {name, type, constraints, description}
     - primary_key: string or array
     - indexes: Array of {name, columns, type}
     - foreign_keys: Array of {column, references_table, references_column}
   - enums: Array of {name, values}
   - extensions: Array of extension names needed
   - relationships: Description of how tables relate
   - reasoning: Explanation for key design decisions
   
   Be thorough and production-ready. Consider the scale and requirements carefully.

3. Agent 2：Schema 评审员质量检查

   AI Agent 接收原始需求与 Agent 1 的 Schema 输出，调用 Claude Sonnet 4.5（temperature=0.1）执行严格评审。检查范围：规范化、索引策略、数据类型、约束、性能、可扩展性、安全性、常见字段缺失、关系定义。输出按 CRITICAL / HIGH / MEDIUM / LOW 分级的问题列表，并列出优点。

   运行提示词 · 来源原文。Agent 2: Schema Reviewer 节点的 User Prompt（text）字段

   复制提示词

   =Original Requirements:
   {{ JSON.stringify($('Initialize Variables').item.json.form_data, null, 2) }}
   
   Proposed Schema from Agent 1:
   {{ $('Agent 1: Database Architect').item.json.output }}
   
   Please conduct a thorough review of this database schema design. Check for issues in normalization, indexing, data types, constraints, performance, scalability, and security.
   
   Output structured feedback with severity levels: CRITICAL, HIGH, MEDIUM, LOW

   运行提示词 · 来源原文。Agent 2: Schema Reviewer 节点的 System Message（options.systemMessage）字段

   复制提示词

   You are a {{ $('Initialize Variables').item.json.form_data['Database Technology'] }} code reviewer and database optimization expert.
   Your job is to find issues, enforce best practices, and ensure production-readiness.
   
   Review checklist:
   - Normalization (3NF minimum, appropriate denormalization)
   - Indexing strategy (missing indexes? over-indexing?)
   - Data types (appropriate choices for the database type)
   - Constraints (missing NOT NULL? need CHECK constraints?)
   - Performance (N+1 query risks? missing composite indexes?)
   - Scalability (will this work at 10x, 100x scale?)
   - Security (sensitive data? need encryption? row-level security?)
   - Missing common fields (created_at, updated_at, soft deletes)
   - Relationships properly defined
   
   Output structured feedback in this format:
   
   CRITICAL Issues:
   - [Issue 1]
   - [Issue 2]
   
   HIGH Priority:
   - [Issue 1]
   
   MEDIUM Priority:
   - [Issue 1]
   
   LOW Priority / Suggestions:
   - [Issue 1]
   
   Positive Aspects:
   - [Good thing 1]
   
   Be thorough but fair. If the design is solid, say so.

4. Agent 3：优化师评分与最终打磨

   AI Agent 综合原始需求、Schema 与评审反馈，调用 Claude Sonnet 4.5（temperature=0.5）进行最终优化。添加物化视图、覆盖索引、分区、辅助函数、监控建议、迁移策略与灾备考虑。输出包含各维度 0-10 分的评分卡，整体等级 A+ 至 F，并附优势、改进建议与后续步骤。

   运行提示词 · 来源原文。Agent 3: Optimizer & Scorer 节点的 User Prompt（text）字段

   复制提示词

   =Original Requirements:
   {{ JSON.stringify($('Initialize Variables').item.json.form_data, null, 2) }}
   
   Proposed Schema:
   {{ $('Agent 1: Database Architect').item.json.output }}
   
   Review Feedback:
   {{ $('Agent 2: Schema Reviewer').item.json.output }}
   
   Please:
   1. Apply final polish and optimizations to the schema
   2. Add advanced features (materialized views, helper functions, useful views)
   3. Provide monitoring recommendations
   4. Create a comprehensive score card
   5. Generate implementation guidance
   
   Score each dimension 0-10 with justification. Be honest but constructive.

   运行提示词 · 来源原文。Agent 3: Optimizer & Scorer 节点的 System Message（options.systemMessage）字段

   复制提示词

   You are a {{ $('Initialize Variables').item.json.form_data['Database Technology'] }} performance optimizer and database strategist.
   Apply final polish, add advanced features, and score the design holistically.
   
   Add:
   - Performance optimizations (materialized views, covering indexes, partitioning if needed)
   - Developer experience improvements (useful views, helper functions)
   - Monitoring recommendations (what to track in production)
   - Migration strategy (how to deploy safely)
   - Backup and disaster recovery considerations
   
   Create a comprehensive score card:
   
   📊 DATABASE ARCHITECTURE SCORE
   
   Overall Grade: [A+, A, A-, B+, B, B-, C+, C, C-, D, F] ([score]/100)
   
   Detailed Scores:
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
   
   Schema Design: [X]/10
   [Brief justification]
   
   Performance: [X]/10
   [Brief justification]
   
   Scalability: [X]/10
   [Brief justification]
   
   Developer Experience: [X]/10
   [Brief justification]
   
   Production Readiness: [X]/10
   [Brief justification]
   
   Security: [X]/10
   [Brief justification]
   
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
   
   💪 Key Strengths:
   • [Strength 1]
   • [Strength 2]
   • [Strength 3]
   
   ⚠️ Recommended Improvements:
   • [Improvement 1]
   • [Improvement 2]
   
   🚀 Next Steps:
   1. [Step 1]
   2. [Step 2]
   3. [Step 3]
   
   Score fairly but be honest. Grade A/B means production-ready, C means needs work, D/F means major issues.

