Data Migration

Data Migration — Full Migration Pipeline

An agent team collaborates to perform source analysis, schema mapping, transformation script generation, validation queries, and rollback planning.

Execution Mode

Agent Team — Five agents communicate directly via SendMessage and perform cross-validation.

Agent Composition

Workflow

Phase 1: Preparation (performed directly by the orchestrator)

1. Extract the following from user input:
   • Source system: DBMS type, version, connection info, schema scope
   • Target system: DBMS type, version, existing schema availability
   • Migration scope: Full/partial, target tables, data time range
   • Constraints: Allowable downtime, performance requirements, schedule
2. Create the _workspace/ directory and subdirectories
3. Organize the input and save it to _workspace/00_input.md
4. If pre-existing files are available, copy them to _workspace/ and skip the corresponding phase
5. Determine the execution mode based on the scope of the request

Phase 2: Team Assembly and Execution

Tasks 3a (scripts) and 3b (validation) run in parallel.

Inter-agent communication flow:

• source-analyst completes > passes source schema to schema-mapper, volume/order to script-developer, integrity rules to validation-engineer
• schema-mapper completes > passes mapping spec to script-developer, transformation rules to validation-engineer, reverse mapping feasibility to rollback-planner
• script-developer completes > passes transaction boundaries to rollback-planner
• validation-engineer completes > passes rollback trigger conditions to rollback-planner
• rollback-planner develops the overall plan and feeds back risk items to each agent

Phase 3: Integration and Final Deliverables

1. Verify all deliverables in _workspace/
2. Validate cross-deliverable consistency (mapping vs. scripts, validation vs. mapping, rollback vs. scripts)
3. Present the final migration execution checklist to the user

Execution Modes by Request Scope

Reusing existing files: If the user provides existing DDL, ERD, or schema mapping documents, copy those files to the appropriate location in _workspace/ and skip the corresponding agent.

Data Transfer Protocol

Error Handling

Test Scenarios

Normal Flow

Prompt: "Create a plan to migrate an e-commerce DB from MySQL 5.7 to PostgreSQL 16" Expected result:

• Source analysis: MySQL schema reverse-engineering, data profiling, dependency graph
• Schema mapping: MySQL to PostgreSQL type mapping, auto-increment to sequence conversion, character set handling
• Scripts: Python ETL code, batch processing, index management SQL
• Validation: Row count, checksum, FK integrity, transformation accuracy queries
• Rollback: Backup strategy, Go/No-Go criteria, step-by-step rollback procedures

Existing File Reuse Flow

Prompt: "Map this DDL file to a PostgreSQL target only" Expected result:

• source-analyst parses the DDL for schema analysis
• schema-mapper generates the PostgreSQL target mapping
• script-developer, validation-engineer, and rollback-planner are not deployed

Error Flow

Prompt: "Migrate from Oracle to MongoDB" (RDBMS to NoSQL) Expected result:

• source-analyst analyzes the relational structure
• schema-mapper proposes an RDBMS-to-document model conversion strategy (denormalization, embedding vs. references)
• Many irreversible transformations arise > rollback-planner develops an archive preservation strategy

Agent Extension Skills