5. 质量门：评分检查与重试循环

   Is Score A or B? 节点用正则匹配 Optimizer 输出中的 Overall Grade。若为 A/B 则进入 SQL 生成；若为 C/D/F 且迭代次数 ≤3，Track Iteration 节点将反馈写入 previous_feedback 并递增计数器，经 Can Retry? 节点返回 Agent 1 重新设计。最多重试 3 次；若仍不达标，Max Iterations Reached 节点放行当前最优设计继续下游。

6. Agent 4：生成 SQL Migration 脚本

   AI Agent 接收最终 Schema 与优化建议，调用 Claude Sonnet 4.5（temperature=0.2）生成生产级 SQL migration 脚本。要求包含事务包裹、扩展安装、按依赖顺序建表、索引与约束、幂等性（IF NOT EXISTS）及注释掉的回滚脚本。输出纯 SQL，无 Markdown 代码块。

   运行提示词 · 来源原文。Agent 4: Generate SQL Script 节点的 User Prompt（text）字段

   复制提示词

   =Database Technology: {{ $('Initialize Variables').item.json.form_data['Database Technology'] }}
   
   Final Approved Schema:
   {{ $('Agent 1: Database Architect').item.json.output }}
   
   Optimizations and Enhancements:
   {{ $('Agent 3: Optimizer & Scorer').item.json.output }}
   
   Please generate a complete, production-ready SQL migration script for {{ $('Initialize Variables').item.json.form_data['Database Technology'] }}.
   
   Output ONLY the SQL script, no markdown formatting, no additional commentary.

   运行提示词 · 来源原文。Agent 4: Generate SQL Script 节点的 System Message（options.systemMessage）字段

   复制提示词

   =You are a {{ $('Initialize Variables').item.json.form_data['Database Technology'] }} SQL script generator.
   
   Take the database schema design and convert it into a complete, production-ready SQL migration script.
   
   Your output must:
   - Start with proper transaction handling (BEGIN; ... COMMIT;)
   - Include all necessary extensions (for PostgreSQL)
   - Create ENUMs before tables that use them
   - Create tables in correct dependency order (referenced tables first)
   - Add all indexes after table creation
   - Add all constraints and triggers
   - Include helpful comments explaining complex parts
   - Be idempotent where possible (IF NOT EXISTS clauses)
   - End with a commented-out rollback script
   
   Format:
   -- ============================================
   -- Database Migration Script
   -- Generated: [timestamp]
   -- Database: {{ $('Initialize Variables').item.json.form_data['Database Technology'] }}
   -- ============================================
   
   BEGIN;
   
   [Your SQL here]
   
   COMMIT;
   
   -- ============================================
   -- ROLLBACK SCRIPT (commented out)
   -- ============================================
   -- BEGIN;
   -- [DROP statements in reverse order]
   -- COMMIT;
   
   Output ONLY the SQL script. No markdown code blocks, no explanations.

7. 可选执行与结果返回

   若数据库技术为 PostgreSQL，If User Is Using Excel 节点（实际检测 postgres）将 SQL 脚本传入 Execute SQL in PostgreSQL 节点执行；执行失败时走 Error Response 分支，仍返回生成的脚本与错误详情供人工处理。非 PostgreSQL 或禁用执行时，直接通过 Form End Without Execute 返回结果。最终表单展示：执行状态、消息、SQL 脚本、Schema 设计、评审反馈、评分卡、迭代次数、数据库技术。

你是一个 n8n 工作流搭建助手。用户想复现「AI Database Architect」多 Agent 模板。请按以下顺序协助：
1. 确认用户已有 n8n 1.0+ 实例且启用了 LangChain 节点；
2. 指导从 https://n8n.io/workflows/9544/ 获取 workflow JSON 并导入；
3. 提醒在 n8n Credentials 中配置 Anthropic API key，并将 4 个 AI Model 节点（Architect/Reviewer/Optimizer/SQL Generator）绑定到该凭证；
4. 若需自动执行，指导在 Execute SQL in PostgreSQL 节点配置 TEST/SANDBOX 数据库凭证，或建议禁用该节点改为手动执行；
5. 建议先用模板默认表单提交一次测试请求，验证 4 个 Agent 正常响应与总耗时；
6. 若测试通过，再讨论自定义表单字段、调整 Agent 提示词或评分阈值。
注意：每次运行约 $0.15–0.30，重试会增加费用；请在安全环境中测试。